agentsclimarketplace

Code review apply

Skill darthmolen/ai-plugins-and-skills/skills/code-review-apply

Use when reviewed code-review slices are ready for triage and the developer wants to walk findings, classify them accept/merge/reject/flag, and apply accepted changes to code — third punch-in command of the code review workflow, after code-review-plan-execute completesFrom its SKILL.md

Install
npx -y skills add darthmolen/ai-plugins-and-skills --skill code-review-apply

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

7.3 KB, as published. Nobody here has run it

Code Review Apply

Walk the open findings from reviewed code-review slices, classify each one with technical rigor, and apply the accepted/merged ones to the codebase.

Core principle: Same as superpowers:receiving-code-review — verify before implementing. No performative agreement. Technical rigor always. The slice spec is the audit trail; never delete a reviewer's words, only annotate Status at HEAD.

When to Use

After code-review-plan-execute finishes — slices are in planning/code-reviews/needs-review/reviewed/ with status: reviewed. Or any time you want to triage findings on a single slice that's been reviewed (e.g., one that was queued via the manual code-send-review escape hatch).

Process

Step 1: Find Slices to Apply

Slice spec filenames may be in either of two forms — current (*-code-review-<initiative>-slice-NN-*.md) or legacy (*-code-review-slice-NN-*.md). Both are honored.

ls planning/code-reviews/needs-review/reviewed/*-code-review-*-slice-*.md \
   planning/code-reviews/needs-review/reviewed/*-code-review-slice-*.md 2>/dev/null \
   | sort -u

Filter to those with status: reviewed. (Skip triaging files unless the user is explicitly resuming a stopped session — see Step 2.)

  • No files: "No reviewed slices to apply. Run code-review-plan-execute first if you have queued slices."
  • One file: Confirm with user: "Apply findings from <filename>?"
  • Multiple files: Show list grouped by initiative (frontmatter), ask user: "All slices in order?", "Pick one initiative?", or "Pick one slice?". Initiative grouping makes the choice clearer when slices from multiple efforts are queued together.

Step 2: Resume Check

If any file in reviewed/ already has status: triaging:

  • Alert user: "Slice <filename> is mid-triage (status_updated: <timestamp>). Resume, restart, or skip?"
  • Resume: continue where the open findings list left off (re-read; recompute open).
  • Restart: reset status: reviewed and start over from the first finding.
  • Skip: leave it for now.

Step 3: Mark Triaging

Before walking findings, update the slice's frontmatter:

status: triaging
status_updated: <ISO 8601 now>
status_note: ""

The file stays in reviewed/ while triaging.

Step 4: Walk Findings

Read the ## Code Review section. Iterate by severity (Critical → Important → Minor). For each finding with Status at HEAD: open or Status at HEAD: partially-addressed-at-*:

For each open finding:
  1. PRESENT to user (verbatim from reviewer):
     - ID, severity, file:line
     - Description
     - Suggested fix
     - Reviewer's Status at HEAD annotation (open or partially-...)

  2. VERIFY against current code:
     - Read the file at the cited line
     - Confirm the issue still exists (don't trust Status at HEAD blindly —
       the code may have moved since the reviewer wrote the finding)

  3. ASK the developer to classify:
     - accept   — apply the suggested fix as written
     - merge    — apply with adaptation; user describes the variant
     - reject   — finding is wrong or doesn't apply; capture why
     - flag     — needs more thought; leave open with annotation
     - defer    — valid but punted to a separate ticket / follow-up branch

  4. ACT on the classification:
     - accept / merge → edit the code; verify (compile/lint/test if quick);
       update the finding's "Status at HEAD" line to:
         "Status at HEAD: resolved-at-<new SHA after edit, or 'pending-commit'>"
     - reject         → update the finding's "Status at HEAD" to:
         "Status at HEAD: rejected — <user's reasoning>"
     - flag           → leave "Status at HEAD: open" but append annotation:
         "Status at HEAD: open — flagged: <user's note>"
     - defer          → update to:
         "Status at HEAD: deferred — <ticket/branch ref or note>"

  5. RECORD the decision in-place in the slice file.
     Never delete or rewrite the reviewer's original Description/Suggested fix.
     Only the "Status at HEAD" line gets updated, plus an optional one-line
     "Apply note" appended below it.

Step 5: Forbidden Responses

Same as superpowers:receiving-code-review:

  • No "Great feedback!" or "Excellent points!"
  • No "You're absolutely right!"
  • No implementing before verifying the issue still exists in the code
  • No accepting all findings without per-item evaluation
  • No thanking the reviewer

Step 6: Mark Completed

When every finding in the slice has a non-open Status at HEAD:

  • Move file to completed/:
    mv planning/code-reviews/needs-review/reviewed/<file>.md planning/code-reviews/needs-review/completed/<file>.md
    
  • Update frontmatter:
    status: completed
    status_updated: <ISO 8601 now>
    status_note: ""
    

If some findings remain open (flagged or deferred), leave the file in reviewed/ with status: triaging so you can come back to it. Update status_note to summarize what's left (e.g., "2 flagged findings awaiting design decision on intervention handler").

Step 7: Notify

After processing each slice (or at end of batch), summarize to chat:

  • Slice ID + name
  • Counts: accepted, merged, rejected, flagged, deferred
  • File location
  • If batch: re-run code-synthesize-reviews once finished — open findings have shifted, fresh synthesis tells you what's actually left.

Verifying Edits

For accepted/merged findings, when applying:

  • Read the file at the cited line BEFORE editing — confirm context.
  • Make the smallest edit that resolves the finding.
  • For C# files: respect CODING-STANDARDS.md (StyleCop, CRLF, etc.).
  • For tests / behaviour-changing edits: run the relevant test suite if it's quick.
  • If the edit is risky or non-obvious, pause and discuss with the user before committing.

The slice's head_sha is the snapshot the reviewer reviewed against. Your edits land at HEAD-of-branch (whatever's checked out now). When recording resolved-at-<SHA>, use the SHA AFTER your fix is committed — or pending-commit if you haven't committed yet.

Do Not

  • Auto-accept Critical findings without per-item user confirmation.
  • Mass-apply findings via batch script — each finding is a judgment call.
  • Rewrite the reviewer's text. The slice spec is the audit trail.
  • Move a slice to completed/ with open findings still present.
  • Skip the verification step (Step 4.2) — Status at HEAD can be stale.
  • Treat flag as reject — flagged findings stay visible in synthesis.

Red Flags

ThoughtReality
"Reviewer is obviously right, just apply"Verify the issue exists in the current code first.
"All these Minor findings can be batch-rejected"Each one is a judgment call. Walk them.
"I'll mark completed even though one is flagged"No — flagged findings keep the slice in triaging.
"I can edit the reviewer's description to clarify"No. Append a note; never modify the original.
"I don't have time to verify each one"Then this isn't the right time to apply. Stop.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.