10x-impl-review-ci

Reviews pull requests against their implementation plan in GitHub Actions CI.

Updated Jul 4, 2026
One-click install
npx skills add https://github.com/Assamir/ai-toolkit --skill 10x-impl-review-ci-assamir
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: 10x-impl-review-ci
Source: https://github.com/Assamir/ai-toolkit/tree/main/.cursor/skills/10x-impl-review-ci
Command: npx skills add https://github.com/Assamir/ai-toolkit --skill 10x-impl-review-ci-assamir

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Plan-driven development loses value when nobody verifies that a merged PR actually matches its plan. This Skill runs an automated, non-interactive implementation review inside GitHub Actions, comparing the PR diff against the plan it claims to implement and publishing an auditable verdict. ## Core Features & Use Cases - Plan drift detection: Discovers the plan at context/changes/<change-id>/plan.md (repo root or monorepo subtree), cross-references changed files against planned files, and flags MATCH / DRIFT / MISSING / EXTRA per planned change. - Parallel evidence gathering: Spawns three subagents covering plan drift, safety/quality/pattern compliance, and test coverage, then runs the plan's non-test automated checks (lint, build, typecheck). - Auditable output contract: Writes a committed report at <change-dir>/reviews/impl-review.md with a seven-dimension verdict table, posts inline review comments on flagged diff lines via MCP, and posts a summary PR comment with a REJECTED gate signal. - Use Case: A team using a plan-then-implement workflow labels a PR with impl-review; the workflow reviews the diff against the plan, commits the report with [skip ci], and fails the check on a REJECTED verdict unless an impl-review-override label is present. ## Quick Start Ask your AI coding assistant to run the 10x-impl-review-ci skill on this pull request to review the implementation against its plan and post the verdict.

Frequently Asked Questions about 10x-impl-review-ci

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

FAQPage Schema
How do I run automated code review on a pull request in GitHub Actions?

Trigger the skill from a pull_request workflow using claude-code-action with the PR number and base ref as environment variables. It discovers the plan, reviews the diff, commits a report to the branch, and posts inline plus summary comments.

How does the skill find the plan file for a PR?

It searches the PR diff for files matching context/changes/<change-id>/plan.md at the repo root or any monorepo workspace subtree, picking the newest. A `Plan: <path>` line in the PR body overrides the convention.

What happens when a PR has no plan file?

The skill posts a neutral comment explaining that no plan was detected and exits with code 0, so the workflow does not fail. This keeps docs, chore, and hotfix PRs from breaking CI.

Can the review verdict block a pull request from merging?

Yes. A REJECTED verdict makes the workflow's verdict-check step fail and posts a failing commit status. Adding the `impl-review-override` label to the PR bypasses the gate after reviewers examine the findings.

Why does the review commit include [skip ci] in the message?

The [skip ci] marker prevents the bot's own push from retriggering the same workflow, which would cause an infinite recursion loop. The skill verifies the marker exists before pushing and aborts loudly if it is missing.

Does the CI review skill modify source code in the PR?

No. The skill is strictly read-analyze-report: it never edits source files, asks questions, or runs a triage loop. It only writes the review report file, updates change.md status, commits, and posts comments.