Agent-Skills.md

Agent Skills: Deep Review: Full-Context Code Review + Fix

Deep-review a PR or branch with issue and repo context, then fix actionable findings. Use for 'review my changes', 'check this PR', or post-implementation quality/spec review requests. SKIP existing human/bot review comments that need replies/thread resolution; use address-review.

UncategorizedID: gopherguides/gopher-ai/review-deep

Repository

gopherguides/gopher-ai
gopherguidesLicense: MIT
172

Install this agent skill to your local

pnpm dlx add-skill https://github.com/gopherguides/gopher-ai/tree/HEAD/plugins/go-workflow/skills/review-deep

Skill Files

Browse the full folder contents for review-deep.

Download Skill

Loading file tree…

plugins/go-workflow/skills/review-deep/SKILL.md

Skill Metadata

Name
review-deep
Description
"Deep-review a PR or branch with issue and repo context, then fix and commit actionable findings; PR-backed runs push new review commits by default. Use for 'review my changes', 'check this PR', or post-implementation quality/spec review requests. SKIP existing human/bot review comments that need replies/thread resolution; use address-review."

Deep Review: Full-Context Code Review + Fix

Plugin Resource Resolution

<PLUGIN_ROOT> is notation. Replace it with a concrete absolute plugin root before every resource read or command:

  • Codex: Start from the directory containing the absolute selected SKILL.md path, then ascend two directories (skills/<name> -> plugin root).
  • Claude Code: Bind it to the injected ${CLAUDE_PLUGIN_ROOT} value.

Before decisions or delegation, read <PLUGIN_ROOT>/lib/driver-interaction.md. Read <PLUGIN_ROOT>/lib/decision-gates.md before resolving scope or delivery. Bind SKILL_ARGS for $go-workflow:review-deep by reading <PLUGIN_ROOT>/lib/skill-arguments.md with this compatibility payload:

<claude-skill-arguments> $ARGUMENTS </claude-skill-arguments>

Step 0: Arguments and Action Contract

Read <PLUGIN_ROOT>/skills/review-deep/arguments.md completely and execute it. It parses PR/issue targets, post and scope hints, and independent FIX_CHANGES, COMMIT_CHANGES, and PUSH_CHANGES controls.

Action matrix:

  • PR-backed default: fix, commit, push, and verify local/remote head equality.
  • Branch-only default: fix and commit locally without push.
  • --no-fix: report only; no review commit or push.
  • --no-commit: leave owned fixes uncommitted and disable auto-push.
  • --no-push: commit locally and report the unchanged remote.
  • --push: push explicitly, but fail when owned fixes remain uncommitted.

A commit, push, or remote-head verification failure is incomplete, never a successful review.

Steps 1-2: Scope and Context

Read <PLUGIN_ROOT>/skills/review-deep/scope-discovery.md completely and execute PR detection, associated-HEAD fallback, base selection, and context routing. API failure is incomplete discovery, not evidence of no PR.

Then read <PLUGIN_ROOT>/skills/review-deep/context-gathering.md completely and gather PR metadata, linked issues, unresolved threads, inline comments, pending reviews, and repository guidelines. For --issue, gather issue-only context; with neither target, perform a diff-only review. Apply the documented context size guard and bot-noise filtering.

Steps 3-4: Coverage Plan and Static Analysis

Read <PLUGIN_ROOT>/skills/review-deep/static-analysis.md completely and execute it. Generate the correct base/uncommitted diff, preserve SCOPE_HINT, run the shared adaptive planner in <PLUGIN_ROOT>/lib/review-planning.md, and complete every planned unit plus the coordinated cross-cutting pass.

For Go changes, collect vet, staticcheck, and race-test output as review evidence; these commands inform findings rather than independently blocking. For visual work, read <PLUGIN_ROOT>/lib/screenshot-evidence.md, initialize the manifest, inspect each capture, and preserve evidence for reporting.

Step 5: Ordered Review Pipeline

Read <PLUGIN_ROOT>/skills/review-deep/review-criteria.md completely. Review every planned unit for correctness, security, performance, maintainability, Go idioms, tests, documentation, and breaking changes.

In order:

  1. Map each issue/PR acceptance criterion to implementation and tests; identify missing requirements and scope creep.
  2. Re-evaluate each unresolved review thread against the current diff.
  3. Incorporate static-analysis and screenshot evidence.
  4. Run the cross-cutting pass, then verify, deduplicate, confidence-score, and rank findings against the checkout.
  5. Read <PLUGIN_ROOT>/skills/review-deep/output-format.md for the exact findings, spec-compliance, review-status, score, and recommendation formats.

Step 6: Fix, Verify, Commit, and Push

Read <PLUGIN_ROOT>/skills/review-deep/fix-and-verify.md completely and execute it end-to-end. Process P0 through P3; skip invalid, pre-existing, or intentional findings and P3 findings below the confidence threshold. Make minimal fixes, add tests for observable changes, and track only review-owned files.

Review and fix all findings in the current context; never delegate based on finding count. Verify the applicable build, tests, and lint. Pass only owned files to review-deep-post-fix.sh and apply commit/push flags independently.

Step 7: Report and Optional PR Post

Read Step 7: Final Summary and Delivery in <PLUGIN_ROOT>/skills/review-deep/output-format.md and execute it using the structured helper result. If the helper failed, do not render a completed review or imply fixes reached the remote.

Post only when --post or the original request explicitly authorizes a PR comment. Otherwise keep the report in the response. This is driver-resolvable; state Decision, Evidence, and Rationale without requesting redundant input.

Completion Contract

A completed review reports all findings, fixes/skips with reasons, verification status, commit/push result, local/remote heads, quality score, and one of APPROVE|REQUEST_CHANGES|COMMENT. Never report success after an incomplete scope lookup, fix verification, commit, push, or remote-head check.