main
123 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 wolfy
|
e8b2c9952b | fix: revert streaming, add dedup and format enforcement | ||
|
wolfy
|
a847c1f3c9 | fix: improve review inline/body discipline and agent reliability | ||
|
wolfy
|
11dc00b2b8 | chore: some more improvements to try and get as close to original as possible | ||
|
wolfy
|
37d15a338d | chore: more improvement and reproducibility | ||
|
wolfy
|
fc11b91851 | chore: continue improving review feedback | ||
|
wolfy
|
4a1743126e | fix: diagnostic issues | ||
|
wolfy
|
57e6529f97 | fix: comment anchor, remove emojis | ||
|
wolfy
|
5ea8a23d80 | fix: review should leave comments on actual files | ||
|
wolfy
|
1e839d36a9 | fix: review process and cleanup | ||
|
wolfy
|
2aca1a3aa3 | feat: adapt pullfrog for gitea + ollama | ||
 Colin McDonnell
|
01e4daa0b5 |
checkout_pr: refuse unconditionally on dirty working tree (#808)
* checkout_pr: refuse unconditionally on dirty working tree drop the live-HEAD comparison from the guard introduced in #796. any checkout_pr call with staged or unstaged changes now throws, even when HEAD is already on pr-N. no stashing, no idempotent escape hatch. motivation is the zed-industries/cloud (2026-05-18) incident: shared-cwd subagents make "carry edits along" semantics dangerous, and the HEAD-equality predicate let a re-checkout silently inherit working-tree state from a sibling agent. forcing commit/discard before any PR-context operation eliminates the entire carry-forward failure class. error names the PR number, lists dirty paths, and tells the agent to commit/push/restore/clean before retrying. * improve dirty-tree error: precise discard commands copilot caught two sloppy bits in the error string: - "push" alone does not clean a dirty tree (needs commit first) - bare `git clean` is a no-op without `-fd` reword to "commit (then push if needed), or discard with `git restore --staged --worktree .` / `git clean -fd`" so the guidance is actually actionable. * checkout_pr: initial-branch invariant setupGit captures `toolState.initialBranch` at run start via live `git rev-parse --abbrev-ref HEAD`. checkout_pr refuses unless current HEAD matches the run-entry branch or the target `pr-N` (idempotent same-PR re-checkout). uses live rev-parse, not toolState.issueNumber (poisonable per the PR #796 review). refusal error names the current branch, target PR, recovery path (`git checkout <initialBranch>` with the literal branch name), and explicitly states routing around via the `git` tool is not sanctioned. closes the zed-industries/cloud (2026-05-18) shape where a subagent parked HEAD on someone else's `pr-X` and the orchestrator's next checkout_pr inherited that position. * reviewfrog: enforce canonical diff + pre-commit halt; align Build dispatch extend REVIEWER_SYSTEM_PROMPT with two prepended HARD CONSTRAINTS: - first action MUST be `git diff origin/<base>` (single-rev, captures uncommitted). no other diff first; no checkout_pr; no alt-ref fetches; no branch listing; no `gh pr list`. - empty canonical diff + claimed-changes dispatch ⇒ reply exactly with `no changes detected — likely pre-commit Build self-review; orchestrator should commit then re-dispatch` and stop. do not guess PR numbers (the zed thrash that ended in `checkout_pr({2582})`). reshape Build mode reviewfrog dispatch step around a verbatim template that names: (a) the situation is pre-commit, (b) canonical diff command, (c) halt-on-empty-diff rule. orchestrator side now says the same thing as the reviewer's baked-in prompt. delegation-discipline bullets and orchestrator-evaluation guidance kept intact. * checkout_pr: handle detached-HEAD entry in initial-branch invariant pullfrog incremental review caught a defense-in-depth gap: `git rev-parse --abbrev-ref HEAD` returns the sentinel string `"HEAD"` on detached entry, which is the default `actions/checkout` state for `pull_request` events. with the previous string-typed `initialBranch`, both the captured value and the live probe would equal `"HEAD"` on any detached state, trivially satisfying the invariant — including a subagent doing `git checkout --detach <sha>`. discriminate the captured HEAD: probe `git symbolic-ref --short HEAD` first (works on named branches), fall back to `git rev-parse HEAD` (SHA) on detached entry. store as `{ kind: "branch"; name } | { kind: "detached"; sha }`. checkout_pr runs the identical probe at call time and compares like-with-like (branch name vs branch name, SHA vs SHA). refusal error renders both heads via a small `describeHead` helper and chooses the right `git checkout` recovery target (branch name or SHA). no inline-discriminant `as` casts — uses a top-level `headsEqual` that narrows via the discriminator. |
||
|
Colin McDonnell
|
6c166ac1cc |
fix: prevent cross-PR push from subagent-induced branch switch (#796)
* fix: prevent cross-PR push from subagent-induced branch switch A workflow_dispatch run for zed-industries/cloud (workflow run 26036155393) force-pushed the orchestrator's work onto an unrelated engineer's PR branch (origin/reactivate-pro-plan, PR #2582). The orchestrator's reviewfrog subagent called checkout_pr({pull_number: 2582}), which (1) moved the shared working tree to pr-2582 and (2) persisted pushDest pointing at reactivate-pro-plan. The orchestrator's subsequent commit + push_branch then clobbered the victim PR. Recovery + disclosure in PR #2584. Three compounding bugs closed here: 1. checkout_pr dirty-tree guard had a first-call hole: the previous condition required ctx.toolState.issueNumber to already be set, so on workflow_dispatch runs the first checkout_pr (commonly from a subagent) bypassed the guard entirely. Now any PR switch with a dirty tree is refused, including the first switch of a run. Idempotent same-PR re-checkouts are still absorbed by alreadyOnBranch inside checkoutPrBranch. 2. push_branch trusted sticky pushDest blindly. Added a backstop: refuse pushes where the local branch matches /^pr-(\d+)$/ AND pushDest.remoteBranch differs from it AND the current run is not scoped to PR N (event.is_pr === true && event.issue_number === N). This catches subagent-induced silent branch switches even if a future bug reintroduces a first-call hole in fix #1. 3. Build-mode self-review prompt told the orchestrator to ship "the output of git diff" to the reviewer. The model in this run synthesized git diff main...HEAD, which excludes uncommitted work — and Build self-review runs BEFORE the commit, so the reviewer saw an empty diff and thrashed, eventually calling checkout_pr on a random PR to find something to look at. Prompt now specifies git diff origin/<base-branch> (two-dot, no HEAD), which compares the working tree against the remote base. Refs: zed-industries/cloud workflow run 26036155393 zed-industries/cloud#2582 (victim) zed-industries/cloud#2584 (disclosure) * review: key dirty-tree guard on current branch + drop 'two-dot' misnomer Address review feedback on PR #796. 1. checkout_pr dirty-tree guard now keys off the live current branch (git rev-parse --abbrev-ref HEAD), not ctx.toolState.issueNumber. issueNumber is ALSO set by get_issue / get_issue_comments / get_issue_events, so a subagent doing get_issue(N) followed by checkout_pr(N) on a dirty tree would have bypassed the original guard (issueNumber === pull_number). The current branch is the actual primitive for "would this call move HEAD" — querying it directly avoids correlating on toolState that other tools write to. 2. modes.ts: drop the wrong "two-dot" label on git diff origin/<base>. That's the single-rev form, not two-dot. Copilot was right that the label was confusing/contradictory with the actually-shown command. |
||
|
David Blass
|
2f1f136da8 |
modes: actually call resolve_review_thread on addressed PR feedback (#749)
* modes: instruct IncrementalReview + AddressReviews to actually call resolve_review_thread empirically (per #672 audit on 195 prod runs across 4 repos) only 1 in 195 runs called resolve_review_thread. IncrementalReview's prompt only mentioned prior reviews as a dedup filter — the agent had the data in hand but no instruction to retire addressed threads, so it almost never did. AddressReviews mentioned resolve as a one-line bullet at the end of step 6, which the agent treated as optional (the one observed AddressReviews run replied to 3 comments and resolved 0). IncrementalReview step 4 now: fetch prior reviews → for each open Pf-originated thread, decide if the new commits addressed it (anchor moved, isOutdated, or substantive concern resolved on a re-read), reply + resolve when addressed, leave open when uncertain. Conservative scope: only Pf-originated threads — human-reviewer threads stay theirs to mediate. AddressReviews step 6 now pairs reply + resolve in the same beat with explicit rules: resolve when you made the change OR replied substantively; do NOT resolve when you pushed back and the disagreement is unresolved. addresses #672. Co-authored-by: Cursor <cursoragent@cursor.com> * anneal: tighten auto-resolve decision rules correctness fixes from /anneal pass on IncrementalReview step 4 + AddressReviews step 6: - `[OUTDATED]` no longer "strong signal of address" — it just means GitHub moved the anchor (line shift / reformat / force-push); agent must re-read code at new location - explicit Pf-origin detection rule (first `comment author=pullfrog[bot]` tag), clarifies `*` marker is unrelated to thread root - explicit dual-ID separation: numeric `id=` for `reply_to_review_comment.comment_id`, GraphQL `thread=` for `resolve_review_thread.thread_id` (was silently 422-ing) - reformatter / partial-fix loophole closed: lines being modified isn't enough; all concerns in multi-concern comments must be addressed - AddressReviews push-failure path explicit: STOP and report_progress, do NOT reply or resolve when fix isn't live - step 4 → step 8 wiring rationale corrected (step 8 dedups by line range, not thread state) Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
Colin McDonnell
|
7e90e5cae6 |
Align Plan-mode prompts on report_progress as the canonical plan tool (#786)
* fix: align Plan-mode prompts on report_progress as the canonical plan tool Fixes #673. Three sites disagreed on where Plan output should be posted, letting a model synthesize a broken third interpretation (initial post via `report_progress({ target_plan_comment: true })`, which then misses the `existingPlanCommentId` precondition). This PR aligns all three on `report_progress` as canonical, with `target_plan_comment` reserved for revisions only: - `action/modes.ts` Plan step 4 — spell out that the initial plan post uses `report_progress` WITHOUT `target_plan_comment`, and that revisions go through `select_mode`'s PlanEdit override. - `action/mcp/comment.ts` `target_plan_comment` flag description — make the "revisions only" precondition explicit and call out the initial-post path by name. - `action/utils/instructions.ts` Progress reporting paragraph — drop the misleading "(e.g., Plan comments)" parenthetical that read as "use create_issue_comment for plans". `PlanEdit` (in `action/mcp/selectMode.ts`) was already correct and is unchanged. Intentionally out of scope (to keep the fix minimal): a `publish_plan` tool, removing the vestigial `create_issue_comment({ type: "Plan" })` branch, hardening the run-end cleanup guard for the `target_plan_comment but no existingPlanCommentId` fallthrough, and renaming `target_plan_comment`. * align create_issue_comment description with report_progress as canonical plan tool |
||
|
Colin McDonnell
|
0d7955d87d | add agent-browser fallback rule for unreachable chrome devtools mcp | ||
|
Colin McDonnell
|
3514bbc39f |
review prompt: tighten body-section bar + inline technical-details (#770)
* review prompt: tighten body-section bar + add inline technical-details Two layers of tightening to the Review/IncrementalReview prompts in PR_SUMMARY_FORMAT (and the per-mode aggregate-&-draft step): 1. Reframe inline-vs-body split. Body `### ` sections are now reserved for concerns that genuinely have no line to anchor to — absence, sequencing, design decisions, scope questions, architectural risk. Drop the "cross-cutting concerns" framing (misled the agent into either filing nothing in the body or filing multi-file anchored findings there). 2. Add a "Hunt for non-anchored concerns" sub-step to both Review (step 6) and IncrementalReview (step 8) aggregate phases. Diagnosis from PR #767's auto-review: on substantial PRs the agent surfaced findings but routed all of them inline, producing reviews with zero `### ` body sections even on diffs where non-anchored concerns clearly existed. 3. Replace the abstract `### ` example with a concrete non-anchored one ("Legacy `opencode.ts` has no documented deletion plan") so the agent pattern-matches the absence-shaped finding, not a line-bug. 4. Add an "Inline technical details" subsection to PR_SUMMARY_FORMAT so inline comments can carry a `<details>Technical details</details>` block when the fix has cross-file implications. Rename the existing "Agent details" inline collapsible to "Technical details" for consistency with body sections. 5. (Carried over from prior uncommitted work) Restructure the review metadata block from `<details>Review metadata</details>` into an HTML comment + an italic TL;DR commit-range line. The HTML comment keeps the metadata addressable for downstream agents without eating user-visible review real estate. No tests touched. * wiki: document multi-model end-to-end eval pattern |
||
|
Colin McDonnell
|
ddbc610569 |
review prompt: friendly green callouts + per-section severity emojis (#756)
* review prompt: friendly green callouts + per-section severity emojis - Replace `[!NOTE]` informational tier and the no-callout minor-suggestions tier with friendly green blockquotes (`> ✅` / `> 💡`). The two loud tiers (`[!CAUTION]` / `[!IMPORTANT]`) keep their GitHub admonitions. - Add a per-`##`-section severity-emoji rule (🚨/⚠️/💡/ℹ️) for cross-cutting review concerns that don't anchor to a line and would otherwise be buried in summary content. - Drop the `<br/>` between summary sections — heading + blank line carries enough visual spacing. - Skip the post-run learnings-reflection turn for `IncrementalReview`. It's the lowest-novelty mode (delta review against existing PR with prior summary already loaded) and almost never produces durable learnings — reflection there costs ~$0.50-0.80/run for nothing. - Surface real error info on `agent-browser` skill install failures (exit code + stdout + stderr + spawn error). The skills CLI uses a TUI that prints errors to stdout, so the prior stderr-only logging silently swallowed every failure. * review prompt: per-bullet severity emoji + bullets-only sections Section headings are plain again (no leading severity emoji). Severity moves to individual bullets so a section that mixes a 🚨 and a 💡 isn't mislabeled by either. Section bodies are now bullets only — paragraph prose under a heading is harder to scan and tends to bury the actionable point. Bullets can carry indented continuation content (sub-bullets, code fences, blockquotes) by indenting two spaces under the parent. * review prompt: cap section length + identifier discipline Bound each summary section to at most 4 bullets at most 2 lines each, and explicitly call out identifier-heavy prose as an anti-pattern. The reader is often a manager or non-author; identifier-dense paragraphs ('foo calls bar.fetch which dispatches to baz via qux...') are unreadable for them. Default to plain-language behavior descriptions, name an identifier only when it's the subject of an actionable concern or a public surface a reader would recognize, target 2-3 backtick tokens per bullet. Move the deep-explanation pattern from open blockquote to a default- collapsed details/summary so depth doesn't dominate the visible body. * review prompt: hard cap on bullet identifier density + worked rewrite example Soft 'aim for 2-3 tokens' guidance was ignored — first big-PR e2e showed 12 of 19 actionable bullets exceeded the target (avg 4.8 tokens, several over 8). Promote to a hard cap of 3 backticked tokens per bullet and pair with a concrete bad/good rewrite the agent can pattern- match against. Also tighten the per-bullet length cap from ~240 to ~200 chars and explicitly call it 'hard cap, not target'. * review prompt: tighten bullet length cap to 160 chars, dramatize the worked example V2 e2e test: token discipline improved (4.8 -> 3.3 avg, 12/19 -> 6/14 violations) but length got worse (235 -> 286 chars, 13/14 over the 200 cap). The agent compensated for fewer identifiers with more prose. Two changes: (1) tighten the cap from ~200 chars to 160 chars / 1 visual line and call out wrap-to-multiple-lines as the failure mode; (2) rewrite the worked example so the good version is genuinely half the length of the bad one, not just lower token count. The example was the thing the agent pattern-matches against; making the good version ~130 chars vs the bad version's ~290 chars sets the right shape. * review prompt: drop fixed bullet-count cap, keep length + identifier caps Per user feedback — section length should be governed by content, not an arbitrary count. Soft guidance ('past ~6, ask whether to split') is fine; the hard '≤ 4 bullets per section' rule was the wrong shape. Length cap (160c) and identifier cap (3 backtick tokens) stay; those target the actual scanability problem. * review prompt: drop ## subsystem sections, flat 'Issues found' list Per-section structure forced every concern into a subsystem frame and made the body read like a series of mini-essays. Replace with two parts: (1) TL;DR + Key changes as the dispassionate overview, (2) flat '### Issues found' list ordered by severity, intermixed across files and subsystems. Per-bullet rules (≤160c, ≤3 backtick tokens, severity emoji prefix, optional indented continuation) carry over unchanged. * review prompt: full v6 structure — preamble + cross-cutting H3s + nitpicks Replaces the flat 'Issues found' bullet list with the iterated v6 shape: - Preamble is a bolded inline 'Reviewed changes' lead-in plus bullets plus a collapsed 'Review metadata' block (mode/files/commits/refs/ reviewed commits list/prior pullfrog review/staleness note). - Each cross-cutting concern gets a '### emoji Title' section. The visible problem write-up is human-friendly and DESCRIBES THE PROBLEM ONLY — no asks, no suggested fixes, no 'the right thing to do is'. - Each section carries a collapsed 'Technical details' block wrapped in a 4-backtick markdown fence (so it can hold its own 3-tick code fences cleanly, agent-readable, one-click copyable). Standard four inner sections: Affected sites, Required outcome, optional Suggested approach, optional Open questions for the human. - '### ℹ️ Nitpicks' at the bottom for body-only nits that don't inline; simple bullets, no technical-details collapse. - Anti-paragraph-wall rule: never two successive plain paragraphs in visible '### ' sections; alternate prose with structure. - Inline-vs-body discipline: anything that anchors to a single line goes inline, body is for cross-cutting only. - Drops legacy '### Key changes', '### Issues found', '<b>TL;DR</b>', and the '<sub>Summary</sub>' line. * model effort: bump Gemini + GPT to high effort; drop Gemini Pro→Flash subagent E2E review eval against a substantive billing-module diff surfaced two related quality gaps: 1. Gemini Pro at thinkingLevel=medium (#663's CI-timeout fix) reviewed the diff only, took the 0-lens path, and missed a catastrophic camelCase/snake_case service-vs-schema mismatch. Bumping back to high — review work is exactly the wrong shape for the medium/high tradeoff #663 was optimizing for; the per-turn TTFT cost is worth paying when reasoning IS the value. 2. GPT had no reasoningEffort override, defaulting to upstream medium. Same diff, similar shallow result vs Claude. Adding reasoningEffort: high for the curated direct-OpenAI slugs, mirroring the Gemini pattern (Anthropic separately uses --effort high via the Claude Code CLI flag in claude.ts). 3. Gemini Pro's subagentModel was 'gemini-flash' — but Google has no in-between tier between Pro and Flash, and Flash is a meaningful capability cliff for review work. Dropping the override so subagents inherit Pro. Cost stays reasonable since Gemini Pro is already the cheapest of the flagship trio. Other providers unchanged: Anthropic opus→sonnet and OpenAI gpt→gpt-5.4 remain (each is a one-tier drop to a still-capable sibling). * model effort: revert orchestrator override, set explicit high on reviewfrog subagent Reshape the effort design after eval: - Drop the explicit Gemini and GPT model-level overrides — orchestrators now run at upstream defaults (Gemini high, GPT-5.x medium). Gemini's upstream IS high, so this is a no-op there; GPT goes back to upstream medium for orchestrator-level routing work. - Add explicit 'high' on the reviewfrog subagent via agent.options. OpenCode merge order is base ← model.options ← agent.options ← variant per session/llm.ts:141, so the subagent always runs at high regardless of which orchestrator dispatched it. Both thinkingConfig.thinkingLevel (Gemini) and reasoningEffort (GPT) keys included; irrelevant keys are ignored per provider. - Bump providers-live timeouts (12min job / 10min step, from 8/6) to budget for Gemini's TTFT variance at high effort. #663's 4min timeout was sized for the medium-effort override that's now removed. * model effort: restore Gemini explicit high override (no-override path breaks) Bare 'rely on upstream default' for Gemini failed in e2e — removing the model-level provider config produced 'Function call is missing a thought_signature' API errors on every gemini-pro run. Even though upstream opencode's options() returns the same thinkingLevel: high we were explicitly setting, opencode's resolution path differs subtly between the two cases. v2's explicit override worked; v3's removal broke. Reproducible across two consecutive runs. Restoring the explicit Gemini override (back to v2 design). GPT orchestrator stays UN-overridden — at upstream default (medium) — since removing that override didn't trigger the same failure pattern and the reviewfrog subagent agent.options high override compensates for the extra depth GPT loses at medium. * diag: remove reviewfrog agent.options to isolate Gemini thought_signature failure v3 (no Gemini orch override) failed with thought_signature error. v4 (restored Gemini orch override at v2-equivalent) ALSO failed, even though the orchestrator config matches v2. The variable between v2 (working) and v4 (failing) is the new reviewfrog agent.options block. Removing it to confirm — if Gemini works again, the agent.options addition is the culprit and we need a different shape for it. * opencode-ai: bump 1.1.56 → 1.15.0 + clean up gemini effort config opencode-ai@1.1.56 was published 2026-02-10 (3 months old). The Google API tightened thought_signature validation 24-48h ago (per https://discuss.ai.google.dev/t/gemini-thought-signature-patch/122555), and the bug class hits opencode's session→prompt serializer for MCP tool-call parts (anomalyco/opencode#4832, #8321). Latest stable bumps us through ~3 months of fixes; needed for Gemini-direct to stop dying with 'thought_signature is missing' on every multi-turn run. Companion cleanup: the gemini provider override in opencode.ts had 30-line block of comments, four unused constants, and a 6-line Object.fromEntries map for two entries. Replaced with one source-of- truth helper that loops modelAliases, filters provider==='google', strips the 'google/' prefix, and returns the override map. Adding any future Google alias to the registry now flows through automatically. Test added: action/agents/opencode.test.ts asserts the helper covers every direct-Google alias, strips the prefix correctly, and pins every entry to thinkingLevel high — catches drift in helper logic without hardcoding the API ids the test would have to update in lockstep with the registry. * fix(workflow): tolerate listJobsForWorkflowRun 404 in resolveRun PR #750 (docker testing rewrite) replaced the per-call env allowlist with full process.env passthrough into the test container. That now leaks GITHUB_RUN_ID + GITHUB_JOB into runs whose MCP token is scoped to a DIFFERENT repo (e.g. providers-live smoke runs the action against pullfrog/test-repo with pullfrog/app's run ID). The unconditional listJobsForWorkflowRun call 404s and crashes the entire run, breaking every providers-live job on main since #750 landed. jobId is purely cosmetic (deep-links 'View workflow run' footer to a specific job vs the run-level URL). Wrapping the API call in try/catch so a 404 logs a debug message and falls through to undefined jobId is the right fix — the failure mode is exactly what graceful degradation is for, and the alternative (filter the env vars at the docker boundary) re-introduces the kind of allowlist #750 was getting rid of. * opencode-ai: pin 1.14.51 instead of 1.15.0 (effect refactor breaks JSON output) opencode 1.15.0 (May 15) ships a major architectural refactor onto @effect — the run command boots an in-process server via @opencode-ai/sdk/v2 and the JSON event emission path through that SDK client doesn't surface on stdout the way our parser expects (CI run on 1.15.0 produced 0 stdout events but the agent still completed). Local invocation also hangs at the in-process server boot. The Gemini thought_signature fixes (the original reason for bumping) landed earlier in the 1.14.x line, so 1.14.51 (May 14) gets us the upstream fix without the Effect rewrite. Defer the 1.15.x bump until we're ready to rewire our parser/spawn around the new SDK. * opencode-ai: revert to 1.1.56; gha: filter outer-CI workflow-run vars at the docker boundary Two related changes for the docker testing harness's ergonomics: 1. Revert opencode-ai 1.14.51 → 1.1.56. The 1.14+ line ships an Effect refactor (the SDK-v2 client + in-process server architecture) that our --format json parser doesn't speak — even the 1.14.51 release, pre-dating the 1.15.0 Effect rename, produced 0 stdout events on our skill-invoke smoke. There's no clean pre-Effect version that ships the Gemini thought_signature fix; that fix needs a separate workstream once we're ready to rewire the parser onto SDK v2. 2. Filter outer-CI workflow-run identifiers (GITHUB_RUN_ID, GITHUB_JOB, GITHUB_WORKFLOW, GITHUB_ACTION, GITHUB_REF, GITHUB_SHA, etc.) from gha.ts's --env-file passthrough. PR #750's full-process.env design leaks pullfrog/app's CI run identifiers into runs that act against a different repo (e.g. pullfrog/test-repo); any code path inside the action that uses them as keys (most notably resolveRun's listJobsForWorkflowRun lookup) 404s. Filtering them here means the action sees undefined and skips the lookup, complementing the defensive try/catch in resolveRun (commit addc76d4). GITHUB_REPOSITORY and GITHUB_TOKEN are NOT filtered — those are genuinely needed. Companion to addc76d4 (resolveRun 404 tolerance). The two together make this class of bug 'either fix would have caught it' rather than 'silently breaks the entire test matrix'. * fix(deps): sync pnpm-lock.yaml with opencode-ai 1.1.56 manifest revert Forgot to refresh the lockfile after reverting the manifest in 02c6d8c1. CI's frozen-lockfile install was failing with 'lockfile: 1.14.51, manifest: 1.1.56' mismatch. |
||
|
Colin McDonnell
|
fa7ddcee4a |
prompt: discerning review-feedback handling + elegance bar
strengthen build-mode self-review and addressreviews step 4 to require verifying every reviewer finding, reject AI slop / over-defensive code, and frame the goal as a complete + minimal + elegant solution. mirror the elegance/no-slop bar in AGENTS.md. |
||
|
Colin McDonnell
|
5caeb75344 |
review: 0-or-2+ lens rule, parallel-or-bust, downshifted subagent models (#710)
* review: 0-or-2+ lens rule, parallel-or-bust, downshifted subagent models
PR review wall-time was dominated by two failure modes: orchestrator
serial-dispatching subagents (despite prompt asking for parallel) and
running every lens on the same Opus tier as the orchestrator. Sample of
recent runs showed 25-60min reviews on small PRs, with 8-10min idle
gaps between subagent dispatches.
Three changes:
1. `action/modes.ts` — replace the soft "1 trivial / 2-3 typical /
4-5 high-stakes" lens calibration with a binary 0-or-2+ rule. Default
is 0 lenses (orchestrator handles review solo with optional cheap
tracerfrog dispatches). 2+ parallel lenses only fire for substantive
PRs (>5 files AND >200 lines) or high-stakes-subsystem touches. Never
exactly one. Both Review and IncrementalReview prompts get loud
ALL-CAPS framing on parallel dispatch — emit ALL Task tool_use blocks
in a single assistant turn before reading any result. Drop the
"do NOT lens-review the diff yourself" advice; orchestrator pulls
context aggressively, in parallel with the lens fan-out.
2. New `tracerfrog` subagent for mechanical code tracing ("where is X
used / who calls Y / what depends on Z"). Pure read+grep+report with
no judgment — orchestrator can dispatch many tracers cheaply in
parallel. Defined in `action/agents/reviewer.ts`. Wired into both
claude.ts (`--agents` JSON) and opencode.ts (`agent` config block).
3. Per-subagent model downshifts via `deriveSubagentModels`:
- Anthropic: reviewfrog → Sonnet, tracerfrog → Haiku
- OpenAI: both → gpt-5.4-mini
- other providers (xai, deepseek, gemini, etc.): inherit (no
standard tier triplet to downshift to)
Claude Code path always runs Anthropic so the downshift is hardcoded
inline in claude.ts. OpenCode uses the helper since orchestrator
provider varies.
Both runtimes' subagent-definition formats verified directly against
their source: `--agents` JSON `model` field (claude-code's
`AgentJsonSchema` accepts model+effort+maxTurns+more) and OpenCode's
`agent.{name}.model` config field (parsed via Provider.parseModel,
applied per-task in tool/task.ts line 92). Parallel dispatch is
infra-supported in both — only the orchestrator model's tool_use
emission pattern was the bottleneck.
Tests: subagentModels.test.ts (14 tests covering provider matrix),
subagentRegistration.test.ts (6 source-asserts catching shape
regressions in buildAgentsJson / buildReviewerAgentConfig).
* subagentModels: add openrouter routes (proxy/router mode)
Initial helper missed the openrouter prefix used by Pullfrog's router
proxy. preview-710 e2e showed the OpenCode + openrouter path receiving
no downshift — orchestrator and lenses both ran on opus-4.7 because
'openrouter/anthropic/claude-opus-4.7' didn't match any of the
anthropic/openai prefixes the helper checked.
Add explicit branches for 'openrouter/anthropic/...' (uses dot notation:
claude-sonnet-4.6 / claude-haiku-4.5) and 'openrouter/openai/...'
(gpt-5.4-mini for both reviewer and tracer). Same opus->sonnet,
sonnet->keep-but-haiku-tracer, haiku->no-op semantics as the direct
anthropic path.
* opencode: log resolved subagent models at startup
So we can verify per-subagent model overrides actually take effect at
runtime. Prints once per run alongside the existing model/effort log
lines.
* drop tracerfrog: keep reviewfrog only, LSP-powered tracer planned later
Removes the cheap-haiku-tracer subagent (TRACER_AGENT_NAME +
TRACER_SYSTEM_PROMPT, registrations in claude.ts/opencode.ts, dispatch
guidance in modes.ts). The mechanical-tracing use case will be served
better by an LSP-powered tool than by a separately-prompted subagent.
deriveSubagentModels collapses to a single { reviewer } shape; the
reviewfrog-on-Sonnet downshift stays. Same source-assert + provider-
matrix tests, minus the tracer-specific cases.
modes.ts wording: drop the 'subagent type cheat sheet' bullet, drop
the parenthetical 'often better served by tracerfrog than reviewfrog'
on the impact lens, drop tracerfrog from the same-turn-context-pulling
hint. The 0-or-2+ rule and ALL-CAPS parallel emphasis are unchanged.
* subagentModels: broader downshift coverage (gpt-pro, gemini-pro, grok); drop gpt-mini target
Scanned every resolved orchestrator slug in action/models.ts against
models.dev pricing data. Identified five clear cases where the
orchestrator is meaningfully expensive AND has a cheaper sibling that
remains capable enough for review-style judgment work.
Changes:
- Anthropic: opus → sonnet (kept; -40%)
- OpenAI: gpt → gpt-5.4 (was: gpt-mini; -54% instead of -85% but
preserves review-quality judgment — gpt-mini was too dumb)
- OpenAI: gpt-pro → gpt (NEW; -93%, biggest single unlock —
gpt-5.5-pro is $30/Mtok in vs gpt-5.5 at $5)
- Google: gemini-pro → gemini-flash (NEW; -75%)
- xAI: grok-4.3 → grok-4-1-fast (NEW; -80%)
Every branch handles the three routes in use: direct provider slug,
opencode-vendored, and openrouter-proxied. Variants below the downshift
target (mini/nano/flash/fast/sonnet/haiku) inherit (no further drop).
Skipped:
- DeepSeek: v4-flash ($0.14/Mtok) is too far below review judgment
threshold; v4-pro orchestrator already cheap ($0.55 blended).
- Moonshot: kimi-k2-thinking would only save 32% and slug stability on
OpenRouter is uncertain; revisit if cost matters.
- o3: already mid-tier in OpenAI's reasoning family; no clean target.
* models: hoist subagent downshift into the registry, add hidden flag
The downshift relationship now lives next to each alias's resolve /
openRouterResolve as a sibling field. Two new ModelDef fields:
- subagentModel?: string — alias key (within same provider) of the
cheaper sibling reviewfrog should use as a lens-fanout subagent.
e.g. claude-opus → 'claude-sonnet'.
- hidden?: boolean — exclude from selectable lists (UI dropdown,
CLI init picker). Does NOT affect resolution; for that use
fallback. Used so internal-only subagent targets like openai/gpt-5.4
exist in the registry but never appear as a user-facing pick.
Wiring:
- anthropic.claude-opus → claude-sonnet (-40%)
- openai.gpt-pro → gpt (-93%, biggest unlock)
- openai.gpt → gpt-5.4 (-54%); gpt-5.4 added with hidden:true
- google.gemini-pro → gemini-flash (-75%)
- mirrored across opencode + openrouter providers (each provider
declares its own three-route data so the downshift declaration
is colocated with the rest of the alias definition).
deriveSubagentModels collapses from ~85 lines of prefix-matching to
a ~15-line registry reverse-lookup: find the alias whose resolve OR
openRouterResolve matches the orchestrator's spec, follow its
subagentModel pointer, return the matching field of the target alias.
Filter sites updated:
- components/ModelSelector.tsx: !a.fallback && !a.hidden
- action/commands/init.ts: same
Tests rewritten to exercise the registry through the public surface;
the matrix collapses to one assertion per (provider × route) pair.
* TEMP: log per-step cost+tokens for subagent model verification (PR #710)
* TEMP: also log SUBAGENT step_finish from bus envelope handler
* remove temporary per-step diagnostic logs (verification done)
Verified subagent model downshift takes effect end-to-end on the OpenCode
+ openrouter path. PR #8 in pullfrog/preview-710-review-perf dispatched
3 lenses (billing-subsystem / security / correctness) on the orchestrator's
opus-4.7 session, and per-subagent step_finish events showed actual cost
exactly matching Sonnet pricing rates (60% of what Opus would have cost):
session n actual if-Opus if-Sonnet match
T3VrUuF... 5 $0.2425 $0.4042 $0.2425 Sonnet ✓
93ZZR7E... 4 $0.2253 $0.3754 $0.2253 Sonnet ✓
Fb1Kr7b... 4 $0.2495 $0.4158 $0.2495 Sonnet ✓
The startup '» subagent models: reviewfrog=...' line stays — useful
permanent diagnostic showing the resolved subagent model per-run.
* TEMP: log per-event model from claude.ts assistant handler
* remove temporary per-event model log (claude.ts verification done)
Verified subagent model downshift takes effect end-to-end on the Claude
Code path. PR #9 in pullfrog/preview-710-review-perf dispatched 2 lenses
on an opus-4-7 orchestrator. Per-assistant-event model field from the
SDK's stream-json output, partitioned by parent_tool_use_id:
ORCH (parent_tool_use_id=null): 17 events all model=claude-opus-4-7
SUBAGENT lens:billing-subsystem: 17 events all model=claude-sonnet-4-6
SUBAGENT lens:security: 21 events all model=claude-sonnet-4-6
Zero leakage to opus from either subagent session. The per-subagent
'model' field in --agents JSON is honored by claude-code at the SDK
level, identical to the OpenCode path verified earlier.
* opencode: bump per-call output cap 5K → 16K to unblock large reviews
The 5K cap (added in #616 to lower OpenRouter upfront budget reservation
for low-wallet runs) was capping the entire response of a single LLM call,
not just the budget reservation. A single tool_use response — like a
`create_pull_request_review` with many inline comments — would truncate
mid-stream past 5K output tokens, leave the JSON unparseable, and the tool
would never actually invoke. We hit this on PR #710's verify-downshift PR:
review aggregated from 3 lenses had 11 inline comments + a long body,
truncated at out=5000 on every retry attempt, action exited with 'Review
mode finished without calling create_pull_request_review after 3 retry
attempts'.
Investigated whether OpenCode (or OpenAI/Anthropic/OpenRouter directly)
exposes a separate budget-reservation parameter that could stay small
while letting the response exceed it. They don't — `max_tokens` /
`max_completion_tokens` is the single value all four use for both the
upfront reservation and the hard output ceiling. No way to decouple them
at the API surface.
Bumped to 16K as a middle ground: 8× the prior cap (handles every review
shape we've observed plus headroom), still half of OpenCode's 32K default
so the wallet-burn benefit for low-balance accounts is preserved, just
smaller. For Opus 4.7 a typical ~50K-input call now reserves roughly
$0.65 instead of the prior $0.38.
Updated the constant comment to spell out the trade-off clearly so this
doesn't happen again.
* opencode: drop OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX override entirely
Verified the original rationale for the override is obsolete. From #616
the cap shrunk OpenRouter's per-call upfront budget reservation so a
single call's reservation wouldn't exceed the per-run key cap
(`ROUTER_PER_RUN_LIMIT_USD = 25`) and lock low-balance accounts out of
starting a run.
That per-run gate is gone. `app/api/proxy-token/route.ts` ~line 422
explicitly says: 'No upper cap (the old ROUTER_PER_RUN_LIMIT_USD = 25 is
gone). The natural ceiling is whatever the user has + their buffer.'
Router now mints keys with `keyLimitCents = balance + buffer` ($50 for
autoreload+card, $5 for card-only, $0 for no-card). A single call's
upfront reservation fits comfortably within that — no separate per-call
gate to fail past.
The cap had a real downside as a hard per-call output truncation. A
single `create_pull_request_review` tool_use with many inline comments
would truncate mid-stream past 5K output tokens, the JSON would be
unparseable, and the tool never invoked. Hit on PR #710's
verify-downshift PR.
Removing the override entirely; OpenCode falls back to its 32K default.
Left an explanatory note above the env-var assignment site so the next
person doesn't unknowingly re-add it.
|
||
|
Colin McDonnell
|
4260984257 |
attribute claude subagent log lines + per-session thinking timer; tighten lens calibration (#700)
* attribute claude subagent log lines + per-session thinking timer; tighten lens calibration three orthogonal fixes diagnosed from the 10m PR-699 review run: 1. wire SessionLabeler into the Claude Code harness. claude-agent-sdk stamps every Assistant/User/System message with session_id and a non-null parent_tool_use_id when emitted from a subagent context, so the same FIFO labeler the OpenCode harness uses works here too. parallel reviewfrog dispatches now log with [lens:correctness] / [lens:operational-readiness] / etc. prefixes instead of being indistinguishable from the orchestrator. matches both "Task" and "Agent" tool names per the v2.1.63 rename. 2. one ThinkingTimer per session. the global timer treated cross-session interleaving (parent thinks → child tool_call, child returns → parent dispatches next) as parent thinking time, so individual "thought for Xs" numbers were untrustworthy. each session now owns its own timer and prefixes its own log line. 3. tighten the Review/IncrementalReview lens-add discipline. PR-699 triggered 4 lenses on a typical refactor (no auth/billing/schema) when the prompt's own calibration says 2-3 is typical; the research-validated lens went deep on Resend idempotency window + prisma updateMany lost-updates without either being load-bearing. adds an explicit "name the failure mode this lens would catch that the diff plausibly introduces" bar, and tightens research-validated specifically: only when correctness depends on the third-party contract, not when the API is merely used. side benefits from #1: subagents' TodoWrite events no longer clobber the orchestrator's progress comment; subagent text no longer overwrites finalOutput; system-event handler safely routes through eventLabel even though SDK only emits system:init for the top-level query today. * fix node strip-only mode: declare formatLine as field, not parameter property * key claude subagent labels by parent_tool_use_id, not session_id claude-agent-sdk runs subagents inside the orchestrator's session — they share session_id — and stamps subagent messages with parent_tool_use_id pointing at the Agent tool_use that spawned them. e2e on PR-700 with preview-700-claude-labeling#1 confirmed the original session_id-keyed wiring never differentiated subagent activity (only the dispatch line got [lens:correctness] in the log; the subagent's reads, writes, and todos all rendered as orchestrator). extend SessionLabeler so labelFor accepts an optional parent_tool_use_id and short-circuits to a direct map keyed by Agent tool_use id when set. recordTaskDispatch optionally takes the Agent tool_use id (block.id at dispatch time) and binds it. orchestrator events keep flowing through the sessionID/FIFO path unchanged so opencode wiring is untouched. * drop weak timer test that asserted only field isolation per pullfrog review on PR-700: the 'two timers do not bleed timestamps' test only verified that two ThinkingTimer instances have separate private fields, which has always been true. doesn't earn its keep — the per-session behavior is exercised by integration through claude.ts + opencode.ts. |
||
|
Colin McDonnell
|
a4a5010441 |
gemini-3: default thinkingLevel to medium + restrict eager prep to frozen install (#663)
* gemini-3: default thinkingLevel to medium + don't `npm ci` without a lockfile upstream opencode hardcodes `thinkingLevel: "high"` for every gemini-3 model on the direct google SDK (see `packages/opencode/src/provider/transform.ts` `options()`). that added 30-60s of pre-tool-call TTFT and 5-46s of post-tool jabber per turn, which is overkill for the tool-routing decisions that dominate agentic loops — and the variance caused the `providers-live (google/gemini-pro)` smoke job to time out at 4 minutes (see job 75405504847 on run 25684766415). three changes: - inject `provider.google.models.<api-id>.options.thinkingConfig.thinkingLevel = "medium"` for the two curated gemini-3 slugs in `buildSecurityConfig`. deep-merges over the upstream default; explicit `--variant high` / user opencode config still wins. flash stays at medium too — low-effort flash is visibly worse and the latency win isn't meaningful (flash is already fast). - bump the `providers-live` harness step from 4 → 6 minutes. the job-level 8-minute cap stays as the upper bound, but gemini's intrinsic TTFT variance was eating most of the 4-minute slack on its own. - in `installNodeDependencies`, pick `frozen` only when a lockfile was actually detected. previously a package.json-only repo (like the smoke fixture's `pullfrog/test-repo`) always triggered `npm ci` and emitted a noisy `EUSAGE` error before falling through. * prep: skip eager install when neither lockfile nor `packageManager` field present the previous commit changed the no-lockfile path from `npm ci` (always errored `EUSAGE`, never wrote any artifact) to a successful `npm install`, which had an unintended side effect: it generated `package-lock.json` in the working tree, tripping the post-run dirty-tree gate. the agent then committed the lockfile and opened a real PR — and in the openai/gpt smoke run on PR #663, the agent overwrote the `SMOKE TEST PASSED` output with the PR URL, failing the smoke validator. a repo with `package.json` but no lockfile and no `packageManager` field has not committed dependency state. eagerly installing produces state the repo doesn't track, which is the dirty-tree problem above. skip the eager install entirely in that case; the agent can opt in via `await_dependency_installation` when it actually needs deps. repos with a lockfile or a `packageManager` field keep the existing frozen-install behavior unchanged. * post-run: suppress dirty-tree gate in non-committing modes (Review / IncrementalReview / Plan) the dirty-tree post-run gate currently fires for every mode and tells the agent to commit and push whatever is in the working tree. that's wrong for modes that complete by submitting a review (`Review` / `IncrementalReview`) or posting a Plan comment (`Plan`) — those modes never touch files as part of their contract, so any tree dirt at end-of-run is incidental tool noise on an ephemeral worktree. nudging the agent to commit it can produce a spurious PR, as seen in the openai/gpt smoke run on PR #663 where a stray `package-lock.json` from `npm install` led the agent to open pullfrog/test-repo#32 and overwrite the smoke output. introduce `NON_COMMITTING_MODES` in `action/modes.ts` and consult it in `collectPostRunIssues`. when the selected mode is read-only, log the suppression for visibility but skip populating `issues.dirtyTree`. modes that legitimately commit (`Build`, `AddressReviews`, `Fix`, `ResolveConflicts`, `Task`) keep the existing nudge. * prep: restore eager frozen-install, drop non-frozen fallback eager dependency prep is non-mutating by contract — it runs before the agent starts and any artifact it leaves in the tree (e.g. a generated `package-lock.json`) trips the dirty-tree post-run gate and can lead the agent to open a spurious PR (seen on the openai/gpt smoke run earlier in this PR). revert the previous skip-when-no-lockfile branch: that was the wrong layer to enforce the invariant. instead, run `frozen` (`npm ci` / `pnpm install --frozen-lockfile` / etc.) unconditionally and drop the `|| install` fallback that could silently mutate the tree when `frozen` is missing. frozen commands fail cleanly without writing artifacts when there's no lockfile, which is exactly the safety contract we want. repos that need a real install must opt in explicitly via a `setup` lifecycle hook. * review nits: single getGitStatus call, tighten gemini-3 override scope comment addresses two inline nits from the PR review: - `collectPostRunIssues` was calling `getGitStatus()` (spawns `git status --porcelain`) in both branches of the mode check. lift the call above the conditional and branch on the result; same behavior, one git invocation. - the JSDoc on `GEMINI_3_DIRECT_API_IDS` said the override applies "across the board," but the constant only covers the two curated slugs in `action/models.ts`. tighten the wording to call out that other gemini-3 ids in models.dev keep the upstream "high" default. skipped the bot's yarn-1 concern after reading yarn 1's `install.js`: `bailout()` (lines 461-465) throws `frozenLockfileError` when `frozenLockfile && (!lockfileClean || missingPatterns.length > 0)`, which fires before `linker.init()` writes node_modules or runs lifecycle scripts. the existing comment's claim that frozen commands fail without artifacts holds for yarn 1 too. |
||
|
Colin McDonnell
|
cf94773bf0 |
modes: make task-list authoring the explicit first step in every mode checklist (#665)
* modes: make task-list authoring the explicit first step in every mode checklist The system prompt already instructs the agent to author an internal task list at the start of every run (action/utils/instructions.ts:291), but the rule lives several hundred tokens above the agent's first decision point and references the mode's checklist before the agent has it. Compliance is roughly coin-flip across opus runs — PR #610 dead-air for 9m20s was the extreme case; my own #664 e2e runs split 1-for-1 on `todowrite` compliance. Putting the directive *inside* the checklist that `select_mode` returns co-locates instruction with referent at the moment the agent decides what to do next. Same vocabulary as the existing rule (`task list`, agent-agnostic; the harness already maps to `todowrite`/`TodoWrite` per-agent in agents/opencode.ts and agents/claude.ts). The directive is deliberately non-prescriptive about list contents — the agent authors items based on the work it's about to do, not from a hand-shaped template. Touches all 8 built-in modes and the PlanEdit override: - Build / AddressReviews / Review / IncrementalReview / Plan / Fix / ResolveConflicts / Task: inserts `1. **task list**: create your task list for this run as your first action.` and renumbers existing steps. - action/mcp/selectMode.ts: same insertion in the PlanEdit override checklist. - All internal step cross-references shifted +1 (`step 5` → `step 6`, `skip steps 3–4` → `skip steps 4–5`, etc.) across Review, IncrementalReview, and ResolveConflicts modes. One code-comment reference in IncrementalReview's preamble updated to match. Complements #664 (live progress streaming): streaming guarantees the user sees *something* regardless of compliance; this PR raises the ceiling on what they see when the agent does comply (clean numbered checklist tracking through the run instead of just the latest assistant message). 488 action tests pass; typecheck, lint, format all clean. * postRun: fix stale 'step 7' reference missed during +1 renumbering |
||
|
Colin McDonnell
|
ef394277c1 |
review: synthesize [!NOTE] informational tier with #644 alert judiciousness — 4-callout visual ladder + approved Fix-gate (#653)
* review: NOTE-tier callout + `actionable` flag to suppress Fix buttons
Adds an `actionable` parameter to the `create_pull_request_review` tool
(defaults true) so the agent can opt out of the Fix-it/Fix-all/Fix-👍s
footer affordance on informational reviews. Threaded through
`createAndSubmitWithFooter` so the buttons are omitted when
`actionable: false`.
Updates `Review` and `IncrementalReview` mode prompts with a 4th tier:
`> [!NOTE]` + `actionable: false` for mergeable, FYI-style observations
(prior feedback addressed cleanly, minor stale doc reference, etc.).
Calibration note: `[!IMPORTANT]`/`[!CAUTION]` are reserved for findings
that warrant code changes, because that's what trains users to click
Fix. `[!NOTE]` reviews must not carry inline comments — if a point is
concrete enough to anchor to a line, upgrade the whole review tier.
* review: drop redundant `actionable` flag, key Fix buttons off `approved`
`approved` already encodes "this PR is mergeable, nothing for the Fix
button to act on" — `actionable` was a second flag carrying the same
signal. Drop it from the tool schema and `FooterOpts`; the footer gate
stays `if (!opts.approved)` (unchanged from pre-PR behavior, with a new
comment documenting the UX rationale).
NOTE-tier reviews now use `approved: true` + `> [!NOTE]` body instead of
`approved: false` + `actionable: false`. For repos with
`prApproveEnabled: false`, the runtime already downgrades APPROVE to
COMMENT, so the GitHub-side shape is identical to the prior design.
* review: address Pullfrog feedback — drop ambiguous parenthetical + update postRun nudge
- Review-mode calibration: drop the "(or no callout at all)" parenthetical
that didn't map cleanly to a bullet; replace with explicit "both the
`[!NOTE]` tier and the 'no actionable issues' tier below use approved:
true" so the bullet-list anchor is obvious.
- `buildUnsubmittedReviewPrompt` (Review mode): the fallback nudge for
unsubmitted reviews now defers to the mode prompt's tier matrix and
acknowledges that `> [!NOTE]` informational reviews submit with
`approved: true` alongside the canonical "No new issues found." path.
Previously the nudge only described the pre-NOTE binary world.
|
||
|
David Blass
|
ee479474ce |
action: tighten review alert judiciousness in prompts (#644)
The Review and IncrementalReview prompts unconditionally wrapped any non-critical review body in `> [!IMPORTANT]`, even for trivial nits or "rough edge" observations. The result is alert fatigue — full-width colored callouts dominate the page when the actual finding is a single JSDoc tweak. Adds an explicit judiciousness preamble to both Review step 5 and IncrementalReview step 7, and splits the prior single non-critical tier into two: - must-address non-critical (`[!IMPORTANT]`) — gated on real consequences if shipped (incorrect behavior, missing validation, regressions the author should fix before merge) - minor suggestions only (no alert) — single-line nits, doc/comment polish, defer-able observations, "rough edges" Critical tier wording also tightened to spell out the bar (`bugs, security, data loss, broken core flows`). Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
David Blass
|
10aeaf8c11 |
action: dedupe identical reply_to_review_comment calls within a session (#623)
* action: dedupe identical reply_to_review_comment calls within a session PR #610 reproduced a Kimi K2 stutter where the agent's tool_use surface showed one `pullfrog_reply_to_review_comment` call but GitHub recorded two byte-identical POSTs 3s apart, leaving a duplicate response on `action/mcp/review.ts:14`. Add `duplicateReplyDecision` (mirrors `duplicateReviewDecision`) and track per-session replies on `ToolState.reviewReplies`, keyed by parent `comment_id` + `bodyWithFooter`. Identical re-emissions short circuit with `{ skipped: true, reason }` instead of POSTing again. Body-keyed (not just id-keyed) so legitimate follow-up replies with different content still go through. Tighten `AddressReviews` step 5 to say *exactly once per comment* and note that the runtime dedupes identical bodies, so the agent has both prompt-level guidance and a server-side guarantee. Co-authored-by: Cursor <cursoragent@cursor.com> * address review: drop stale file ref in dedupe comment; soften tool description * remove comment.test.ts --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: pullfrog[bot] <226033991+pullfrog[bot]@users.noreply.github.com> |
||
|
Colin McDonnell
|
85d25a6fe6 |
post-run gate: fail review-mode runs that don't submit a review or progress (#638)
* post-run gate: fail the run when review mode finishes without a review or progress
review-mode runs that ended in a text-only assistant turn ("now I have enough
to draft the review...") were silently swallowed: the progress comment was
deleted by stranded-comment cleanup and no review appeared on the PR. user-
visible result was identical to "the agent never ran." caught in
https://github.com/pullfrog/app/actions/runs/25583698781.
new post-run gate alongside stopHook / dirtyTree / summaryStale: derived
inline from toolState (selectedMode in {Review, IncrementalReview} && !review
&& !finalSummaryWritten && hadProgressComment) — no parallel toolState flag.
when it fires, the resume prompt nudges the agent to call either
create_pull_request_review or report_progress; persistent failure after
MAX_POST_RUN_RETRIES surfaces as AgentResult.error.
also: when the post-run loop returns success=false, write the error to the
progress comment before the stranded-comment cleanup runs, and skip the
delete in that case. previously a !success run from the loop would lose the
error message into the void.
IncrementalReview's trivial-skip branch now calls report_progress with a
brief "no review warranted" note instead of exiting silently — keeps the
contract symmetric with the gate and gives the user a visible signal even
on no-op review runs.
documents the literal-record design rule on the ToolState interface so
future fields don't drift back into derived/absence-encoding state.
* review feedback: mode-aware nudge, gate-error preservation, prompt order
addresses three findings from the auto-review on this PR:
1. Review mode nudge no longer offers `report_progress` as an exit. Review
mode's contract (modes.ts step 5) forbids it; the gate previously sent
contradictory copy. IncrementalReview's nudge still offers both since
its trivial-skip path legitimately allows `report_progress`.
2. `writeJobSummary` is now wrapped in try/catch on the success-path
cleanup. without this, a throw there jumped to the outer catch and
overwrote the gate's failure message in the progress comment with the
(less actionable) writeJobSummary error — restoring exactly the
invisible-failure UX this PR fixes. step-summary writes are
informational; let them fail silently.
3. `buildPostRunPrompt` reorders gates to match the terminal hard-fail
order: `stopHook` → `unsubmittedReview` → `dirtyTree` → `summaryStale`.
when both hard-fail gates co-fire (rare in review modes), the prompt's
emphasis now matches the user-visible failure message.
new test asserts the IncrementalReview nudge offers both exits while the
Review nudge offers only `create_pull_request_review`. e2e validation
already passed against pullfrog/preview-638-review-stop-hook PR #1
(gate fired once; agent recovered on second turn).
* mode-aware terminal error copy
second auto-review caught a residual contradiction: the terminal hard-fail
error string reported "create_pull_request_review or report_progress" for
both modes, even though the new mode-aware nudge tells Review-mode agents
"Review mode does not have a no-submit exit". the error message now mirrors
the nudge — Review names only `create_pull_request_review`,
IncrementalReview lists both. additional Review-mode hard-fail test asserts
the absence of `report_progress` in the error.
|
||
|
Colin McDonnell
|
d6de1c369a |
learnings: edit-in-place tmpfile (drop update_learnings tool) (#635)
* learnings: edit-in-place tmpfile (drop update_learnings tool)
learnings now follow the PR-summary file pattern: server seeds
`pullfrog-learnings.md` from `Repo.learnings` at startup, agent reads
it as part of context, may edit in place during the post-run reflection
turn, server reads back at end-of-run and PATCHes if changed.
motivation: `update_learnings` required the agent to pass the FULL
merged list as a string parameter — an output-token tax that grew
linearly with the learnings size, and a constant prompt-context
expansion since the contents were also inlined into the LEARNINGS
section. for repos with mature learnings the prompt was getting
visibly noisy in CI logs.
key changes:
- new `action/utils/learnings.ts` (seed/read helpers + 10k cap)
- `main.ts`: always seed; `persistLearnings` mirrors `persistSummary`
(success path, error path, exit-signal handler, idempotent guard,
byte-trim equality skip); forwards `model` for `LearningsRevision.model`
- `LEARNINGS` prompt section now contains only the file path + a
one-line "read it" instruction (no contents inlined)
- `update_learnings` MCP tool deleted; `action/mcp/learnings.ts` removed
- reflection turn (`buildLearningsReflectionPrompt`) reframed around
file editing with explicit prune-stale + leave-alone-if-nothing-new
framing
- `learningsStep` removed from every mode checklist — surface lives only
in the LEARNINGS prompt section + the reflection turn now
* learnings: harden seed step + refresh stale docs (review feedback)
Three findings from PR review, all implemented:
1. wrap learnings seed in best-effort try/catch (action/main.ts) —
the always-on seed block ran unconditionally and an unwrapped
`seedLearningsFile` (mkdir + writeFile) failure (ENOSPC, EACCES,
hostile sandbox) would unwind into the outer main() catch and flip
an otherwise-successful run to "❌ Pullfrog failed" before the
agent even started. asymmetric with `persistLearnings`'s own
best-effort contract. wrap and log on failure; downstream
consumers (`persistLearnings`, agent harnesses, `resolveInstructions`)
already handle `learningsFilePath: undefined` cleanly.
2. refresh wiki/main.md — `resolveInstructions` parameter renamed
from `learnings` to `learningsFilePath` in this PR; the data-flow
diagram and the resolver dependency table both still showed the
pre-refactor signature.
3. drop deleted `learnings.ts` from ROADMAP.md + RESEARCH.md
"missing MCP tool tests" bullets — `action/mcp/learnings.ts` was
removed in this PR; the bullets are otherwise still accurate.
|
||
|
Colin McDonnell
|
e4e93ea6d3 |
PR summary as agent-edited tmpfile snapshot (#568)
* PR summary as agent-edited tmpfile snapshot Replaces the comment-based PR summary path (and the in-progress update_pr_summary tool from #534) with a snapshot file the agent edits in place during Review / IncrementalReview / pr-summary Task runs. The server seeds the tmpfile with the previous snapshot (incremental) or a stable scaffold (first run), exposes the path via select_mode, and reads it back at end-of-run to persist to WorkflowRun.summarySnapshot and (when the prSummaryComment toggle is on) splice into the PR description body. Why a tmpfile rather than a tool call: incremental snapshot edits are output-token-cheap when the agent uses native file-editing tools, and range-diff cleanly across runs because section headings are stable. The agent never has to regurgitate the full snapshot to update it. Gating: snapshot generation is opt-in via either prSummaryComment="enabled" (splice into PR body) or prReReview="enabled" (snapshot feeds future incremental review runs as context). Users who disable both pay nothing end-to-end — no seeding, DB write, or body splice. Behavior changes: - Drop the Summarize mode and the Summary comment type entirely; the rolling summary is no longer a separate run shape. - pull_request_synchronize with re-review off and summary on still dispatches a silent pr-summary Task, but it edits the snapshot file instead of posting a fresh comment. - /api/repo/.../pr/.../summary-comment now returns { snapshot: string | null } from the DB instead of fetching a comment via GraphQL. URL kept stable so deployed older actions degrade gracefully. - summaryCommentNodeId is retained on WorkflowRun for legacy data and a future backfill of pre-snapshot comment-based summaries. Supersedes #534. The commit-tool/sub-agent direction in that PR is abandoned in favor of this file-based shape. * address review pass #1: synchronize fallback, splice idempotency, docs * address review pass #2: in-flight skip should not race summary fallback * address review pass #3: signal-handler flush, doc clarifications * address review pass #4: in-flight persist promise + bounded body-splice timeout * address review pass #5: defensive catch on persist worker, doc nit * add summary-stale post-run gate When generateSummary is set, we capture the bytes of the seeded snapshot file and pass them to the agent's post-run loop alongside the file path. After each agent attempt, the loop diffs the current file against the seed; if they're byte-identical the agent never touched it, and we nudge once via a resume turn (similar to the dirty-tree gate, but soft and fire-once so smaller models that legitimately decide no edit is warranted don't burn the retry budget). Mostly defends against forgetful smaller models on the Review path — their mode prompt asks them to edit the snapshot file, but the multi-step instruction can fall through when the diff is large. * trigger: retry vercel preview build * fix(action): drop unused re-export that pulled node:fs/promises into next bundle action/internal/index.ts was re-exporting DEFAULT_PR_SUMMARY_INSTRUCTIONS from action/utils/prSummary.ts, but nothing in the next.js app imports it. prSummary.ts uses node:fs/promises, and pullfrog/internal is aliased into the next bundle by next.config.ts, which made turbopack try to resolve node:fs/promises in client chunks and fail with: the chunking context (unknown) does not support external modules (request: node:fs/promises) drop the re-export — selectMode.ts (the only real consumer) already imports it directly from action/utils/prSummary.ts. * firewall PR summary snapshot from user instructions; resurrect rich format for Review The agent-internal snapshot (the markdown file the agent edits in place across runs) is exclusively durable context for future agent runs — user-supplied summarization instructions warp it and degrade that context. Drop the prSummaryCommentInstructions read path end-to-end: - handleWebhook: stop reading prSummaryCommentInstructions, stop passing prSummaryInstructions through dispatch options - action payload + ToolState + selectMode addendum: drop the instructions appendix; the snapshot prompt is fixed, not user-shaped - TriggersSettings: drop the InstructionsEditor for prSummaryCommentInstructions - prSummary.ts: reframe DEFAULT_PR_SUMMARY_INSTRUCTIONS as agent-targeted (durable context, not human-facing prose) Prisma columns (prSummaryComment, prSummaryCommentInstructions) and the matching zod schema entry stay for graceful retreat. Separately, resurrect PR_SUMMARY_FORMAT (deleted along with the Summarize mode in the original PR) and wire it into Review mode only. Initial PR reviews now include a structured summary section in the review body using the rich format (TL;DR, key changes, ## sections with before/after, file-link trails). IncrementalReview keeps its existing terser bullet-list shape since re-review bodies are deltas, not introductions. The user-facing review summary and the agent-internal snapshot are deliberately separate artifacts with separate prompts and zero shared content. * address review comments: prompt self-consistency + stale-doc cleanup PR 568 self-review (4232488109) flagged a self-contradiction the firewall commit introduced and three stale doc references that survived. - action/modes.ts: Review-mode step 2's trivial-PR shortcut said `submit "Reviewed — no issues found." per step 5`, but step 5's rewrite removed exactly that preamble. Aligned both: trivial PRs and no-actionable-issues PRs now produce a body that opens with "No new issues found." followed by the PR summary, so the user gets the headline up front and still sees what was reviewed. - docs/pr-reviews.mdx: dropped the "customize the summary style with Summary instructions in the console" sentence (the editor was removed in the firewall commit). Replaced with a note that the snapshot uses Pullfrog's built-in format and is not user-customizable. - wiki/prompt.md, wiki/modes.md: rewrote the snapshot-prompt entries to reflect the firewall — DEFAULT_PR_SUMMARY_INSTRUCTIONS is the entire prompt, prSummaryCommentInstructions is no longer wired in. * drop orphaned prSummaryCommentInstructions column Prod audit (455 repos): 5 non-null rows on a single account, all containing the literal placeholder text from the InstructionsEditor we removed in the firewall commit. No account has an intentional preference set, so silent-ignore (the keep-for-retreat option) costs us nothing meaningful while leaving an orphan column in the schema. Drop it. - prisma/schema.prisma: remove the column - prisma/migrations/20260506000000_drop_pr_summary_comment_instructions: ALTER TABLE ... DROP COLUMN - utils/schemas/triggers.ts: drop the matching zod entry * drop body splicing; snapshot is internal-only User-visible PR summarization continues to ship in Review and IncrementalReview review bodies (which already render PR_SUMMARY_FORMAT and "Reviewed changes" respectively). The snapshot tmpfile is now purely durable cross-run agent context — seed, edit-in-place, save to DB, feed the next run. Massive simplification: the body splice mechanics, the two-toggle gating matrix, the summaryHandlingCovered race tracking, and the synchronize summary-only Task fallback all go away. Code: - prSummary.ts: drop splice/strip/marker code (`splicePrSummary`, `stripExistingSummaryBlock`, `buildSummaryBlock`, `extractPrSummary`, PULLFROG_SUMMARY_START/END). keep scaffold, instructions, seed/read. - main.ts: rename persistAndPostSummary -> persistSummary; collapse to a single DB PATCH. drop pulls.get/pulls.update, drop AbortSignal timeout, drop in-flight promise machinery, drop prSummaryToBody plumbing. - ToolState: add summarySeed (replaces local var in main.ts so persist can compare). drop prSummaryToBody and summaryPersistInFlight. - persistSummary now compares against the seed and skips the DB write with a warning when unchanged — saving the seed verbatim is either a no-op or persists the placeholder scaffold, neither useful. - postRun.ts: when summary-stale is the only failing gate and the resume turn itself fails, restore the pre-resume successful result and break. symmetric with the existing reflection-failure preservation. summary-stale can no longer flip a successful run to failed. Webhook: - pull_request_opened: generateSummary follows prReReview only (the snapshot has no consumer when re-review is off). - pull_request_synchronize: collapses to "if prReReview enabled, dispatch IncrementalReview". the summaryHandlingCovered flag, the same-SHA/in-flight coordination it was protecting, and the summary-only Task fallback all delete cleanly. UI / config: - drop SummarizePRsTrigger (the toggle gated body splice; with that gone it has no behavior). drop sidebar entry, console import, Text icon import. - drop prSummaryComment from triggers zod schema, prisma schema, preview settings script. Migration: squash the two existing migrations into one timestamped 20260507000000_pr_summary_snapshot covering all three column changes (add summarySnapshot on workflow_runs, drop prSummaryCommentInstructions and prSummaryComment on repos). repo convention is one migration per PR. Action: bump 0.0.203 -> 0.0.205 (payload contract changed: prSummaryToBody removed; main is at 0.0.204). Out-of-diff cleanup: - review.ts:190 + review.test.ts:651 — "Reviewed — no issues found." -> "No new issues found." to match the canonical body in modes.ts. Verified: pnpm typecheck clean, pnpm lint clean, postRun + review tests pass, dev DB reset against production and the squashed migration applied cleanly (summarySnapshot present, prSummaryComment / prSummaryCommentInstructions both gone). * re-orient snapshot toward functional summary; drop prior-review-feedback section Empirical audit on preview-568 PR #5 showed the snapshot IS load-bearing for the orchestrator: lens-dispatch prompts on incremental runs carried forward context from the snapshot's risk register (e.g. "the JSDoc explicitly scopes to code points — do not flag grapheme-cluster issues" on the surrogate-pair fix run, "consistency with native padStart" on the padStart-added run). The orchestrator was reading the snapshot, reasoning about it, and using it to anti-prime / focus subagents — exactly the high-leverage path. My earlier "snapshot is write-only" claim was wrong. The shape, however, was steering it toward review-history-log instead of functional summary. This commit re-orients: - prSummary.ts: replace the four-section scaffold (~580 chars of placeholder italics under "What this PR does / Key changes / Risk / Reviewed in prior runs") with a minimal seed (~150 chars: just a header + a one-line comment about what the file is for). different PRs warrant different organization; forcing a refactor and a feature into the same template is procrustean. minimal seed also makes the unchanged-from-seed gate in persistSummary more sensitive. - selectMode.ts addendum: rewrite around three principles. (1) the snapshot is a FUNCTIONAL summary of what the PR does and the risks it carries, not a chronological review log — commit history can already be reconstructed from list_pull_request_reviews. (2) the orchestrator should USE the snapshot during triage and dispatch — concrete example given of carrying snapshot context into subagent lens prompts. (3) structure is the agent's call; stable headings make snapshots range-diff cleanly when they fit, but riff when they don't. - modes.ts IncrementalReview: drop the "Prior review feedback" checklist from the user-facing review body (step 6b gone, step 7 ELSE IFs cleaned up). It duplicated content that's already covered by the Reviewed-changes bullets and tracked durably in the snapshot for the next agent run; in the user-facing body it was noise. step 3 still fetches prior reviews but its role is now just filtering aggregation in step 5, not rendering. - AGENTS.md: codify "no follow-ups" rule. when an issue is identified during code review, fix it in this PR — PR scope does not constrain quality. follow-up TODOs are forbidden as a substitute for doing the work now. Empirical evidence supporting the re-orientation: - Run 25568912293 (PR#5 incr1, surrogate-pair fix): orchestrator's correctness lens dispatch said "Do NOT flag grapheme-cluster issues — the JSDoc scopes to code points." The grapheme-cluster framing was not in the diff; it was downstream of the snapshot's prior risk-section framing of truncate's contract. Snapshot influencing dispatch. - Run 25569054779 (PR#5 incr2, padStart added): orchestrator's correctness lens dispatch enumerated edge cases including "consistency with native String.prototype.padStart contract" and "fill = multi-code-point string (e.g. emoji)". Both threads carried over from the snapshot's prior truncate code-point-vs-code-unit discussion. Snapshot informing the shape of what was looked for. The cost of maintaining the snapshot (~800 tokens, ~$0.005/run) is trivially affordable when it materially improves orchestrator triage on the 1-5 lenses dispatched per review. |
||
|
Colin McDonnell
|
3c8b493aee |
modes: soften "two-out-of-three" rule from veto to look-harder signal
The previous phrasing ("not enough — still degrades the codebase") read as a
categorical claim that elegance vetoes correctness, which inverts the usual
hierarchy and risks giving the agent a clean rationalization for rejecting
genuine correctness fixes. Reframe as a prompt to keep searching for a fix
that gets all three before accepting the trade — preserves the pressure
without the absolute.
|
||
|
Colin McDonnell
|
588badd1b0 | run audit cron every 8h | ||
|
David Blass
|
b835d53d83 |
add /anneal + pullfrog-reviewer named subagent + Build self-review polish (#550)
* cherry-pick updated /anneal command from billing branch + add as Claude Code slash command mirrors origin/billing:.cursor/commands/anneal.md (commit 4f389a8f) into both .cursor/commands/ and .claude/commands/ so the parallel-lens annealing prompt is available in both editors. content is identical between the two files. * anneal: drop REVIEW.md pointer, surface-agnostic dispatch wording, fix modes.ts self-review contradictions Anneal pass over the /anneal slash command and the Build-mode self-review step: - Drop REVIEW.md references in both anneal.md copies. The file does not exist on the Claude Code surface (only .cursor/commands/), and its contents (correctness/security/impact framing) directly contradict the prescribed single-lens, no-pre-shaping discipline. - Replace "Task tool calls" with surface-agnostic "parallel subagent calls" so the meta-prompt does not couple to either CLI's tool naming. - Hedge the "verify via web search" instruction to acknowledge subagents may not have web search available. - modes.ts: drop "and the changed files" — the same step's don't-list forbids handing subagents a curated reading list (in-file contradiction). - modes.ts: restore the "skim only, don't pre-review" warning that the long-form treats as load-bearing. - modes.ts: drop "NO MCP tools" — overbroad; the actual safety property is captured by "no writes, no shell commands, no side effects". Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * anneal: two-round self-anneal of /anneal + modes.ts self-review Expand the multi-lens parallel-review protocol with fixes surfaced by running /anneal on this branch twice. Material additions: /anneal canonical (.claude/commands/anneal.md + .cursor mirror): - promote orientation-vs-defect-hunting distinction to a load-bearing framing in the opening paragraphs - add an empty-target early exit ("nothing to anneal" stop) at §1 - spell out the read-only constraint with the no-op-if-reverted test, and forbid recursive subagent dispatch (incl. agentic MCP tools) - add cleanup-and-debt sub-categories (env vars, feature flags, dangling symbols), supply-chain, test-integrity lenses to the catalog - §1 lens-count rule: explicit trivial/typical/high-risk tiers; "treat as typical" tiebreaker for the unsure case - §2 example uses bare `git diff <primary-branch>` to capture uncommitted edits (three-dot syntax is committed-only) - §5 targeted-follow-up cross-references the fresh-eyes carve-out in Delegation discipline - final-message format spells out coverage shape, findings-table shape, dry-run fix-plan branch, and plan/doc summary branch - stopping criteria distinguish "trivial" from "small / low-risk" action/modes.ts Build mode step 4 (self-review one-pass anneal): - empty-diff early exit; "step 4 mandatory whenever there is a diff" resolves the prior contradiction with the always-runs assertion - lens count by risk (2-3 typical / 4 high-risk single-round-cap / exactly 1 trivial) with separate Tiebreaker - expand swap-in lens menu (research-validated assumptions, security, user-journey, ops, integration, test integrity, supply chain, performance, holistic) so the catalog is a starting menu, not a closed set - rename `cleanup & scope` to `diff hygiene` to avoid colliding with the canonical's broader `cleanup & debt` - delegation discipline bulletized (don't lens-review yourself, don't summarize, don't curate, don't pre-shape, don't mention other lenses); independence rationale stated inline - explicit research-discipline reminder for any lens that touches external contracts (web search, quote URLs) - comment block enumerates deliberate omissions vs the canonical (dry-run, severity categorization, read-only shell) and the deliberate scope decision (sibling diff-producing modes stay solo) action/modes.ts Review + IncrementalReview subagent-dispatch wording: - propagate the no-recursive-dispatch rule (was missing) Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * add set_plan/get_plan + restructure Review/IncrementalReview as parallel-subagent orchestrators Build mode's self-review and Review/IncrementalReview now follow the multi-lens parallel-subagent fan-out pattern from the canonical /anneal protocol. New set_plan/get_plan MCP tools (orchestrator-only) persist the implementation plan in tool state so the self-review's plan-adherence lens can verify the diff against the original intent rather than reconstructing it post-hoc. Subagent "read-only / no further dispatch" is currently enforced via prompt prose only — neither claude-code's --disallowedTools nor opencode's per-agent tools allowlist is configured to scope subagent MCP access. Documented as a deferred ~30-50 LOC follow-up in the modes.ts header comment. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * revert Review/IncrementalReview mode prompts to main; keep Build self-review changes E2e testing on this branch only exercised the trivial-1-lens path for Review (preview repo had only docs PRs). Multi-lens Review fan-out was never directly validated against a real code PR. Splitting the Review/IncrementalReview restructure to its own branch (review-mode-orchestrator, draft PR #555) pending focused validation. Keep on this branch: - set_plan/get_plan MCP tools - Build mode multi-lens self-review (Test 3 directly validated 2-subagent parallel fan-out on a 2-file diff) - /anneal command updates (.claude/ and .cursor/ mirrors) Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * require plan parameter when selecting Build mode Adds an arktype .narrow on SelectModeParams that rejects select_mode({mode:"Build"}) unless a non-empty 'plan' string is also provided. When valid, the plan is stored into ctx.toolState.plan at mode-selection time, so step 4's plan-adherence lens always has a comparison target. This closes the e2e finding that agents never reached for set_plan on their own (5 of 6 runs in production). Build mode prompt updated to reflect that plan is already populated at mode selection; set_plan remains as the mid-task replan tool. Other modes are unaffected. Validation surfaces the error to the agent with a descriptive message including the path ('plan') and recovery instructions, so a failing call is recoverable on the next turn rather than a hard fail. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * move Build-mode plan-required check from arktype .narrow to execute() arktype .narrow predicates aren't JSON-Schema serializable — FastMCP's toJsonSchema() emitted a {code: "predicate", predicate: Function} object instead of a serialized schema. Effect: agents couldn't see select_mode in their tool list (verified by 5 consecutive runs across two models silently bypassing select_mode entirely after the prior commit). Fix: keep the param schema clean (.narrow removed) and check selectedMode.name === "Build" && !params.plan in the execute() body, returning a structured error response. The agent now sees select_mode normally, gets a clear actionable error if it forgets the plan, and can recover on the next turn by retrying with the plan included. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * flip lens architecture: Build = single fresh-eyes subagent, Review/IncrementalReview = multi-lens Build mode self-review previously fanned out 1-4 lenses on the agent's own diff. The bias-mitigation argument for fan-out is weaker for self-review than for reviewing someone else's PR — the orchestrator just wrote the code, so what matters is one fresh-eyes subagent that doesn't share the implementation context, not breadth across parallel angles. Build now dispatches exactly one subagent that gets the original user request and the diff and evaluates whether the diff fulfills the request. Review and IncrementalReview now use the multi-lens orchestrator pattern (triage → parallel read-only fan-out → aggregate → draft comments → submit). For someone else's PR, parallel lenses (correctness, security, research-validated, user-journey, etc.) provide breadth that a single subagent can't carry coherently. Was previously parked on the review-mode-orchestrator branch (PR #555). Removes set_plan/get_plan MCP tools, ToolState.plan field, and the plan parameter on select_mode. Validated end-to-end that those didn't cause agents to actually use plan tracking (5 of 6 e2e runs skipped them); the original user request from the prompt body is the source of truth and the orchestrator already has it. Drops timeout test plan-param workaround that was added for the prior validation. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * split Review/IncrementalReview multi-lens back out to review-mode-orchestrator branch The multi-lens orchestrator restructure for Review/IncrementalReview was bundled into this branch in commit e964ae0c, but it hasn't been validated against a real code-heavy PR (the e2e exercised it only on docs PRs). Splitting it back out keeps this branch focused on the validated half — Build → single fresh-eyes subagent — and lets the Review changes ship in a focused PR (#555 reopened). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * anneal: fix Build prompt contract bugs found by 3-lens review Major fixes: - checkout_pr returns the field as `base`, not `baseRef` (per checkout.ts:611-616). The prompt was telling agents to read `result.baseRef` which would be undefined. - The base-ref fallback "after fetching" is unreachable via the `git` MCP tool (it blocks `fetch` per AUTH_REQUIRED_REDIRECT). Now names `git_fetch` explicitly. - Boundary-tag wrapping for the user request had no escape rule for input that contains the literal close marker, and no fallback for an empty request. Both are now documented with a nonce-suffix mitigation. - PR reference updated #555 → #557 (the active PR for the multi-lens review-mode-orchestrator branch; #555 was closed after the rebase). Minor fixes: - Retry predicate tightened: "errors out (tool error) or returns an empty body", not "returns nothing usable" (which is unfalsifiable and lets an orchestrator declare any output not-usable to skip review). - Subagent read-only constraints rephrased as prescriptive ("MUST NOT call") rather than descriptive ("you have only"), since on inheriting runtimes the subagent does in fact have access to write tools and the constraint is prompt-only. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * anneal round 2: tighten Build prompt edge cases (workflow_dispatch, base-ref, footer-strip, skip marker) Cross-lens findings from holistic + user-journey + research-validated lenses: - workflow_dispatch + empty diff: report_progress silently no-ops when there's no parent issue/PR. Now also call set_output with a "no-op" summary so the user gets surfacable feedback. - base-ref resolution: clarified `base` from checkout_pr is a bare ref name, added explicit `git remote show origin` path for repos whose primary is not `main` (master, trunk, etc.). - bare `git diff` description: tightened from "shows working tree" to "shows unstaged working-tree changes" — bare diff misses staged changes too, not just committed ones. - prompt-body stripping: explicitly call out the leading `> ` blockquote prefix (added by the *YOUR TASK* section formatting) and the entire Pullfrog footer block, not just one example link. - boundary-tag nonce: always-on now, not conditional on detecting a close marker. Cost is one random short string; failure mode (prompt injection if input contains literal close marker) is silent. - subagent-skip marker: structured `Self-review: SKIPPED (subagent error: ...)` on its own commit-message line, so the gap is greppable. Header comment also documents: - AddressReviews/Fix/Task asymmetry (deliberately deferred) - Subagent-runtime-fence deferred fix must explicitly deny Skill / agentic MCP tools, not just destructive tools (claude-code blocks recursive Task spawn but not alternative dispatch paths). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * anneal round 3: targeted re-review of round-2 changes catches real regressions Round 2's "fixes" introduced two real bugs that round 3's targeted correctness re-review caught: CRITICAL (fixed): tier-3 base-ref resolution used `git remote show origin`, which requires network auth — the MCP `git` tool runs commands through plain spawn() without auth, so this hangs on private repos. Replaced with `git symbolic-ref refs/remotes/origin/HEAD` (local symref, no network), which actions/checkout populates. MAJOR (fixed): the eventInstructions fallback was incoherent — the agent has no separately-addressable eventInstructions field; whatever it received in *YOUR TASK* is its only input. Removed the misleading reference. MAJOR (fixed): per-line `> ` strip was ambiguous, could destructively flatten user-pasted markdown blockquotes. Now: "strip exactly one leading `> ` per line". MAJOR (fixed): tier-1 base-ref preferred bare `<base>` over `origin/<base>`, which fails on the rare alreadyOnBranch path in checkout_pr where the local ref isn't re-created. Now prefers `origin/<base>` (always populated post-fetch). MINOR (fixed): footer-strip anchor was `<sup>`/`<picture>`, both of which appear in legitimate user content (footnotes, etc.). Switched to the PULLFROG_DIVIDER sentinel which is purpose-built for this. MAJOR (acknowledged, partial fix): 4-hex nonce is theatrical security; bumped to 8 hex and explicitly noted it's a typo-guard, not a security boundary, and that the structural fix (separate task() argument) is the real solution. REJECTED (verified false positive): subagent claimed `set_output` is not registered for workflow_dispatch. Verified at action/utils/payload.ts:118 — workflow_dispatch from `gh workflow run` resolves to trigger:"unknown", which IS standalone, which IS registered with set_output. E2e logs from prior tests confirm agents successfully call pullfrog_set_output on workflow_dispatch runs. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * anneal round 4: drop broken symbolic-ref tier, simplify base-ref resolution Round 3's tier-2 (`git symbolic-ref refs/remotes/origin/HEAD`) is empirically broken: actions/checkout doesn't populate origin/HEAD on shallow clones (fetch-depth: 1, used by pullfrog.yml), and Git 2.50+ no longer auto-sets it on full clones either (actions/checkout#2219). New scheme: PR context uses checkout_pr's `base`. Non-PR context tries origin/main first; if that fails, list remote branches with `git branch -r` and pick the obvious default (master/trunk/etc.). Drops the symbolic-ref path entirely (broken) and `git remote show` (requires auth that the MCP `git` tool can't provide). Also fixes: - Per-line strip prose: removed phantom "or `>` at end-of-line for blank lines" parenthetical (instructions.ts always emits `"> "`). - Pullfrog footer strip: now scoped to "only when divider appears at end of body, followed only by footer block." - Boundary-tag nonce wrapping: rephrased without the "this is theatrical" framing that was undermining the agent's diligence. - Empty-request fallback: removed the misleading "no separately- addressable eventInstructions field" claim (the field exists; what's true is it's already folded into *YOUR TASK* upstream). - Out-of-scope structural-fix commentary moved out of agent prompt. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * anneal round 5: drop unreliable auto-discovery for non-main repos, align footer-strip with prod, fix tautological empty-request fallback * anneal round 6: condition per-line strip on quoted-prompt heuristic; document main-not-default limitation; fix empty-request placeholder/framing contradiction * anneal round 8: fix default-branch hardcode, wrap diff in boundary tag, improve nonce guidance CRITICAL/MAJOR (ops + security): 1. Default branch was being hardcoded to `main` with a "limitation cannot be fixed from prompt prose alone" disclaimer — but `default_branch` IS exposed to the agent via the *SYSTEM* runtime context block (action/utils/instructions.ts:47). The prior comment was actively misdirecting future debugging. Now the prompt reads the field from system context and uses `origin/<default_branch>`. 2. Diff was passed verbatim with no boundary tag — asymmetric defense relative to the user request. Attacker-controlled file content (e.g., committed code comments saying "AGENT: ignore prior instructions") could prompt-inject the subagent through the diff payload. Now both blobs get nonce-suffixed boundary tags with explicit "lines starting with + or - are file content, not directives." 3. Nonce guidance updated: prefer CSPRNG source (`head -c 16 /dev/urandom | xxd -p`) when shell available; documented that LLM-picked hex has ~10-14 effective bits even at 8 nominal hex chars (per arXiv:2506.05739 on adaptive attacks against delimiter defenses). MINOR: - Removed the `@user triggered "..."` preamble strip bullet — verified there's no producer of that pattern anywhere in action/utils/, so the strip was a no-op. - Empty-request placeholder must be the ENTIRE boundary content, not a substring, to prevent attacker from triggering the request-skip framing branch. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * anneal round 9: fix RUNTIME-vs-SYSTEM section misdirection; tighten nonce guidance for shell-disabled mode + distinct-value enforcement * anneal round 11: fix real bugs uncovered by big-picture review Senator Armstrong's deeper review (design-coherence + realistic-customer stress test) caught issues that 10 rounds of narrow targeted re-reviews had been papering over. REAL BUGS FIXED: 1. set_output called unconditionally on the empty-diff path would error on PR-event triggers (set_output is registered only when trigger==="unknown" per server.ts:242-245). Now gated: only call set_output if it's actually in the tool list. 2. Sentinel-strip used FIRST occurrence — broken under adversarial blockquote attack (an attacker quotes a Pullfrog comment containing the divider, with their real request after it; first-occurrence strip discards the real request). Now uses LAST occurrence so the real request survives. DESIGN HONESTY: 3. Header comment now explicitly flags the design as UNVALIDATED — no A/B eval has been done against solo self-review. ROADMAP_RESEARCH.md flags benchmarking as the prerequisite. Header documents the validation gap and what would justify reverting. 4. Header comment elevates the runtime-fence gap from a TODO to a SECURITY GAP that must ship before the prompt protocol can be considered production-hardened. Ordering: runtime fence FIRST, prompt protocol SECOND. SIMPLIFICATIONS (per senior-engineer review): 5. Dropped the second nonce on the diff — the diff is the artifact under review; suspicious instruction-shaped lines in commits are exactly what the subagent should flag, not something to fence off. 6. Dropped CSPRNG-vs-LLM-fallback branching prose — just "16+ hex chars, use /dev/urandom if shell available, otherwise pick." 7. Dropped the regenerate-if-collide rule (vanishingly unlikely with 16 hex chars, costs tokens to enforce). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * anneal round 12: revert round-11 regressions (sentinel-strip, set_output gate, diff nonce) Round 12's sharper review caught three regressions round 11 introduced: 1. Sentinel-strip last-occurrence was strictly worse than first-occurrence for the common "user references a prior Pullfrog comment" case. The adversarial-quote scenario it was defending against is contrived (an attacker can put hostile payload anywhere; strip discipline doesn't change attack surface). Reverted to first-occurrence to align with canonical stripExistingFooter() and avoid silently swallowing user reference context. 2. set_output "gate" via "if it's in your tool list" relied on tool introspection that LLMs cannot reliably perform. Replaced with: just call report_progress; document the workflow_dispatch limitation as acceptable (job log is feedback-of-last-resort) rather than asking the agent to conditional-call a tool that may not exist. 3. Diff was de-nonced in round 11 on the assumption runtime fence ships first, but until that runtime fence lands the plain label is forgeable (committed file content can include "--- END DIFF ---" + injection). Restored nonce wrapping. The cost is one extra hex string; the benefit is real until runtime fence ships. Also added explicit caveat on the self-attested skip marker: the proper fix is MCP-layer dispatch-counting, not commit-message annotation. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * ruthless cut: revert Build self-review elaboration to compact form main already had subagent dispatch (4 compact lines). This branch added 70+ lines of elaboration — header warnings, base-ref dance, footer-strip rules, nonce- suffixed boundary tags, retry-once skip markers, delegation-discipline list — all predicated on a runtime fence that doesn't exist and validation that never ran. Senior-engineer review (round 11) explicitly recommended cutting; ROADMAP_RESEARCH flags A/B benchmarking as the prerequisite for this design. Net change vs main now matches what the user actually asked for: - drop the optional plan step (and its "follow the plan" / Notes references) - subagent receives the original user request alongside the diff, evaluated against base ref, with explicit no-further-dispatch constraint Everything else reverts to main's prose. ~10 lines net change instead of 70+. * anneal round 13: tighten self-review prompt inputs to runtime-resolvable values Two underspecified inputs flagged by parallel holistic + mechanics review: 1. "the original user request" is empty for non-@pullfrog-tagged auto-triggers (sync, check_suite, opened, etc.); only YOUR TASK is reliably present in the assembled prompt across all event types. Replace. 2. "base ref (PR base or repo default branch)" requires the agent to resolve and fetch the default branch on non-PR runs (origin/<default> typically not fetched). Drop the elaboration — bare git diff captures all changes at step-3 time since step 2 doesn't commit. Aligns with 3ed2c55a's ruthless-cut philosophy: less elaboration, not more. Verified in round 14: YOUR TASK is the literal section header in instructions.ts (buildTaskSection); bare git diff scope is correct. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * restore plan step to Build mode prompt The plan step was removed alongside the MCP-contract plan-required work, but the user only wanted it gone from the MCP contract, not from the prompt itself. Restores step 1 (plan), the "follow the plan" build sub-bullet, the trailing Notes section, and renumbers learningsStep back to 6. Made-with: Cursor * add pullfrog-reviewer named subagent; standardize review fence to non-mutative+non-recursive Defines a constrained `pullfrog-reviewer` named subagent for the Build mode self-review and /anneal lens dispatch, with a single source of truth in action/agents/reviewer.ts (allowed tools, denied mutating MCP tools, system prompt). Enforcement: - opencode: real fence via agent.pullfrog-reviewer block in buildSecurityConfig — denies edit/bash/task and globs each mutating pullfrog_* MCP tool to false. - claude-code: forward-looking only. Per-agent disallowedTools is upstream-broken (anthropics/claude-agent-sdk-typescript#172, open as of latest update Mar 2026 — subagent child processes still see and can call disallowed tools, including Task). The --agents JSON is defined anyway so the fence becomes real when upstream fixes #172; until then the prompt prose constraint is the actual fence. The PreToolUse hook workaround that does enforce is out of scope. Read-only MCP tools (get_*, list_*) intentionally remain enabled so the reviewer can pull PR/issue/check context without dispatching state changes. Both modes.ts Build self-review and the two anneal.md files now share the same "non-mutative + non-recursive" framing — file reads, grep, search, web search/fetch, read-only shell, and read-only MCP queries allowed; writes, state-changing MCP, and nested subagent dispatch denied. Resolves the previous inconsistency where /anneal allowed read-only shell and Build self-review banned all shell. Made-with: Cursor * Build self-review: pass build-phase failure summary to reviewer subagent Adds an instruction in step 4's dispatch: along with YOUR TASK and git diff, pass a tight plain-text summary of any lint/typecheck/test failures fixed during build (what broke, root cause, the fix) — or "no build-phase failures" if clean. Goal: let the reviewer check that fixes addressed root causes rather than suppressed symptoms (e.g., editing a test to make it pass instead of fixing the bug). Implemented as agent self-summarization rather than piping raw build output to avoid context flooding — typecheck/test output can be hundreds to thousands of lines per failure. The agent has the failure trail in its own conversation history and summarizes from memory; the reviewer sees a few lines per failure, not raw stderr. Caveat: this is a plausible-but-unvalidated quality improvement. The mechanical justification (signal already produced, currently not passed on) is real; "this catches more bugs" is a hypothesis that will need actual run data to confirm. Downside is bounded (reviewer gets slightly more context, no behavior change if the summary is empty or ignored). Made-with: Cursor * Build self-review: distill /anneal delegation + research discipline into dispatch instructions Lifts the codified learnings from /anneal's "Delegation discipline" and "Research discipline" sections into Build mode step 4. These rules are about how-to-prompt the reviewer (not about parallelism), so they transfer losslessly to single-agent dispatch and address bias modes the prior prompt was silent on: - Don't summarize what you implemented (biases toward shape-validation) - Don't curate a reading list (your curation is itself a lens) - Don't pre-shape output with severity/category (leaks hypotheses) - Don't defect-hunt in parallel (reintroduces the implementation bias the subagent is meant to mitigate) - For diffs touching third-party API contracts / SDK semantics / framework directives / DB engine specifics, instruct the reviewer to verify load-bearing claims via web search and quote URLs rather than trust training data Restructures step 4 from one paragraph into three (constraints, inputs, discipline) plus a final review-and-commit paragraph for readability. These are validated learnings from many anneal rounds, not theoretical best practices — they're the single substantive piece this branch was missing. Made-with: Cursor * pullfrog-reviewer: drop MCP deny-list, rely on prose constraint Per-PR-review feedback: hand-maintaining MUTATING_MCP_TOOLS against action/mcp/server.ts was fragile — a future mutating tool added to the MCP server without updating this list would silently grant write access to the reviewer. Inverting to an allowlist or adding a structural test both keep the drift problem. Drop the list and all per-agent runtime denies (claude disallowedTools, opencode tools/permission map). Strengthen REVIEWER_SYSTEM_PROMPT to spell out the categories of state-changing MCP tools by example and explicitly tell the model to apply the no-op-if-reverted invariant to tools added after the prompt was written — the rule is the invariant, not the enumeration. Keep the named subagent so the prompt is reliably injected. Update modes.ts and both anneal.md copies to drop the runtime-enforces-where-supported claim. Co-authored-by: Cursor <cursoragent@cursor.com> * pullfrog-reviewer: fix description to allow read-only shell The description field was overstating the constraint as 'must not shell', but the system prompt explicitly allows read-only commands like git diff, git log, cat, ls. Align description with the actual contract. Co-authored-by: Cursor <cursoragent@cursor.com> * restructure Review/IncrementalReview as multi-lens parallel-subagent orchestrators For someone else's PR, parallel lenses (correctness, security, research-validated claims, user-journey, etc.) provide breadth across angles that a single subagent can't carry coherently. The orchestrator does triage → parallel read-only subagent fan-out → aggregate → draft comments → submit. Lens count by risk: 1 lens for trivial PRs, 2-3 for typical, 4 for high-risk surfaces (billing, auth, migrations). This branch contains ONLY the Review/IncrementalReview multi-lens prompts. Build mode keeps its single-fresh-eyes-subagent shape (different problem — orchestrator just wrote the code; bias-mitigation comes from one subagent that doesn't share the implementation context). The Build changes ship in a separate PR (self-review-subagents → main). Pending validation against a real code-heavy PR before merge — e2e on a docs-only preview repo only exercised the trivial-1-lens path. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Review/IncrementalReview: dispatch fan-out via reviewfrog named subagent The fan-out steps previously said "launch one read-only subagent per lens" without naming the subagent. That bypassed the only enforcement layer the named subagent provides: a baked-in system prompt that restates the non-mutative + non-recursive contract regardless of what the orchestrator sends. Both modes now dispatch via REVIEWER_AGENT_NAME (matching Build mode's self-review wiring) and restate the constraint inline so the rule is present twice. * rename pullfrog-reviewer → reviewfrog Mechanical rename of the named subagent. Constant names (REVIEWER_AGENT_NAME, REVIEWER_SYSTEM_PROMPT) and file paths (action/agents/reviewer.ts) stay as-is — only the agent identifier string and prose references in anneal.md and code comments change. * modes/anneal: trivial PRs skip review entirely; lens count is judgment, not table; allow subsystem lenses Three coupled changes to Review/IncrementalReview/Build self-review and the canonical /anneal command: 1. Trivial-skip: trivial diffs (single-line, formatting/comment-only, doc typo, low-risk dep bump, no behavior change) skip the fan-out / self-review entirely. Build mode skips its self-review subagent; Review submits a bare "Reviewed — no issues found." without dispatching lenses; IncrementalReview takes the existing non-substantive submit path. Tiebreaker on uncertainty: treat as non-trivial. 2. Drop prescriptive lens counts. Replaces "2-3 typical / 4 high-risk cap / 1 trivial" with judgment-based guidance: pick as many lenses as the target has distinct surfaces of risk worth investigating independently; one is sometimes enough; bias toward more (and toward follow-up rounds in /anneal) for high-stakes subsystems; 5+ is a smell that lenses are overlapping rather than covering distinct ground. 3. Subsystem lenses. Adds an explicit second flavor of lens — domain-scoped frames like "the auth lens", "the billing lens", "the schema-migration lens" — alongside the existing themed lenses (correctness, security, user-journey, etc.). Stack themed + subsystem freely. modes.ts and anneal.md (.cursor/ + .claude/, kept byte-identical) move together so the canonical pattern doc and the orchestrator prompt agree on the protocol. * add SessionLabeler so parallel subagent log lines are differentiable When the orchestrator dispatches multiple `reviewfrog` subagents in a single assistant turn (the parallel fan-out the multi-lens prompt now requires), their tool_use / tool_result / text events arrive on opencode's NDJSON stream tagged with distinct `sessionID`s but go through a single `[Pullfrog]` log prefix. Result: log readers can't attribute which lens issued which tool call, making CI logs unreadable for any review with 2+ lenses. SessionLabeler: - Binds the first-seen sessionID to "orchestrator" and subsequent new sessionIDs to FIFO-popped lens labels seeded from task tool_use inputs. - Derives labels from `lens: <name>` markers in the dispatch prompt, the Task `description` field, the `subagent_type`, or `subagent#N` fallback. - Keeps state local to a single runOpenCode invocation. Wiring: - opencode.ts: every event handler (init, message, text, tool_use, tool_result) now looks up the per-event label and prefixes log output via formatWithLabel(). Subagent finalOutput/token-reset paths gated on ORCHESTRATOR_LABEL so child sessions can't clobber parent state. - claude.ts: claude rolls subagent activity into a single tool_result block (no per-event session_id), so it gets a minimal "» dispatching subagent: <label>" log line on Task tool_use as the only attribution. - modes.ts (Review + IncrementalReview): orchestrator instructed to set the Task `description` to the lens name, since that's what the labeler reads when no explicit `lens:` marker is in the prompt. Tests: 18 unit tests covering label derivation, FIFO binding, interleaved sessions, fallback paths, and a realistic four-lens parallel fan-out simulation. Full action test suite stays green (400 passing). This is the pre-flight instrumentation that the multi-lens validation runs depend on — without it, post-hoc log analysis can't tell two subagents apart. * log subagent dispatch + finish at info level for per-lens visibility OpenCode's runtime currently encapsulates subagent execution inside the `task` tool — subagent-internal tool_use/tool_result events do not surface on the parent's NDJSON stream. The SessionLabeler I added in 0c4647f4 therefore can't actually differentiate concurrent subagent log lines (there are no concurrent log lines on the parent stream to differentiate). What CAN be observed on the parent stream is the dispatch and the result of each `task` tool call. This patch surfaces both at info level: » dispatching subagent: lens:security (subagent_type=reviewfrog) ... » subagent finished: lens:security (15.3s, status=completed) — ... Without this, a 4-lens parallel fan-out looks like 4 dispatches in close succession followed by a long quiet gap and then an aggregation turn — you can't see when each lens finished or how the durations overlapped. With it, parallel execution is visible from the timestamps on the "finished" lines. The dispatched label comes from SessionLabeler.recordTaskDispatch (so both lines share the same lens identity). taskDispatchInfo maps callID to {label, startedAt} so the matching tool_result can compute duration and emit the finished line. Also added a defensive comment on the SessionLabeler instantiation documenting that the per-event session-prefix path is currently dormant in the opencode runtime, but kept in place so attribution flips on automatically if/when opencode begins streaming subagent sessions. * fix subagent-finished log: hybrid exact+FIFO callID matching opencode does not consistently surface a tool_result callID matching the originating tool_use callID for the `task` tool, so the previous exact-match-only finish line never fired. Now we: - Dual-index task dispatches by callID AND in a FIFO queue. - Track non-task callIDs so we can identify "unrecognised callID" results as likely-task-with-mismatched-id. - On tool_result, exact-match first; fall back to FIFO when the output looks like a subagent reply (>300 chars) and the callID is unknown. - Flush leftover dispatches at run end with an "(inferred at run-end)" suffix so the gap is visible if subagent results arrive entirely off the tool_result event path (e.g. inlined into the next assistant message). * fix subagent-finished log: move run-end flush to post-subprocess block Investigation on T3 + finish-log-validation runs revealed two real issues with my prior attempt: 1. The `result` event handler is dead — opencode never emits a `result`-typed event over its NDJSON stream, so the inferred-at-run-end flush I had placed there never fired. Move the flush to right after `runSubprocess` returns where it actually executes. 2. The FIFO heuristic was too strict — the >300-char output check excluded short or empty outputs that opencode's `task` tool_result appears to carry (the subagent's full reply seems to arrive via a separate channel, not the result event itself). Drop the size check; rely solely on `knownNonTaskCallIDs` to keep genuinely-non-task tool_results from popping a pending task. Net effect: every `task` tool dispatch gets a matching `» subagent finished` line in the logs, either from the FIFO fallback during the run or from the run-end flush as a backstop. * modes/anneal: anchor lens calibration in worked examples The prior trivial-skip definition ("single-line fix, formatting-only, …") was anchored on diff size, but real-world risk is anchored on diff *shape*: a 5000-line lockfile regen IS trivial, and a 1-line SQL operator flip in a billing path is NOT. The prior lens-count guidance ("there's no fixed count, bias toward more for high-stakes subsystems") gave the agent no concrete shapes to anchor against, so runs varied between under-pick (4 generic lenses on a billing PR) and over-pick (5 overlapping themed lenses on a refactor). This commit hardens both: - Trivial definition gets explicit "looks trivial but isn't" anti-patterns: SQL operator flips, money/tax/timeout constants, feature-flag defaults, comparison operator changes, semantic 1-liners buried in whitespace, public-API renames, new direct deps. Skip lists get explicit "size doesn't matter" calibration for lockfile regens and mechanical renames. - Lens count gets a worked-example ladder: 1 lens (refactor / new test file / isolated fix), 2-3 lenses (typical features), 4-5 lenses (high-stakes subsystem touches), 6+ is a smell. - Subsystem lenses get an explicit recommendation to lead over generic themed equivalents for high-stakes domains, with the reasoning: domain framing primes the subagent for domain-specific failure modes (double-charges, refund races, dispute flows) the generic lens misses. Mirrored byte-identical into both anneal.md copies; modes.ts updates all three review surfaces (Build self-review, Review triage, IncrementalReview triage). * fix harness false-failure when Review submits without todowrite Review and IncrementalReview prompts explicitly forbid calling report_progress (the review IS the durable record). The post-run harness in action/utils/run.ts errors with "agent completed without reporting progress" when toolState.wasUpdated is false at exit. Until now, the only path that set wasUpdated for these modes was the todoTracker's debounced publish — which only fires if the agent happens to call todowrite during the run. Adversarial run on PR #16 (misleading-trivial billing tweak) hit exactly this case: agent went straight from triage → fan-out → review submission with no todowrite calls, and the harness reported failure even though the substantive review was successfully submitted with two inline comments. Fix: create_pull_request_review now marks wasUpdated=true (and finalSummaryWritten=true) on every terminal path — successful submit, empty-content skip, and all-comments-dropped skip. Submitting a review is unambiguously a "done" signal in these modes. Found via adversarial testing of the multi-lens orchestrator on a 1-line tax constant change. Logged in /tmp/pullfrog-validation/v3/. * fix harness false-failure when Review submits without todowrite (correctly) Replaces the prior fix (acc2bd65) which set wasUpdated=true inside create_pull_request_review. That approach worked for the harness check but broke the orphan-comment cleanup: with wasUpdated=true and finalSummaryWritten=true, the (!wasUpdated || trackerWasLastWriter) condition in main.ts evaluated false and the "Leaping into action" progress comment was left behind on every Review run — the exact behavior the cleanup logic was designed to prevent (see plans/review_progress_comment_cleanup_b0120f6c.plan.md). Correct fix: change the harness check in action/utils/run.ts to recognize a submitted PR review as an alternate completion signal alongside wasUpdated. wasUpdated stays false on purpose so cleanup deletes the orphan, but the run no longer false-fails when the agent followed the Review-mode contract (submit a review, never call report_progress). The bug was discovered during adversarial testing of PR #16 (misleading-trivial billing tweak) where the agent went straight from triage → fan-out → review submission without using todowrite, causing the harness to error even though the substantive review (a CAUTION blocking review with two inline comments catching a 10x tax cut) was successfully posted. * fix harness false-failure for Review modes (mode-based carve-out) Replaces the prior carve-out (4c0f69aa) which gated on toolState.review.id. That worked for runs where the review tool actually populated the toolState (validation-2 succeeded), but failed for runs that took a slightly different path where the assignment didn't propagate visibly to handleAgentResult — even when the review verifiably posted to GitHub. Found this empirically: PR #19 (pure mechanical rename across 20 files) opened with the prior fix in place, the agent picked exactly one impact lens (correct calibration!), confirmed no stale references, submitted "Reviewed — no issues found." successfully (visible in GitHub API), and the harness STILL errored with "agent completed without reporting progress." Same SHA, same branch, same code as validation-2 which passed. The toolState.review.id check turns out not to be reliably visible from the run.ts handler in all paths. Better fix: gate on toolState.selectedMode. Review and IncrementalReview modes are designed to never call report_progress (the review is the durable record, and IncrementalReview's non-substantive path produces no artifact at all by design). The harness completion check makes no sense for these modes — skip it entirely. The agent's clean subprocess exit is the completion signal. This also handles edge cases the previous fix missed: IncrementalReview's non-substantive path (no review submitted by design) and any future Review-flow shape that doesn't end at create_pull_request_review. * ci: trigger Test run to validate models-live timeout/concurrency changes * ci: prune passthrough models from live smoke matrix openrouter/* aliases and keyed opencode/* aliases are routing-layer wrappers around models we already smoke-test directly. running every passthrough burns CI minutes (~30 min/run) without catching anything the direct smoke doesn't — slug drift is already covered by the models-catalog job. keep one canary per routing layer (openrouter/claude-sonnet, opencode/claude-sonnet) to validate auth + tool-call translation. free opencode models stay in the matrix since they're unique to the provider. INCLUDE_ALL_PASSTHROUGHS=1 bypasses the prune for full validation. matrix size: 37 → 20 jobs. * fix isRateLimited false-positive on UUIDs/timestamps containing 429 The bare "429" substring pattern was matching MCP session IDs (e.g. `...-4429-...`) and microsecond timestamps in agent stdout, sending transient failures down the 60s rate-limit retry path. With the new 4-minute per-step CI timeout, that backoff plus a slow retry pushed the step past its budget and timed out. Switch to regex patterns and gate the numeric code on `\b429\b` so word boundaries prevent the substring false-match. Verified locally that the UUID `97287d2f-ae1d-4429-8627-73e2454e80ca` and timestamp `02:04:50.9429654` no longer match while real `HTTP 429` / `"status":429` strings still do. --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Colin McDonnell <colinmcd94@gmail.com> Co-authored-by: pullfrog[bot] <226033991+pullfrog[bot]@users.noreply.github.com> |
||
|
David Blass
|
57bd10d6dd |
run-issues fixes: #5, #11, #12, #15, #16/#25, #20, #21, #22, #31 (#546)
* fix(#15): precompute diff anchors in checkout_pr TOC * test(#15): update TOC snapshot for precomputed diff anchors * chore(tests): skip codex-mini-latest models.dev check + refresh latest-by-provider snapshot * fix(#22): add commitCount and commitLog to checkout_pr return * fix(#21): include PR body in checkout_pr return * fix(#5): force-fetch PR refspec to overwrite stale local branch * fix(#31): rename git tool parameter from subcommand to command * fix(#11): soft-fail post-checkout hook, bump timeout to 10min * fix(#16): strengthen diff file usage guidance Agent was bypassing diffPath and running `git diff` instead. Tighten instructions in `checkout_pr` result and remove the mixed-signal "log, diff" listing in the global Git guidance. `git log` and `git diff --stat` remain allowed for commit-range overview. * fix(#20): drop invalid inline review comments instead of failing review Previously, a single inline comment anchored outside a diff hunk would 422 the entire review submission. Pre-validate comments against the PR file patches via listFiles, drop the invalid ones, and append a note to the review body listing what was skipped. Include the dropped list in the tool response so the agent can retry targeted fixes. * fix(#12): stop MCP server on inner activity kill + filter reconnect noise Inner-activity-kill zombies were burning multi-hour runner time because mcp-proxy's SSE reconnect and provider-error retry lines kept the outer activity timer alive long after the agent subprocess was killed. - Filter [mcp-proxy] / "provider error detected" chunks so they don't count as outer-timer activity. - Add onActivityTimeout callback to spawn + thread through agent runs. - main.ts wires that callback to stop the MCP HTTP server (so reconnects finally fail instead of looping) and arms a 5min safety-net timer that force-rejects the outer timer if the agent promise is still pending. * audit: harden #12 lifecycle + cover #20/#12 with unit tests Bugs found during Ralph audit of the prior run-issues fixes: - main.ts's 5min safety-net setTimeout was never cleared on the happy path; also activityTimeout.stop() didn't null the internal rejectFn, so a late forceReject from the safety-net could still reject a long-resolved promise. Timer now cleared in finally; stop() now disarms forceReject. - mcp server disposal was non-idempotent, so the inner-kill path ran server.stop() twice once the outer `await using` block exited. Made the returned disposer idempotent. Tests: - action/mcp/review.test.ts: 14 tests for commentableLinesForFile (multi-hunk, no-count hunks, no-newline marker, empty) and validateInlineComments (file not in diff, wrong side, out-of-range line and start_line, partitioning batches, default side). - action/utils/activity.test.ts: 6 tests for isActivityNoise covering mcp-proxy lines, provider-error lines, mixed chunks, Buffer input. * audit(#22): cap commitLog at 200 + scope git-diff restriction to PR review - cap git log --oneline at 200 entries so a PR with thousands of commits cannot blow up the MCP tool response; expose commitLogTruncated so callers can warn the agent when the log was clipped - tighten instruction wording so `git diff` / `git diff --cached` remain available for inspecting an agent's own uncommitted changes, while PR review content must still come from diffPath Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#11,#22,#31): surface hook/commit warnings in instructions + polish git tool - append hookWarning + commitLogTruncated advisories to checkout_pr instructions so the agent actually sees the warning inline, not just as a field it may skip - fix stale 'subcommand' wording in git tool redirect for `pull` and in the `command` parameter description; the MCP parameter is named `command` now, and that's what the agent binds to Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * fix(#20): reassign params.comments even when all inline comments dropped if every inline comment fails pre-validation, the earlier guard skipped reassigning params.comments, so the submission still carried the bad comments and GitHub 422'd on the whole review. always reassign to validation.valid so the downstream 'nothing left to post' skip fires and an otherwise-empty review is no-oped cleanly. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#22): degrade gracefully when base ref isn't resolvable checkout_pr used to assume \`origin/<base>\` is always reachable, but it isn't guaranteed after a shallow fetch that only pulled down the PR head. Failing the whole checkout over metadata we added for ergonomics would be a regression, so wrap the rev-list / log in a try/catch and return empty commit metadata instead. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#12): anchor noise patterns to line start to avoid false positives before this, a line like "agent said: [mcp-proxy] was there" or "context: provider error detected in log" in real agent output would have been treated as noise and failed to reset the outer activity timer. both patterns now anchor at the start of the (optionally debug-timestamped) line, matching only lines mcp-proxy or our own log.info actually emit. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#20): export and unit-test formatDroppedCommentsNote covers single-line `path:N`, multi-line `path:start-end`, and startLine==line fallback so changes to the dropped-comments note format surface in test diffs instead of only in GitHub UI. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#20): cap dropped-comment note to stay under GitHub body limit a pathological run (agent emits hundreds of invalid inline comments on a huge PR and they all get dropped) would push the review body past GitHub's ~65KB limit and fail the whole submission with a body-too-long 422 — the exact all-or-nothing failure #20 was meant to prevent. cap the detail list at 50 entries with a "…and N more" line so the note stays bounded. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#20): distinguish binary/no-patch files in dropped-comment reason previously a comment on a binary file (or pure rename / mode-only change) was dropped with "line X is not inside a diff hunk", which misleads the agent into retrying with different line numbers. call out the no-textual-diff case explicitly so the agent knows to move that feedback to the review body instead. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#11): replace lifecycle timeout string-match with typed sentinel spawn() now rejects with SpawnTimeoutError (code === SPAWN_TIMEOUT_CODE or SPAWN_ACTIVITY_TIMEOUT_CODE) instead of a plain Error. executeLifecycleHook now branches on that code so rewording the error message in subprocess.ts can no longer silently misroute timeouts into the "transient — retry" warning. * audit(#12): route agent hung-vs-failed via typed SpawnTimeoutError claude.ts and opentoad.ts decide between "hung" and "failed" log wording based on the subprocess error. move them off the literal "activity timeout" substring match onto the same SPAWN_ACTIVITY_TIMEOUT_CODE sentinel used by lifecycle.ts so all three call sites agree on the source of truth. * audit(#20): delete leftover pending review when submit fails Why: `createAndSubmitWithFooter` creates a PENDING review first so we can mint Fix-links with the review ID, then submits. If submitReview fails (e.g. 422 from a race where the diff moved between pre-validation and submission), the draft was left on the PR. GitHub only allows one pending review per user, so the agent's retry would then fail with "already has a pending review" — an error the agent has no tools to clean up from. Best-effort cleanup: delete the pending draft on submit failure before re-throwing the original error, so retries start from a clean slate. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#31): point agent to concrete alternative when rebase/bisect blocked Why: in disabled-shell mode, `git rebase` and `git bisect` are blocked as arbitrary-code-execution escape hatches. Previous error messages explained *why* but left the agent without a next step — especially painful right after the `pull` redirect, which suggested "merge or rebase locally." The agent would follow that advice, hit the rebase block, and loop without knowing what to try next. Now: rebase block explicitly says "use 'merge' instead"; bisect block notes that manual bisect is also unavailable through this tool; pull redirect no longer recommends rebase in shell-disabled contexts. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: import security tables into security.test to prevent drift Why: the security tests re-declared AUTH_REQUIRED_REDIRECT, NOSHELL_BLOCKED_SUBCOMMANDS, and NOSHELL_BLOCKED_ARGS inline with hand-copied message strings. When the runtime messages in git.ts were tightened (recent rebase/bisect guidance updates), the test copies drifted and tests validated a stale version of the logic while passing clean. A missing or mistyped entry in git.ts could therefore slip through. Now: export the tables from git.ts and import them into the test file. If a runtime message changes, the tests exercise the new string automatically; if an entry is added or removed, tests covering that command see the change without manual sync. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: widen pending-review cleanup to cover pre-submit throws getApiUrl() (invoked in footer build) can throw if API_URL is misconfigured, which would leak a pending draft between createReview and the previous submitReview try/catch. Move the try/catch to wrap the entire post-create body so any throw routes through deletePendingReview cleanup. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: reject leading-dash refs/branch names to block flag injection git's parseopt accepts options intermixed with positional args, so a ref like "--upload-pack=evil" passed to git_fetch could be parsed as a flag rather than a refspec. Add a narrow rejectIfLeadingDash helper to git_fetch (ref), delete_branch (branchName), and push_branch (branchName). HTTPS remotes ignore --upload-pack server-side, but the hygiene matters for defense in depth (ssh remotes, future code paths). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: validate the resolved branch in push_branch too When branchName is omitted, rev-parse surfaces the current branch name, which could start with '-' if git state was tampered with. Move the leading-dash check to after the branch is resolved so both the explicit and derived paths go through validation. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: cache commentable-lines snapshot at checkout to match review anchor Review comments are anchored to checkoutSha (commit_id), but validation was hitting pulls.listFiles at review time — latest HEAD, not the SHA the agent actually reviewed. If the PR was updated mid-run, valid comments could be silently dropped (or invalid ones admitted). Snapshot the commentable lines during checkout_pr so review-time validation matches the anchor exactly. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#12): route activity monitor's own debug output around the write wrap startProcessOutputMonitor monkey-patches process.stdout.write to mark activity, then called log.debug(...) every 5s to report idle time — which landed right back in its own wrapper, failed isActivityNoise, and called markActivity. with ACTIONS_STEP_DEBUG=true (common on reruns) the idle counter reset every interval and the timeout could never fire, re-creating the #12 zombie-run bug for any debug-enabled run. Fix: capture the original stdout.write and use it directly for the monitor's own diagnostics so they bypass the feedback loop. Added a tight-timeout regression test that asserts the timeout still rejects in debug mode. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#12): noise-filter subprocess.ts monitor logs so outer timer survives debug activity.ts's own monitor output already bypasses the wrap (c35cd3fb), but subprocess.ts's spawn activity timer uses log.debug — which goes straight through process.stdout.write and would still mark activity on every interval when debug logging is enabled. Pattern-filter those '(spawn|process) activity (check|timer|monitor)' lines in both local ([DEBUG] ...) and GH-runner (::debug::...) formats so they don't reset the outer agent-hang timer. Kept scoped to those specific monitor messages — a blanket [DEBUG] filter would silently classify any coincidentally-debug-prefixed agent output as idle, which is a worse failure mode than the one we're fixing. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#11): surface spawn ENOENT-style errors in stderr buffer spawn() resolved with exitCode=1 and an empty stderr when the command itself couldn't start (missing binary, bad permissions). lifecycle.ts then reported 'output: (empty)' to the user, who was explicitly told 'retry if the failure looks flaky' — so every run hit the same wall with no diagnostic trail. Append the '[spawn] <cmd>: <node error>' line to stderrBuffer before resolving so the real cause (ENOENT, EACCES, …) flows through to the hook-warning message. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit(#11,#12): cover executeLifecycleHook typed-timeout routing the typed SpawnTimeoutError + sentinel-code branching introduced in d7ee7fd2 / ea8dd2c4 classifies hung vs failed lifecycle hooks — critical for whether agents retry — but had no unit coverage. add tests for all four branches (no script, exit 0, non-zero exit with retry-if-flaky guidance, timeout with do-NOT-retry guidance, transient spawn failure). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: re-verify clean tree after prepush hook the pre-prepush check guarantees we enter the hook with a clean tree, but if the hook writes tracked files (formatter, type generator, build artifacts), the push still only sends the pre-hook commit — the hook's edits silently disappear from the upstream branch while the tool reports "successfully pushed". add a post-hook status check so the agent sees the dropped mutations and can commit or discard them before retrying. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: reject push_tags refspec injection via ':' in tag name without tag validation, a tag like "foo:refs/heads/main" concatenated into "refs/tags/${tag}" becomes a valid <src>:<dst> refspec — git pushes the local refs/tags/foo's commit to remote main, bypassing push_branch's default-branch guard. same shape blocks leading '-' (flag injection) and other refspec metacharacters (~ ^ ? * [ \) via an allow-list regex. only reachable in push=enabled today, so this is defense-in-depth, but hardens the tool in case push_tags is ever exposed in restricted mode. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: stop pointing agents at an internal constant they can't change the lifecycle-hook timeout warning told agents to "bump LIFECYCLE_HOOK_TIMEOUT_MS" — but that's a hard-coded constant in the action, not something the agent or repo owner can tune. the agent would plausibly loop hunting for where to change it. redirect to the actual lever they control: ask the repo owner to simplify the hook. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: drop inverted inline-comment ranges locally with precise reason validateInlineComments only checked that both line and start_line anchor inside a hunk, not that start_line <= line. an inverted range (e.g. start=44, line=42) would pass local validation and GitHub would 422 with "invalid line numbers" — opaque to the agent and unfixable without reading docs. reject locally with a reason that names the constraint. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: don't let usage-summary write error mask main's outcome writeGitHubUsageSummaryToFile is called in main's finally block. it can throw on ENOSPC / EACCES / missing parent dir. a throw here propagates past the try's successful return or the catch's error return, hiding the actual run outcome behind an I/O failure on a purely informational file. swallow the write error (debug-logged) — the summary is nice-to-have, not load-bearing. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: don't mislabel agent handler errors as JSON parse failures the onStdout event loop wrapped both JSON.parse and the handler call in one try/catch that logged every caught error as 'non-JSON stdout line'. if a handler threw (e.g. todowrite state shape drift), the error was silently classified as a parse error, making diagnosis impossible. split the try blocks so JSON errors and handler errors get distinct, identifying log lines. * audit: reject leading-dash PR refs before they reach git commands PR head/base refs come from GitHub and are attacker-controlled on fork PRs (the PR author picks headRef freely). they flow straight into `git fetch origin <ref>`, `git checkout -B <ref>`, and config writes. without a leading-dash check, a ref named like '-upload-pack=evil' could be parsed as a flag instead of a refspec. validate both refs at the top of checkoutPrBranch (before any async work) and cover the two attack shapes with unit tests. * audit: cover ActivityTimeout.stop()'s forceReject disarming main.ts's safety-net-timer path depends on ActivityTimeout.stop() nulling out rejectFn so a late safety-net fire after a successful agent run is a no-op. that behavior had no direct coverage — removing the \`rejectFn = null\` in stop() would silently break the happy path (unhandled rejection / spurious failure) without failing any test. add three tests covering: forceReject rejects with the reason, stop() disarms forceReject, and forceReject after timer rejection is an idempotent no-op. * audit: stabilize activity-timeout idleSec against late stdout race * audit: reject 0ms timeout parses to avoid insta-fail from '0m' * audit: surface raw GitHub error on review 422 instead of assuming anchor cause * audit: key commentable-lines cache by PR number to prevent cross-PR drift * audit: enumerate concrete 422 causes and name checkout_pr in review error * audit: stop shipping ralph-loop runtime state in PR history .claude/ralph-loop.local.md and .claude/ralph-loop-prompt.md were accidentally staged in an earlier audit commit. the .local.md suffix is conventional for gitignored runtime state, and the prompt file is per-run harness config — neither should merge to main. ignore the pattern and untrack the existing entries (files remain on disk so the active loop keeps working). * audit: pin commentable-lines cache to checkoutSha, not just PR number a second checkout_pr(N) call advances toolState.checkoutSha at line 305 or 334, then runs fetchAndFormatPrDiff + cache population at line 549. any throw between those two points (rate limit, 5xx, network blip) left the old snapshot keyed to (pullNumber=N) while checkoutSha now points at a different sha. review_pr(N) would reuse the stale snapshot, silently validating comments against the wrong anchor — the original failure this cache was meant to prevent. track commentableLinesCheckoutSha alongside the pull number and require both to match before returning the cache. if either has moved, fall back to listFiles like any other miss. * audit: auto-clear leftover pending review from killed prior runs a workflow timeout or OOM between createReview PENDING and submitReview leaves GitHub holding a pending draft. the next run hits GitHub's one-pending-per-user-per-PR limit and 422s at pending-create, with no way to recover short of a human cleaning up manually. catch 422 at pending-create, list the PR's reviews (GitHub only exposes our own pending to us, so the filter is safe), delete the leftover, and retry once. 404/422 on the cleanup are treated as no-ops (race with another concurrent cleanup or the draft was submitted); any other cleanup error rethrows so the real cause reaches the caller. * audit: extract + unit-test stranded-pending-review cleanup the recovery branch inside createAndSubmitWithFooter had no direct test coverage. a regression in any of its guards (status check, message match, listReviews filter, 404/422 tolerance, non-retryable rethrow) would silently cause either destructive deletes of unrelated reviews or the old failure mode where a stranded pending draft blocks every retry. extract to clearStrandedPendingReview so the cases can be exercised with a mocked octokit, and add tests for each branch — including the load-bearing negative cases (non-422 passthrough, non-pending-review 422 passthrough, no-leftover-found passthrough, non-retryable cleanup error passthrough). no behavior change at the call site. * audit: document concurrent-run race in clearStrandedPendingReview two runs on the same PR using the same GitHub App installation token would both see each other's PENDING draft via listReviews (GitHub exposes PENDING only to the author, and both runs share authorship). the loser's recovery path would delete the winner's active draft, causing the winner's submitReview to 404. no reliable in-request signal distinguishes a genuinely-stranded prior-run draft from an active peer's draft — PENDING reviews have no created_at, and the user field is the same bot in both cases. the correct fix is workflow-level concurrency (a per-PR concurrency key), not a heuristic here. document the limitation so future readers don't try to bolt on a broken heuristic. * audit: report signal-killed subprocesses as failures, not exit code 0 node's close event delivers (code=null, signal=<name>) when a child is killed by signal (OOM killer, segfault, external SIGTERM). the close handler captured only exitCode and coerced null to 0 via `exitCode || 0`, so lifecycle hooks killed by signal were silently reported as successful — lifecycle.ts's `if (result.exitCode !== 0)` check skipped the warning and callers proceeded as if setup/post-checkout/prepush had completed. now capture signal, append "killed by signal <name>" to stderr, and resolve with exitCode=1 when code is null but signal is set. adds a regression test that spawns `kill -KILL \$\$` and asserts a non-zero exit plus the signal-kill marker in stderr. * audit: untrack RUN_ISSUES*.md ralph-loop working docs same pattern called out in 4f14dbf1: these files are per-run harness state and analysis scratch, not merge-to-main deliverables. the TODO literally opens with "Ralph loop instructions:", so it's unambiguously in the same category as .claude/ralph-loop-prompt.md was. files stay on disk so the active loop keeps working. * audit: block refs/... + symbolic-ref bypass of default-branch guard push_branch's restricted-mode guard compared the resolved remoteBranch against defaultBranch with exact-string equality. an agent passing branchName "refs/heads/main" flowed through: rejectIfLeadingDash passed, getPushDestination's fallback preserved the refs/heads/main string as remoteBranch, so "refs/heads/main" !== "main" and the block was skipped, yet git push happily resolved refs/heads/main to the local main commit and pushed to the remote main branch. symbolic refs (HEAD / FETCH_HEAD / ORIG_HEAD / MERGE_HEAD) are the same class of bypass — they resolve to whatever commit they point at, unconstrained by the name-based guard. add rejectSpecialRef to enforce bare branch names at the tool entry, use it in push_branch and delete_branch. checkout_pr only ever assigns pr-<number> as the local branch, so nothing legitimate relied on the refs/... form here. * audit: keep original 422 visible when listReviews fails during pending-review cleanup if listReviews threw (e.g. transient 502, rate limit) during the stranded pending-review recovery path, the listing failure replaced the original 422 "pending review" error when it propagated up through the tool's outer catch. agents then saw a generic server error with no mention of the real blocker and stopped retrying the cleanup. now the listing failure is logged at debug but does not mask the original 422. the caller's retry re-attempts cleanup, which succeeds if the listing failure was transient. * audit: block default-branch deletion even under push: enabled delete_branch required push: enabled, but within that mode the agent could delete the default branch with no local guard. GitHub branch protection usually catches this at the remote, but not every repo has protection configured — and even when it does, relying on remote config for local safety is wrong. pushing to main is reversible (revert, force-push old HEAD); deleting main is not (reflog recovery only, 30-day window). block deletion of the resolved default_branch in DeleteBranchTool regardless of push permission. push: enabled authorizes pushes, not wholesale removal of the repository's primary branch. * audit: attach no-op catch to agentPromise so a late rejection can't crash cleanup agentPromise raced against activityTimeout.promise (and the --timeout timeoutPromise), both of which had .catch(() => {}) handlers. agentPromise did not. if a timeout won the race, agentPromise became stranded and its subsequent rejection was an unhandled rejection — under node 15+'s default unhandled-rejection policy that terminates the process, which would kill main() mid-cleanup and lose the error-reporting and usage-summary work queued in the catch/finally blocks. the race still sees the rejection (the original promise is shared); this catch only prevents node from treating a post-race rejection as unobserved. * audit: close push_branch refspec-injection via ':' / '+' in branchName rejectSpecialRef only forbade leading-dash, `refs/` prefix, and symbolic refs. git push accepts `[+]src[:dst]` refspec syntax, so an agent under push:restricted could smuggle a full refspec through branchName and bypass the downstream exact-string default-branch guard: "evil:refs/heads/main" → push local 'evil' to remote main ":refs/heads/main" → delete remote main ":other" → delete arbitrary branches (outside grant) "+main" → force-push refspec prefix reject ':', '+', '^', '~', '?', '*', '[', '\\', and whitespace — git's own check-ref-format forbids all of them in branch names, so the allow-list cannot false-positive against a legitimate branch. add regression tests. * audit: stop suggesting blocked 'rebase' in push_rejected advice under shell=disabled Why: when push fails with non-fast-forward, the advice told the agent to run 'git rebase origin/...'. In shell=disabled mode the git MCP tool blocks rebase (as an arbitrary-code-execution escape hatch), so the agent's only path forward was to hit the block, read the fallback message, and try merge — one wasted round trip. Now: under shell=disabled we directly suggest 'git merge origin/...', which always works. Under other modes the advice keeps the rebase/merge choice but leads with merge so the example is copy-pastable either way. * audit: harden includeIf cleanup against shell-injection via subsection names setupGit read `includeif.*` keys via `git config --get-regexp`, split on the first space, and fed the result into `execSync(\`git config --unset "${key}"\`)`. git config subsection values preserve arbitrary characters, so a crafted `[includeIf "gitdir:$(touch${IFS}/tmp/pwn)safe"]` entry round-trips through `--get-regexp` with its `$(...)` command substitution intact, survives the split-on-space filter (IFS-bypass leaves the payload space-free), and gets evaluated when interpolated into the shell command. Confirmed reachable as an RCE sink in local repro. Switch to `--get-regexp -z` (null-terminated, no ambiguity on whitespace) and call `$("git", ["config", "--unset-all", key])` which uses spawn-array and never hands the key to a shell. Extract the logic into `removeIncludeIfEntries` and add regression tests covering the injection payload, whitespace-in-subsection keys, benign entries, and the no-op case. * audit: clear SIGKILL escalator on clean SIGTERM exit the overall-timeout path scheduled a 5s SIGKILL follow-up without capturing the timer id. if the child cooperated with SIGTERM and `close` fired promptly, the escalator stayed pending in the event loop for up to 5s — delaying any subsequent clean shutdown (e.g. the main action exiting after an agent timeout) by that long. capture sigkillEscalatorId alongside timeoutId and clear it in both close and error handlers. regression test asserts the active-timer count does not grow past the pre-spawn baseline after a timed-out child exits on SIGTERM. * audit: correct rebase-availability hints to reflect shell=restricted the MCP git tool only blocks rebase when shell=disabled (NOSHELL_BLOCKED_SUBCOMMANDS check in GitTool). under shell=restricted, git({command: "rebase"}) works fine through the tool — NOSHELL_BLOCKED_SUBCOMMANDS doesn't apply. but two agent-facing messages implied rebase is only available with shell=enabled: - AUTH_REQUIRED_REDIRECT["pull"] said "rebase is only available when shell is enabled" - push-rejected integrateStep (non-disabled branch) said "(or 'rebase' if shell is enabled)" under shell=restricted, agents reading these would wrongly think they had to pick merge — pushing them toward merge commits when rebase would have been cleaner. the push-rejected branch is already ternary-gated on shell !== "disabled", so the qualifier there was just redundant noise. * audit: block difftool/mergetool under shell=disabled git difftool -x <cmd> is the short form of --extcmd. the args blocklist only matches --extcmd / --extcmd=*, so -x slipped through and let an agent run arbitrary commands even when shell=disabled. globally blocking -x would false-positive on git cherry-pick -x, which only appends metadata, so block difftool (and mergetool, same shape via mergetool.<name>.cmd) at the subcommand level instead. agents have no legitimate need for either — diffs go through diff/show and merges are resolved by file edits. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * audit: recover stranded PENDING drafts on no-body createReview too The body path already clears a stranded PENDING draft from a prior crashed run via createAndSubmitWithFooter's own try/catch. The no-body path (approve-with-no-feedback or comments-only) called createReview directly — so a PR whose previous body-path run crashed between createReview(PENDING) and submitReview would permanently 422 any subsequent no-body review with "already has a pending review" until a body-path run happened to clear it. Factored out createReviewWithStrandedRecovery so both paths get the same recovery treatment, and added regression tests covering the no-stranded / stranded-and-retry / non-stranded-422-no-retry cases. * audit: reject timeouts past node's setTimeout ceiling a user-supplied timeout like "999h" parses fine (parseTimeString has no upper cap) but falls off the 2^31-1 ms limit setTimeout clamps to 1ms. the agent run would reject with "timed out after 999h" in a single tick. extract a resolveTimeoutMs helper that centralizes the zero/overflow/ unparseable checks (previously scattered behind inline boolean logic in main.ts) and cover the behavior with unit tests including the boundary value. * fix(#22): replace parameter property in SpawnTimeoutError node --experimental-strip-types rejects readonly/public/private param properties in constructors. tests run via node directly (no tsc), so CI was hitting ERR_UNSUPPORTED_TYPESCRIPT_SYNTAX on every action-agents / action-agnostic job before any test code ran. declare the field and assign in the body instead. * audit: tighten git tool description and delete_branch refspec - `git` tool description previously implied `pull` had a dedicated MCP tool alongside `push_branch`/`git_fetch`. it doesn't — the redirect sends the agent back to the same git tool with `command: "merge"` (or `rebase`). update the description to teach this directly instead of letting agents discover it through the redirect error. - `delete_branch` now passes `refs/heads/${branchName}` to `git push --delete` so a same-named tag can't be silently deleted when both exist on the remote. `rejectSpecialRef` already guarantees the bare-name invariant, so the template construction stays injection-safe. Made-with: Cursor * audit: polish review.ts per anneal findings - drop `as "LEFT" | "RIGHT"` cast in `validateInlineComments` — octokit types `side?: string` at the createReview endpoint, so narrow via `c.side === "LEFT" ? "LEFT" : "RIGHT"`. no cast, no redundant annotation — TS infers the literal union from the ternary. - consolidate `clearStrandedPendingReview` from 3 params to 2 by folding `originalErr` into `params`, per AGENTS.md "max 2 parameters" rule. updates both call sites (`createReviewWithStrandedRecovery`, `createAndSubmitWithFooter`) and all 7 test paths. - upgrade `listReviews`-during-cleanup failure log from `log.debug` to `log.info` so operators not running at debug still see that recovery was attempted before the original 422 bubbles up. message now reads "surfacing original 422" to make the intent unambiguous. Made-with: Cursor * audit: signal partial commit metadata in checkout_pr previously a rev-list/log failure (e.g. shallow fetch where `origin/<base>` isn't reachable) silently returned `commitCount: 0, commitLog: ""` — indistinguishable from "this PR has no commits past base", which could mislead review reasoning about scope. add a `commitLogUnavailable: boolean` field to `CheckoutPrResult`, set when the rev-list/log calls throw. instructions footer now tells the agent to treat the values as "unknown" rather than "no commits" in that case. message phrased to cover the rare case where rev-list succeeds but git log throws (partial, not strictly zero) metadata. Made-with: Cursor * audit: fix parseDiffTocEntries to match production ' · diff-<sha>' TOC suffix the regex required $ right after the line range, but formatFilesWithLineNumbers in checkout.ts appends ` · diff-<sha256>` so agents have the GitHub "Files Changed" anchor precomputed. result: tocEntries was always empty on real PR reviews, breakdown.files was empty, and runDiffCoveragePreflight never fired its one-time "read the diff" nudge. add an optional suffix to the regex and a regression test that uses the exact production TOC shape. Made-with: Cursor * audit(#20): skip empty downgraded-APPROVE reviews before they 422 GitHub rejects `event: "COMMENT"` reviews with no body and no inline comments (HTTP 422 "Unprocessable Entity", verified empirically on repos/pullfrog/preview-546-run-issues-fixes/pulls/1). the runtime `prApproveEnabled` downgrade folds approved=true into event=COMMENT when the repo flag is off, so an agent asking to APPROVE a PR with no other feedback produces exactly that rejected shape — but the existing empty-review skip only fired for !approved cases, so the tool POSTed the doomed COMMENT, octokit returned what looked like a success-with- no-persisted-review shape, and agents reported a phantom reviewId that 404s on any subsequent GET. extract the skip decision into `reviewSkipDecision` and add a second branch for approved + !prApproveEnabled + empty. the function returns null when the review should be submitted, so a real bare APPROVE (approved + prApproveEnabled + empty) still goes through unchanged — GitHub accepts empty APPROVE reviews because the stamp itself is the content. surfaced in the PR #546 preview e2e run 24678139563 (reviewId 4141786854 reported by the agent but absent from every reviews listing). TC13 run 24680349445 re-ran the same scenario with prApproveEnabled=enabled and the review persisted correctly, isolating the cause to the downgrade + empty interaction. * audit(#31): drop misleading rebase mention from pull redirect AUTH_REQUIRED_REDIRECT["pull"] and the git tool's top-level description both said "use git_fetch then this tool with command 'merge' (or 'rebase' unless shell is disabled)". the "(or 'rebase' unless shell is disabled)" qualifier is active misinformation when the agent is already running under shell=disabled: rebase is blocked there by NOSHELL_BLOCKED_SUBCOMMANDS, so the suggestion sends the agent into a second block on the next tool call. 3b83ee97 already fixed this pattern for the push-rejected advice at line 248, but the pull redirect at line 280 and the tool description at line 351 were missed. the right copy isn't a conditional qualifier that agents have to parse against their own shell mode — it's just naming the one alternative that works everywhere (merge). agents under shell=restricted/enabled who want rebase can invoke it directly; the redirect doesn't need to advertise it. verified in preview e2e run 24679728733 (TC8 probe 6) where the agent correctly captured the verbatim redirect message under shell=disabled and explicitly flagged the "(or 'rebase' unless shell is disabled)" clause as confusing — the new test in security.test.ts asserts the message names merge and never rebase in every shell mode. * audit: drop vestigial entry/post references + add preview-546 settings util followup to d79860c6 "refactor: flatten action entrypoints" (Apr 10), which moved action.yml from built `entry`/`post` files to source `entry.ts`/`post.ts` but left three stale references lying around: - .gitignore: `action/run/entry` / `action/dispatch/entry` paths no longer exist anywhere in the build. - .github/workflows/pull-from-action.yml: agent instruction told the upstream sync agent to "Ignore `entry` files (they are built artifacts and .gitignored in this repo)". there are no built entry artifacts anymore — entry.ts is source. - .cursor/settings.json: search.exclude pattern "**/entry" excluded the old built files that no longer exist. none of these were load-bearing on their own, but the same drift had already broken preview e2e end-to-end: the pullfrog/template workflow's three-file copy step (cp .../entry, cp .../post) silently failed with cp: no such file on every preview PR since Apr 10. that template fix went to pullfrog/template@7ec7c8d and the preview-546 mirror at @17ab585, which is what unblocked this PR's full e2e validation. also adds scripts/preview-546-settings.ts, the helper used during the e2e validation to show/set/reset DB-level repo settings on the Neon preview branch (push, shell, prApproveEnabled, hook scripts). scoped to this preview repo ID so it can't accidentally mutate prod. * audit(#11): scope removeIncludeIfEntries to repoDir under inherited GIT_* the function takes `repoDir` as the target, but plain execSync / $(...) inherit GIT_DIR, GIT_WORK_TREE, and GIT_INDEX_FILE from the parent process — and `git config --local` honors GIT_DIR over cwd. when this runs as a child of another git invocation (notably the pre-push hook, but also any future caller embedded inside a git subcommand), the cleanup silently targets the outer repo instead of repoDir. latent today because the real caller is ASKPASS setup, which runs before any git-subcommand ancestor exists, but the function's contract still promised the wrong thing — and the test suite hit exactly this bug when invoked through `git push`. - envScopedToRepo() strips GIT_* before both the get-regexp and unset calls, so cwd wins. - swap the $(...) shell helper for execFileSync on the unset call. $() would merge our scoped env with a "restricted" base that's tuned for hook execution (no tokens) — overkill here and it re-introduces the shell-vs-argv distinction this function was explicitly hardened against in a9aa3b2b. execFileSync with argv is the right tool for a call where the key can contain arbitrary characters. - setup.test.ts also strips GIT_* in its own execSync harness so the suite passes identically under `pnpm vitest run`, `pnpm -r test`, and `git push`'s pre-push hook. --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> Co-authored-by: Colin McDonnell <colinmcd94@gmail.com> |
||
|
Colin McDonnell
|
56a5d29598 |
add diff coverage preflight for PR review submissions (#544)
* add one-time diff coverage preflight for PR reviews track diff read coverage from agent tool-use events and run a one-time pre-flight before review submission, with explicit coverage skip reasons for low-value files like lockfiles. Made-with: Cursor * add manual dispatch fallback for preview deploy workflow allow preview repo and preview sync jobs to be run via workflow_dispatch with explicit PR number and branch inputs, so preview provisioning can be retriggered when pull_request events fail to fire. Made-with: Cursor * fix manual preview dispatch PR input wiring use normalized PR number and branch env values for comment creation and script env wiring so workflow_dispatch preview runs can create and update PR-specific preview resources. Made-with: Cursor * remove obsolete snapshots invalidated by checkout instructions change * fix diff coverage read offset handling and add local sanity-check guidance normalize read offset semantics for diff coverage tracking, reuse shared range counting in review preflight, add focused diff coverage unit tests, and document the local play.ts testing workflow in AGENTS.md. Made-with: Cursor * add regenerated mcp test snapshots capture snapshot files generated by the review comment and checkout formatting tests during pre-push validation so the branch remains clean and reproducible. Made-with: Cursor * add diff coverage preflight instrumentation logs log diff coverage initialization in checkout_pr and emit preflight state/breakdown diagnostics in create_pull_request_review to debug missing coverage enforcement in preview e2e runs. Made-with: Cursor * add env override to force local cli execution in action runtime support explicit local-cli execution via PULLFROG_FORCE_LOCAL_CLI so preview workflows can run branch action code instead of the npm fallback package during e2e debugging. Made-with: Cursor * add preview e2e debugging learnings for action runtime validation capture the preview execution-path gotchas and one-time preflight verification pattern in AGENTS.md so future investigations validate the real runtime and avoid npm fallback confusion. Made-with: Cursor * reduce diff coverage log noise while preserving failure visibility downgrade verbose diff coverage lifecycle diagnostics to debug, keep a concise info-level pre-flight failure signal, and document preview runtime debugging learnings in AGENTS.md. Made-with: Cursor * WIP * tune sync.md: ff override + softer overlap verification Made-with: Cursor * chore: bump models snapshot for claude-opus-4-7 Made-with: Cursor * rip out coverage_skips waiver from diff coverage pre-flight Made-with: Cursor --------- Co-authored-by: pullfrog[bot] <226033991+pullfrog[bot]@users.noreply.github.com> |
||
|
Colin McDonnell
|
4b3c5ca905 |
rename agent key to opencode and add skill invocation coverage (#541)
* rename agent key to opencode and add skill invocation coverage. add skill-invoke tests for claude and opencode with local play-based validation signals, update CI matrices, and include the current tracked refactors in this branch for review. Made-with: Cursor * exclude agent-specific skill-invoke tests from wrong agent in CI matrix * address review follow-up and preserve workflow run UI tweak. switch changed-agents ci coverage to exercise the opentoad implementation path while keeping the opencode expectation, and include the local workflow run client interaction updates requested on this branch. Made-with: Cursor * remove opentoad agent filename from runtime. rename the opencode harness implementation file from opentoad.ts to opencode.ts and update ci coverage input accordingly so action code no longer carries the old filename. Made-with: Cursor * ensure security prompt bypass is set on every test fixture. this keeps adversarial and permissions harnesses from being blocked by the default prompt-injection refusal path during CI security tests. Made-with: Cursor --------- Co-authored-by: pullfrog[bot] <226033991+pullfrog[bot]@users.noreply.github.com> |
||
|
Colin McDonnell
|
8ee9e3176a |
remove generate-proxies postinstall hack, resolve pullfrog source via bundler config
the postinstall script referenced scripts/generate-proxies.ts which isn't included in the published npm package, silently breaking every npx install. replaced the proxy stub approach with turbopack resolveAlias and webpack conditionNames so both bundlers resolve pullfrog imports to TypeScript source directly — matching what tsc already does via customConditions. also moves PR summary format from handleWebhook into modes.ts so the summarize mode prompt includes it directly. Made-with: Cursor |
||
|
Colin McDonnell
|
255f29efb8 |
omit prior review feedback section entirely when nothing was addressed
Made-with: Cursor |
||
|
Colin McDonnell
|
b282e8b599 |
Improve incremental review output and fix todo tracker race (#529)
* improve incremental review output and fix todo tracker race - reviewed changes section: summarize at logical-change level with past-tense verbs, not per-file enumerations - add TodoTracker.completeAll() to mark all non-cancelled items as completed before snapshotting the collapsible in review/progress posts Made-with: Cursor * completeAll -> completeInProgress: only mark in-progress items as completed Pending items that were genuinely skipped stay as-is in the collapsible, so the task list honestly reflects what the agent actually did. Made-with: Cursor |
||
|
Colin McDonnell
|
2759206a67 |
update stale model snapshot (glm-5.1 replaced qwen3.6-plus-free)
Made-with: Cursor |
||
|
Colin McDonnell
|
08101a0e67 |
summarize mode: drop subagent delegation and dead effort hint
the "delegate a subagent" instruction doubled LLM sessions for every summary run, and "use mini or auto effort" was a no-op since the agent always runs at high/max effort. Made-with: Cursor |
||
|
David Blass
|
1c730300b6 | Clarify push, prepush, and progress errors in agent prompts (#521) | ||
|
Colin McDonnell
|
4bb280cd0a |
incremental review: improve no-new-issues body text
Made-with: Cursor |
||
|
Colin McDonnell
|
8f7145e716 |
simplify incremental review summaries to bullet points
Made-with: Cursor |
||
|
Colin McDonnell
|
b9b6503315 |
reorder prompt sections: task-first with dynamic TOC (#513)
* reorder prompt sections: task-first with dynamic TOC put the actual task at the top of the prompt for primacy, add a dynamic table of contents, and push system/runtime metadata to the end. new section order: TOC → YOUR TASK → PROCEDURE → EVENT CONTEXT → SYSTEM → LEARNINGS → RUNTIME Made-with: Cursor * enforce clean working tree: continue session if agent leaves uncommitted changes after each agent run, check `git status --porcelain`. if dirty, resume the same session with instructions to commit on a new branch, push, and open a PR. retries up to 3 times before giving up. - claude code: capture session_id from result event, use --resume <id> - opencode: use --continue to resume the last session - remove --no-session-persistence from claude (needed for --resume) - update Task mode to clarify branch/push/PR is the default finalize step Made-with: Cursor * log full prompt in collapsible group for debugging Made-with: Cursor * fix: format tool refs in buildCommitPrompt via formatMcpToolRef * enforce clean git status: general instructions, stop hook, and Task mode Made-with: Cursor * fix: rename stale titleBody references after body leak fix Made-with: Cursor --------- Co-authored-by: pullfrog[bot] <226033991+pullfrog[bot]@users.noreply.github.com> |
||
|
Colin McDonnell
|
6b93e6b368 |
review/incremental-review: always submit review, never call report_progress (#516)
* WIP * WIP * review/incremental-review: always submit review, never call report_progress The progress comment is auto-deleted by the stranded-comment cleanup in main.ts when the agent skips report_progress. This makes reviews the sole PR artifact for both modes, reducing noise. - soften report_progress tool description to allow mode opt-out - Review mode: always submit exactly one review (approve or request changes) - IncrementalReview mode: submit review for substantive outcomes, silently exit for non-substantive changes (formatting-only pushes produce zero artifacts) Made-with: Cursor * incremental-review: clarify approval condition for substantive no-issues case Made-with: Cursor * report_progress: s/completed/current task list Made-with: Cursor * system instructions: align report_progress guidance with mode opt-out Made-with: Cursor |
||
|
Colin McDonnell
|
70d56ebc89 |
improve review quality: add --effort flag, subagent guidance, remove dead prompts (#508)
* Update waitlist, run ralph experiments * improve review quality: add --effort flag, subagent guidance, remove dead prompts - add --effort high/max to Claude Code CLI (max for Opus, high for Sonnet/Haiku). default was silently dropped from high to medium in March 2026. - add subagent guidance to Review/IncrementalReview modeGuidance for parallel investigation of large cross-cutting PRs (read-only, no side effects). - remove "THINK HARDER" from mode prompts (vestigial, no longer controls thinking). - remove redundant mode.prompt bodies from modes.ts — the actual guidance lives in modeGuidance (selectMode.ts) and mode.prompt was dead code for all built-in modes since the delegation system was removed in March. Made-with: Cursor * make Mode.prompt optional, remove ModeSchema dead code prompt is only needed by custom user-defined modes (validated by Zod modeSchema in utils/schemas/modes.ts). built-in modes get their guidance from modeGuidance in selectMode.ts. the arktype ModeSchema was never imported anywhere. Made-with: Cursor * make modes.ts the single source of truth for mode guidance move all mode guidance from modeGuidance in selectMode.ts into mode.prompt in modes.ts. selectMode.ts now only contains the runtime tool logic (resolving modes, merging user instructions, handling PlanEdit/SummaryUpdate overrides). this eliminates the confusing fallback chain where someone editing mode instructions had to know to look in selectMode.ts rather than modes.ts. Made-with: Cursor * add self-review subagent step to Build mode, update wiki Build mode now delegates a read-only subagent to review the diff before committing, catching bugs/logic errors/edge cases that the builder might miss. Also updates wiki/modes.md to reflect the single-source-of-truth architecture (modes.ts owns all guidance, selectMode.ts is pure runtime logic). Made-with: Cursor * update model snapshot (openrouter qwen3.6-plus rename) Made-with: Cursor |
||
|
Colin McDonnell
|
6c5d228c04 |
allow external_directory reads — not a security boundary
Made-with: Cursor |
||
|
Mateusz Burzyński
|
a7b8dcbced |
Rework incremental diffing (#499)
* Improve our deepening logic * Use consistent SHA for PR-related operations in CheckoutPrTool * compute `deepenDepth` at more appropriate time * fix stale comment * add comments for `alreadyOnBranch` * ensure before sha is available * small cleanup * computeIncrementalDiff * move the util * improve algorithm * improve algorithm further * get rid of temp result array * add comment * compute incremental diff and updte instructions * add comment * update stale comment * get rid of redundant rev-parse call * improve comment * strenghten the instructions * make diff paths unique |
||
|
David Blass
|
6b18b6730b |
live todo tracking, collapsible task list in final progress, hide set_output outside standalone (#492)
* fix false "without reporting progress" error + live todo tracking clean up orphaned progress comments when review is skipped or only set_output is used, preventing the false positive in handleAgentResult. parse todowrite events from OpenCode's NDJSON stream and render a live markdown checklist in the PR progress comment (2s debounce). agent's explicit report_progress always takes priority. Made-with: Cursor * fix contradictory review/progress prompting align Review and IncrementalReview mode prompts with their guidance — mode prompts said "always submit" while guidance said "skip if clean." remove the empty-approval submission that was silently dropped by the tool. make progress comment lifecycle explicit: created on first call, updated in place, removed after review submission. Made-with: Cursor * centralize todo tracking into shared TodoTracker module extract inline todo tracking logic (~95 lines) from opentoad.ts into action/utils/todoTracking.ts. the tracker is created once in main.ts and passed to agents via AgentRunContext.todoTracker, making it agent-agnostic and reusable for future agent implementations. Made-with: Cursor * fix todoTracker optional type to match file convention add | undefined to todoTracker in AgentRunContext, matching every other optional property in the same interface. Made-with: Cursor * instruct agents to always maintain a task list for live progress system prompt now tells agents to create an internal task list at the start of every run. the tracker renders it to the progress comment automatically. report_progress is reserved for final results only — no more intermediate "Checking..." messages that cancel the tracker and leave stale text on the comment. Made-with: Cursor * require report_progress summary at end of every run agents must always call report_progress with a final summary — the completed task list should never be the end state of the progress comment. updated all review mode prompts to call report_progress after submitting (or not submitting) a review. Made-with: Cursor * keep progress comment after review with final summary stop deleting the progress comment after review submission — the agent now always calls report_progress with a summary at the end, and that summary should persist as a record of what was done. Made-with: Cursor * harden stranded progress comment cleanup - main.ts: detect when tracker was last writer (agent never called report_progress) and delete the stranded checklist instead of leaving it as the final comment state - postCleanup.ts: expand stuck-comment detection to also catch stranded todo checklists (regex match for checklist patterns) when the process is killed before normal cleanup runs - modes.ts + selectMode.ts: add report_progress step to Summarize and SummaryUpdate modes (only modes that were missing it) Made-with: Cursor * fix stale comments, typo, and build mode redundancy - comment.ts: update deleteProgressComment docstring and inline comment to reflect current usage (stranded-comment cleanup, not post-review) - modes.ts: merge duplicate report_progress steps (8 + 10) into single step 9, fix "optimizatfixons" typo - wiki/post-cleanup.md: document checklist detection regex Made-with: Cursor * collapsible completed todos in final progress, hide set_output outside standalone mode - add renderCollapsible() to TodoTracker, append completed task list as <details> section when agent calls report_progress - cancel tracker after agent's final report_progress so it doesn't overwrite with raw checklist - conditionally register SetOutputTool only in standalone mode or when output_schema is provided - remove unconditional set_output instruction from orchestrator task section - update Summarize/SummaryUpdate mode guidance to not reference set_output Made-with: Cursor * show completion count in collapsible task list summary Made-with: Cursor * only count completed (not cancelled) in collapsible task list summary Made-with: Cursor * reinforce concise summary prompting across system prompt, modes, and tool description Made-with: Cursor * address review feedback: wasUpdated bypass, tracker false-positive, race condition - remove wasUpdated=true from cleanup paths so handleAgentResult correctly detects genuinely silent runs - add hadProgressComment to ToolState as immutable snapshot for the safety check - use todoTracker.hasPublished instead of enabled for stranded-comment cleanup - serialize onUpdate calls via inflightPromise chain with post-cancel guard - add settled() to wait for in-flight updates before writing final summary Made-with: Cursor * address round-2 review: hasPublished after success, finalSummaryWritten flag - set hasPublished only after onUpdate resolves (not before) so failed writes are not counted as published - add finalSummaryWritten flag to ToolState, set after successful non-plan reportProgress; decouple cleanup detection from todoTracker.enabled so it survives API failures where cancel() ran but the write didn't succeed Made-with: Cursor |
||
|
Colin McDonnell
|
b31800c213 |
clarify PR summary instructions for readable section titles
Made-with: Cursor |
||
|
Colin McDonnell
|
4a8c432a48 |
add Summarize mode for updatable PR summary comments (#470)
* add Summarize mode for updatable PR summary comments Introduces a Summarize mode that manages a single summary comment per PR, updated in place on subsequent pushes. Mirrors the Plan/PlanEdit pattern: API endpoint for existing-comment lookup at select_mode time, node ID tracking on WorkflowRun, and SummaryUpdate guidance for edits. Also fixes summary format instructions: Before/After uses inline <br/> to avoid double line breaks, metadata line placed after key changes, SHA-256 anchor instructions strengthened against fabrication. Made-with: Cursor * fix pre-existing lint error in checkout.ts Made-with: Cursor * fix dead restricted param in deepenForBeforeSha GitAuthOptions dropped the restricted field in the ASKPASS refactor (#478) but deepenForBeforeSha (#471) still passed it. Remove the field and the now-unused shell param from DeepenForBeforeShaParams. Made-with: Cursor |