upgrade Claude to Opus 4.6 with effort levels (#256)

* upgrade Claude to Opus 4.6 with --effort max for --max mode

- mini: haiku → sonnet
- auto: opusplan → opus (Opus 4.6)
- max: opus → opus + --effort max (Opus 4.6 max effort)
- bump @anthropic-ai/claude-agent-sdk 0.2.7 → 0.2.39

Co-authored-by: Cursor <cursoragent@cursor.com>

* update action lockfile for claude-agent-sdk 0.2.39

Co-authored-by: Cursor <cursoragent@cursor.com>

* add tool_use_summary handler for SDK 0.2.39

Co-authored-by: Cursor <cursoragent@cursor.com>

* integrate gpt-5.3-codex with runtime model availability detection

checks GET /v1/models at agent start to determine if gpt-5.3-codex is
available for the API key, falling back to gpt-5.2-codex when it isn't.
model resolution runs concurrently with CLI install for zero added latency.
also bumps @openai/codex-sdk from 0.80.0 to 0.98.0.

Co-authored-by: Cursor <cursoragent@cursor.com>

---------

Co-authored-by: Cursor <cursoragent@cursor.com>

agents/claude.ts

@@ -17,16 +17,18 @@ import { type AgentRunContext, agent } from "./shared.ts";
 // model selection based on effort level
 // these are aliases that always resolve to the latest version
 const claudeEffortModels: Record<Effort, string> = {
-  mini: "haiku",
+  mini: "sonnet",
-  auto: "opusplan",
+  auto: "opus",
   max: "opus",
 };
 
-// FUTURE: Consider using Anthropic's "effort" parameter (beta) with Opus.
+// Claude Code CLI --effort level per pullfrog effort
-// This would allow a single model with effort levels ("low", "medium", "high") controlling
+// null = use default (high). "max" is Opus 4.6 only.
-// token spend across responses, tool calls, and thinking. Requires beta header "effort-2025-11-24".
+const claudeEffortLevels: Record<Effort, string | null> = {
-// See: https://platform.claude.com/docs/en/build-with-claude/effort
+  mini: null,
-// This approach could replace model selection if effort proves effective for controlling capability.
+  auto: null,
+  max: "max",
+};
 
 /**
  * Build disallowedTools list from payload permissions.
@@ -81,9 +83,10 @@ export const claude = agent({
     // install CLI at start of run
     const cliPath = await installClaude();
 
-    // select model based on effort level
+    // select model and effort level
     const model = claudeEffortModels[ctx.payload.effort];
-    log.info(`» using model: ${model} (effort: ${ctx.payload.effort})`);
+    const effortLevel = claudeEffortLevels[ctx.payload.effort];
+    log.info(`» using model: ${model}${effortLevel ? ` (effort: ${effortLevel})` : ""}`);
 
     // build disallowedTools based on tool permissions
     const disallowedTools = buildDisallowedTools(ctx);
@@ -110,6 +113,11 @@ export const claude = agent({
       "--verbose",
     ];
 
+    // add --effort flag if specified (e.g. "max" for Opus 4.6)
+    if (effortLevel) {
+      args.push("--effort", effortLevel);
+    }
+
     // add disallowed tools if any
     if (disallowedTools.length > 0) {
       args.push("--disallowedTools");
@@ -301,5 +309,6 @@ const messageHandlers: SDKMessageHandlers = {
   system: () => {},
   stream_event: () => {},
   tool_progress: () => {},
+  tool_use_summary: () => {},
   auth_status: () => {},
 };

agents/codex.ts

@@ -15,14 +15,52 @@ import { type AgentRunContext, agent } from "./shared.ts";
 
 // configuration based on effort level
 // https://developers.openai.com/codex/models/
-// gpt-5.3-codex announced 2026-02-05 but not yet available in codex CLI
 type ModelReasoningEffort = "minimal" | "low" | "medium" | "high" | "xhigh";
 type CodexEffortConfig = { model: string; reasoningEffort?: ModelReasoningEffort };
-const codexEffortConfig: Record<Effort, CodexEffortConfig> = {
+
-  mini: { model: "gpt-5.1-codex-mini", reasoningEffort: "low" },
+// preferred model for auto/max — falls back to gpt-5.2-codex if API key lacks access
-  auto: { model: "gpt-5.2-codex" },
+const PREFERRED_MODEL = "gpt-5.3-codex";
-  max: { model: "gpt-5.2-codex", reasoningEffort: "high" },
+const FALLBACK_MODEL = "gpt-5.2-codex";
-};
+
+function getCodexEffortConfig(model: string): Record<Effort, CodexEffortConfig> {
+  return {
+    mini: { model: "gpt-5.1-codex-mini", reasoningEffort: "low" },
+    auto: { model },
+    max: { model, reasoningEffort: "high" },
+  };
+}
+
+// check if a model is available for the given API key via GET /v1/models
+async function isModelAvailable(ctx: { apiKey: string; model: string }): Promise<boolean> {
+  try {
+    const response = await fetch("https://api.openai.com/v1/models", {
+      headers: { Authorization: `Bearer ${ctx.apiKey}` },
+      signal: AbortSignal.timeout(10_000),
+    });
+    if (!response.ok) {
+      log.warning(
+        `failed to list models (HTTP ${response.status}), falling back to ${FALLBACK_MODEL}`
+      );
+      return false;
+    }
+    const body = (await response.json()) as { data: Array<{ id: string }> };
+    return body.data.some((m) => m.id === ctx.model);
+  } catch (err) {
+    log.warning(`failed to list models: ${err}, falling back to ${FALLBACK_MODEL}`);
+    return false;
+  }
+}
+
+// resolve the best available model for auto/max effort levels
+async function resolveModel(apiKey: string): Promise<string> {
+  const available = await isModelAvailable({ apiKey, model: PREFERRED_MODEL });
+  if (available) {
+    log.info(`» ${PREFERRED_MODEL} is available for this API key`);
+    return PREFERRED_MODEL;
+  }
+  log.info(`» ${PREFERRED_MODEL} not available, using ${FALLBACK_MODEL}`);
+  return FALLBACK_MODEL;
+}
 
 function writeCodexConfig(ctx: AgentRunContext): string {
   const codexDir = join(ctx.tmpdir, ".codex");
@@ -95,14 +133,14 @@ export const codex = agent({
       throw new Error("OPENAI_API_KEY is required for codex agent");
     }
 
-    // install CLI at start of run
+    // install CLI and resolve model concurrently
-    const cliPath = await installCodex();
+    const [cliPath, model] = await Promise.all([installCodex(), resolveModel(apiKey)]);
 
     // write config file (creates ~/.codex/config.toml)
     const codexDir = writeCodexConfig(ctx);
 
     // get model and reasoning effort based on effort level
-    const effortConfig = codexEffortConfig[ctx.payload.effort];
+    const effortConfig = getCodexEffortConfig(model)[ctx.payload.effort];
     log.info(`» using model: ${effortConfig.model} (effort: ${ctx.payload.effort})`);
     if (effortConfig.reasoningEffort) {
       log.info(`» using modelReasoningEffort: ${effortConfig.reasoningEffort}`);

entry

@@ -144125,13 +144125,13 @@ var package_default = {
   },
   dependencies: {
     "@actions/core": "^1.11.1",
-    "@anthropic-ai/claude-agent-sdk": "0.2.7",
+    "@anthropic-ai/claude-agent-sdk": "0.2.39",
     "@ark/fs": "0.53.0",
     "@ark/util": "0.53.0",
     "@octokit/plugin-throttling": "^11.0.3",
     "@octokit/rest": "^22.0.0",
     "@octokit/webhooks-types": "^7.6.1",
-    "@openai/codex-sdk": "0.80.0",
+    "@openai/codex-sdk": "0.98.0",
     "@opencode-ai/sdk": "^1.0.143",
     "@standard-schema/spec": "1.0.0",
     "@toon-format/toon": "^1.0.0",
@@ -144442,10 +144442,15 @@ ${ctx.instructions.user}` : null,
 
 // agents/claude.ts
 var claudeEffortModels = {
-  mini: "haiku",
+  mini: "sonnet",
-  auto: "opusplan",
+  auto: "opus",
   max: "opus"
 };
+var claudeEffortLevels = {
+  mini: null,
+  auto: null,
+  max: "max"
+};
 function buildDisallowedTools(ctx) {
   const disallowed = [];
   if (ctx.payload.web === "disabled") disallowed.push("WebFetch");
@@ -144483,7 +144488,8 @@ var claude = agent({
   run: async (ctx) => {
     const cliPath = await installClaude();
     const model = claudeEffortModels[ctx.payload.effort];
-    log.info(`\xBB using model: ${model} (effort: ${ctx.payload.effort})`);
+    const effortLevel = claudeEffortLevels[ctx.payload.effort];
+    log.info(`\xBB using model: ${model}${effortLevel ? ` (effort: ${effortLevel})` : ""}`);
     const disallowedTools = buildDisallowedTools(ctx);
     if (disallowedTools.length > 0) {
       log.info(`\xBB disallowed tools: ${disallowedTools.join(", ")}`);
@@ -144502,6 +144508,9 @@ var claude = agent({
       "stream-json",
       "--verbose"
     ];
+    if (effortLevel) {
+      args3.push("--effort", effortLevel);
+    }
     if (disallowedTools.length > 0) {
       args3.push("--disallowedTools");
       args3.push(...disallowedTools);
@@ -144646,6 +144655,8 @@ var messageHandlers = {
   },
   tool_progress: () => {
   },
+  tool_use_summary: () => {
+  },
   auth_status: () => {
   }
 };
@@ -144653,11 +144664,43 @@ var messageHandlers = {
 // agents/codex.ts
 import { mkdirSync as mkdirSync4, writeFileSync as writeFileSync8 } from "node:fs";
 import { join as join10 } from "node:path";
-var codexEffortConfig = {
+var PREFERRED_MODEL = "gpt-5.3-codex";
-  mini: { model: "gpt-5.1-codex-mini", reasoningEffort: "low" },
+var FALLBACK_MODEL = "gpt-5.2-codex";
-  auto: { model: "gpt-5.2-codex" },
+function getCodexEffortConfig(model) {
-  max: { model: "gpt-5.2-codex", reasoningEffort: "high" }
+  return {
-};
+    mini: { model: "gpt-5.1-codex-mini", reasoningEffort: "low" },
+    auto: { model },
+    max: { model, reasoningEffort: "high" }
+  };
+}
+async function isModelAvailable(ctx) {
+  try {
+    const response = await fetch("https://api.openai.com/v1/models", {
+      headers: { Authorization: `Bearer ${ctx.apiKey}` },
+      signal: AbortSignal.timeout(1e4)
+    });
+    if (!response.ok) {
+      log.warning(
+        `failed to list models (HTTP ${response.status}), falling back to ${FALLBACK_MODEL}`
+      );
+      return false;
+    }
+    const body = await response.json();
+    return body.data.some((m) => m.id === ctx.model);
+  } catch (err) {
+    log.warning(`failed to list models: ${err}, falling back to ${FALLBACK_MODEL}`);
+    return false;
+  }
+}
+async function resolveModel(apiKey) {
+  const available = await isModelAvailable({ apiKey, model: PREFERRED_MODEL });
+  if (available) {
+    log.info(`\xBB ${PREFERRED_MODEL} is available for this API key`);
+    return PREFERRED_MODEL;
+  }
+  log.info(`\xBB ${PREFERRED_MODEL} not available, using ${FALLBACK_MODEL}`);
+  return FALLBACK_MODEL;
+}
 function writeCodexConfig(ctx) {
   const codexDir = join10(ctx.tmpdir, ".codex");
   mkdirSync4(codexDir, { recursive: true });
@@ -144709,9 +144752,9 @@ var codex = agent({
     if (!apiKey) {
       throw new Error("OPENAI_API_KEY is required for codex agent");
     }
-    const cliPath = await installCodex();
+    const [cliPath, model] = await Promise.all([installCodex(), resolveModel(apiKey)]);
     const codexDir = writeCodexConfig(ctx);
-    const effortConfig = codexEffortConfig[ctx.payload.effort];
+    const effortConfig = getCodexEffortConfig(model)[ctx.payload.effort];
     log.info(`\xBB using model: ${effortConfig.model} (effort: ${ctx.payload.effort})`);
     if (effortConfig.reasoningEffort) {
       log.info(`\xBB using modelReasoningEffort: ${effortConfig.reasoningEffort}`);

package.json

@@ -25,13 +25,13 @@
   },
   "dependencies": {
     "@actions/core": "^1.11.1",
-    "@anthropic-ai/claude-agent-sdk": "0.2.7",
+    "@anthropic-ai/claude-agent-sdk": "0.2.39",
     "@ark/fs": "0.53.0",
     "@ark/util": "0.53.0",
     "@octokit/plugin-throttling": "^11.0.3",
     "@octokit/rest": "^22.0.0",
     "@octokit/webhooks-types": "^7.6.1",
-    "@openai/codex-sdk": "0.80.0",
+    "@openai/codex-sdk": "0.98.0",
     "@opencode-ai/sdk": "^1.0.143",
     "@standard-schema/spec": "1.0.0",
     "@toon-format/toon": "^1.0.0",

pnpm-lock.yaml

@@ -12,8 +12,8 @@ importers:
         specifier: ^1.11.1
         version: 1.11.1
       '@anthropic-ai/claude-agent-sdk':
-        specifier: 0.2.7
+        specifier: 0.2.39
-        version: 0.2.7(zod@4.3.5)
+        version: 0.2.39(zod@4.3.5)
       '@ark/fs':
         specifier: 0.53.0
         version: 0.53.0
@@ -30,8 +30,8 @@ importers:
         specifier: ^7.6.1
         version: 7.6.1
       '@openai/codex-sdk':
-        specifier: 0.80.0
+        specifier: 0.98.0
-        version: 0.80.0
+        version: 0.98.0
       '@opencode-ai/sdk':
         specifier: ^1.0.143
         version: 1.0.143
@@ -111,8 +111,8 @@ packages:
   '@actions/io@1.1.3':
     resolution: {integrity: sha512-wi9JjgKLYS7U/z8PPbco+PvTb/nRWjeoFlJ1Qer83k/3C5PHQi28hiVdeE2kHXmIL99mQFawx8qt/JPjZilJ8Q==}
 
-  '@anthropic-ai/claude-agent-sdk@0.2.7':
+  '@anthropic-ai/claude-agent-sdk@0.2.39':
-    resolution: {integrity: sha512-I1/zcnLah74kZeRkj/1QnDaC6ItJ2m/Bftlm25uoaRkZx7i7SkcpqM9jGE/r2A8PMxnw5WpabP60Xgj99CrTuw==}
+    resolution: {integrity: sha512-wR1TBH62X6E1YwRnWa+A2Eau7AfpTWtfpnwQXO3yRY31FtmzOjPkQb93hbF3AkT0WL7YF9mxBBwJKUa3ZEc5+A==}
     engines: {node: '>=18.0.0'}
     peerDependencies:
       zod: ^4.0.0
@@ -620,8 +620,8 @@ packages:
   '@octokit/webhooks-types@7.6.1':
     resolution: {integrity: sha512-S8u2cJzklBC0FgTwWVLaM8tMrDuDMVE4xiTK4EYXM9GntyvrdbSoxqDQa+Fh57CCNApyIpyeqPhhFEmHPfrXgw==}
 
-  '@openai/codex-sdk@0.80.0':
+  '@openai/codex-sdk@0.98.0':
-    resolution: {integrity: sha512-4+/bZOSjJAPsuM6yceQoVbNUK7UTS03QMKxCrFClObxD1j6n5oh7OSJniyff2MmF0DI7yHnf9omzfFL7RoLWnQ==}
+    resolution: {integrity: sha512-TbPgrBpuSNMJyOXys0HNsh6UoP5VIHu1fVh2KDdACi5XyB0vuPtzBZC+qOsxHz7WXEQPFlomPLyxS6JnE5Okmg==}
     engines: {node: '>=18'}
 
   '@opencode-ai/sdk@1.0.143':
@@ -1664,7 +1664,7 @@ snapshots:
 
   '@actions/io@1.1.3': {}
 
-  '@anthropic-ai/claude-agent-sdk@0.2.7(zod@4.3.5)':
+  '@anthropic-ai/claude-agent-sdk@0.2.39(zod@4.3.5)':
     dependencies:
       zod: 4.3.5
     optionalDependencies:
@@ -2012,7 +2012,7 @@ snapshots:
 
   '@octokit/webhooks-types@7.6.1': {}
 
-  '@openai/codex-sdk@0.80.0': {}
+  '@openai/codex-sdk@0.98.0': {}
 
   '@opencode-ai/sdk@1.0.143': {}
 

post

@@ -41281,13 +41281,13 @@ var package_default = {
   },
   dependencies: {
     "@actions/core": "^1.11.1",
-    "@anthropic-ai/claude-agent-sdk": "0.2.7",
+    "@anthropic-ai/claude-agent-sdk": "0.2.39",
     "@ark/fs": "0.53.0",
     "@ark/util": "0.53.0",
     "@octokit/plugin-throttling": "^11.0.3",
     "@octokit/rest": "^22.0.0",
     "@octokit/webhooks-types": "^7.6.1",
-    "@openai/codex-sdk": "0.80.0",
+    "@openai/codex-sdk": "0.98.0",
     "@opencode-ai/sdk": "^1.0.143",
     "@standard-schema/spec": "1.0.0",
     "@toon-format/toon": "^1.0.0",