agentleFS
lifemixturelifemixture/ai-orchestration-repo/.claude/skills/feature-code-review/SKILL.md

feature-code-review

Skill0 starsChanged 3 months agoMITView on GitHub

Review an implemented multi-repo feature as a whole — the contract agreement between repositories, security across the new seams, fidelity to the plan, coherence and operability. Deliberately does NOT redo the per-phase review the build already gated. Report only, dated, under docs/features/<slug>/code-reviews/.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read PRD.md and IMPLEMENTATION_PLAN.md, plus the latest report in plan-reviews/ if one exists — an unresolved planning finding that …

    line 26
  2. Distinctive:

    Verify the checklist against reality: before trusting it: does each repository actually have that branch, and commits on it?

    line 37
  3. Distinctive:

    Run this first and on its own.

    line 47

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

You are reviewing an implemented feature across every repository it touched. Feature folder: $ARGUMENTS

This skill is report-only: it changes no code, and it never writes into projects/. It produces a dated report that a re-run of /feature-build can act on.

You are not the phase reviewer. During the build, an independent reviewer gated every phase against its acceptance criteria, in its own repository. Redoing that is waste dressed as thoroughness. Your job is the altitude that review could not reach: what only becomes visible when all the repositories are on the table at once.

The context rule applies here too. You do not read diffs. You hand each subagent the command and let it read in its own context. A whole-feature diff across four repositories would fill this conversation and take the report down with it.

Step 1 — Establish scope

Read PRD.md and IMPLEMENTATION_PLAN.md, plus the latest report in plan-reviews/ if one exists — an unresolved planning finding that survived into the code is worth more than a fresh one. Read repos.yaml.

From the plan's status checklist, take the branch name and the repositories touched. For each, the review range is:

git -C projects/<svc> diff <manifest branch>..feature/<slug>

Verify the checklist against reality before trusting it: does each repository actually have that branch, and commits on it? A checklist row claiming work that is not in any commit is itself a blocker-grade finding, and everything downstream of it is fiction.

Step 2 — Do the repositories agree?

Spawn integration-checker over the repositories on both sides of every contract the feature changed, with the contract and the plan's deploy order.

Run this first and on its own. If the two sides do not agree, the rest of the review is commentary on something that does not work.

Step 3 — The four lenses, in parallel

Spawn one feature-reviewer per lens — security, fidelity, coherence, operability — giving each the feature folder, every repository with its diff command, and its lens. They are independent and run concurrently.

Merge the results. Where two lenses report the same thing, keep the sharper one.

Step 4 — Write the report

Get today's date (date +%F). Write docs/features/<slug>/code-reviews/<YYYY-MM-DD>-code-review.md:

  • Verdict — is this ready for a human to push? Blockers?
  • Contract agreement — the integration-checker result, before the findings. Everything else is conditional on it.
  • Findings, grouped by severity, each with: id, severity (blocker | high | medium | low), category (contract | security | fidelity | coherence | operability | checklist), repository, location (file:line), finding, recommendation.
  • Map refresh needed — list the seam files whose services this feature changed. They are stale by definition now: .claude/rules/services/<svc>.md still describes the system as it was before this branch. Name the command: /system-map --only <svc,...>. Left undone, the next feature plans against a map this one invalidated.
  • Plan drift — where PRD.md or IMPLEMENTATION_PLAN.md no longer describe what exists.

Output

Print the report path and a concise summary: the contract verdict, counts by severity, the blockers, and the services needing a map refresh. Modify no code.

Suggest addressing blockers via /feature-build, then refreshing the map, and only then handing the branches to a human to push in the plan's stated order.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.