Colin McDonnell
b8ac42e875
* mcp: embed example calls in top-level tool descriptions
agents (esp. claude sonnet) hallucinate param names from training-data
priors — `pr_number` instead of `pull_number`, `summary` instead of
`body`, full subcommand strings jammed into `git({command})` like it
were `shell({command})`. each error burns a tool round-trip plus a
follow-up ToolSearch, ~40+ events / 24h, no observable recovery cost
to us but visible to users in agent logs.
cheapest fix: add a sample formatted function call to every affected
tool's top-level description. example anchors are more reliable than
schema descriptions alone because the model treats descriptions as
narrative but call examples as canonical structure. for `git` and
`shell` (whose `command` fields collide), include explicit
counter-examples disambiguating which tool owns which shape.
no schema aliases / coercion yet — try the cheap thing first; if the
next audit window still shows the same hallucination rate, layer
aliases on top per #585's recommendation.
closes #585, closes #701
* mcp: drop negative anchors from tool descriptions
negation is a footgun in tool descriptions — telling the model "NOT
pr_number" makes pr_number more salient, not less. let the positive
example carry the schema and trust the model to read it.
removes:
- "the parameter is pull_number (a number), NOT pr_number" and
similar across checkout_pr, get_pull_request, list_pull_request_reviews,
get_review_comments, create_pull_request_review
- "NOT summary, message, or content" on report_progress
- "WRONG: git({ command: 'log --oneline' })" counter-example on git
- redundant param-type restatements after the example (e.g. "depth is a
number, not a string" on git_fetch, "description is required" on shell)
keeps a single positive example per tool. for tools with multiple call
shapes (git, git_fetch, push_branch), two positive examples instead of
one + a counter-example.
64 lines
1.9 KiB
TypeScript
64 lines
1.9 KiB
TypeScript
import { type } from "arktype";
|
|
import type { ToolContext } from "./server.ts";
|
|
import { execute, tool } from "./shared.ts";
|
|
|
|
export const IssueInfo = type({
|
|
issue_number: type.number.describe("The issue number to fetch"),
|
|
});
|
|
|
|
export function IssueInfoTool(ctx: ToolContext) {
|
|
return tool({
|
|
name: "get_issue",
|
|
description:
|
|
"Retrieve GitHub issue information by issue number. " +
|
|
"Example: `get_issue({ issue_number: 1234 })`.",
|
|
parameters: IssueInfo,
|
|
execute: execute(async ({ issue_number }) => {
|
|
const issue = await ctx.octokit.rest.issues.get({
|
|
owner: ctx.repo.owner,
|
|
repo: ctx.repo.name,
|
|
issue_number,
|
|
});
|
|
|
|
const data = issue.data;
|
|
|
|
// set issue context
|
|
ctx.toolState.issueNumber = issue_number;
|
|
|
|
const hints: string[] = [];
|
|
if (data.comments > 0) {
|
|
hints.push("use get_issue_comments to retrieve all comments for this issue");
|
|
}
|
|
hints.push(
|
|
"use get_issue_events to retrieve cross-references and commit references (relationships not reflected in current state)"
|
|
);
|
|
|
|
return {
|
|
number: data.number,
|
|
url: data.html_url,
|
|
title: data.title,
|
|
body: data.body,
|
|
state: data.state,
|
|
locked: data.locked,
|
|
labels: data.labels?.map((label) => (typeof label === "string" ? label : label.name)),
|
|
assignees: data.assignees?.map((assignee) => assignee.login),
|
|
user: data.user?.login,
|
|
created_at: data.created_at,
|
|
updated_at: data.updated_at,
|
|
closed_at: data.closed_at,
|
|
comments: data.comments,
|
|
milestone: data.milestone?.title,
|
|
pull_request: data.pull_request
|
|
? {
|
|
url: data.pull_request.url,
|
|
html_url: data.pull_request.html_url,
|
|
diff_url: data.pull_request.diff_url,
|
|
patch_url: data.pull_request.patch_url,
|
|
}
|
|
: null,
|
|
hints,
|
|
};
|
|
}),
|
|
});
|
|
}
|