3-verify-pr-fix

Verifies whether a candidate GitHub PR fixes a reproduced issue by replaying the same MCP tool call and diffing outputs.

Updated Aug 29, 2026
One-click install
npx skills add https://github.com/larscolombia/lubot-hunter-mcp-2026 --skill 3-verify-pr-fix-larscolombia
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: 3-verify-pr-fix
Source: https://github.com/larscolombia/lubot-hunter-mcp-2026/tree/main/.agents/skills/3-verify-pr-fix
Command: npx skills add https://github.com/larscolombia/lubot-hunter-mcp-2026 --skill 3-verify-pr-fix-larscolombia

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When reviewing a pull request that claims to fix a bug in the LinkedIn MCP server, maintainers need proof that the fix actually works against the real LinkedIn site, not just a code read. This Skill automates that verification: it checks out the PR, restarts the MCP server on the PR branch, replays the exact tool call that failed on main, and diffs the outputs to produce a clear verdict. ## Core Features & Use Cases - Baseline Replay: Reuses the on-main failure baseline and tool-call metadata captured by /2-repro-issue, so the exact same MCP tool call is re-run on the PR branch. - Safe PR Checkout: Fetches the PR head into a detached HEAD, installs an EXIT trap to restore the original branch, and never edits, merges, or pushes. - Verdict Classification: Classifies the PR as fixes, fixes with concerns, does not fix, cannot run, or cannot verify (skill limitation), with evidence from the output diff. - Scraping Rule Audit: Audits the PR diff against CLAUDE.md rules for locale-independent detection, DOM-class selector avoidance, section-to-URL mapping, and test coverage. - Use Case: A maintainer asks "does PR #498 fix issue #470?" — the Skill checks out the PR, replays the failing get_person_profile call, diffs the JSON responses, and reports a verdict with audit flags. ## Quick Start Ask the assistant to verify PR #498 against the baseline captured earlier by running /2-repro-issue on the linked issue.

Frequently Asked Questions about 3-verify-pr-fix

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify a GitHub PR actually fixes a reported issue?▼

Check out the PR branch, restart the MCP server, and re-run the exact tool call that failed on main using the saved baseline metadata. Then diff the PR-branch output against the on-main baseline to classify the result as fixes, fixes with concerns, or does not fix.

How to test an MCP server tool call over HTTP?▼

Start the server with streamable-http transport, POST an initialize request to /mcp to obtain an Mcp-Session-Id header, send notifications/initialized, then POST a tools/call JSON-RPC request with the tool name and arguments.

What happens if there is no reproduction baseline for the issue?▼

The Skill stops and instructs you to run /2-repro-issue first, since that skill is the canonical source of the failing tool call and on-main response. It never silently re-runs the reproduction itself.

Does this Skill modify or merge the pull request?▼

No. It checks out the PR as a detached HEAD, verifies, and reports only. An EXIT trap restores your original branch and cleans up temp files, and it never pushes, edits, or merges.

Why does a working fix still get a concerns verdict?▼

The verdict is downgraded if the diff uses locale-dependent text matching, LinkedIn CSS class selectors, multi-navigation section mappings, or lacks test coverage, because these violate the project's scraping rules even when the fix works on the target profile.

What are the limitations of this PR verification approach?▼

Bugs that only manifest as crashes, under stdio transport, or in specific environments produce a success baseline, so the diff cannot show before/after. The Skill reports this as a skill limitation and falls back to regression checks plus code audit.