Generate and edit Draw.io, Mermaid, and Excalidraw diagrams from natural language using a structured JSON spec.
Coding
consolidate
Try itConsolidate AI PR feedback into classified findings and post an AI Review Summary.
What it does
Runs a seven-topic workflow (pr → collect → internal → classify → decide → post → next) that checks out the PR branch into a worktree, collects CodeRabbit/Copilot AI reviews, falls back to an Internal Code Review when needed, and dual-labels each finding by Type and Severity. After the user decides which findings to act on, it posts an AI Review Summary plus a Formal Review. The `--interactive` flag (or intent keywords in args) pauses before each POST and shows the draft via AskUserQuestion for approval, edit, or reject.
When to use it
- Triaging CodeRabbit or Copilot AI review comments on a PR
- Posting a consolidated AI Review Summary before merge
- Internal Code Review fallback when external AI reviewers are unavailable
- Reviewing and editing draft PR comments before they are published
The skill document
Consolidate
Consolidate and respond to external feedback on PRs and issues.
Options
| Flag | Default | Behavior |
|---|---|---|
--interactive | off | Before each POST (Internal Review at Step 3.5.3 / Summary at Step 7), pause and AskUserQuestion the drafted body to the user. User can approve, request edits, or reject. Edits applied → re-ask. Reject → abort POST for that artifact. |
Auto-activation by args keywords (HARD STOP)
When the consolidate caller passes args containing any of the following intent classes (any language — match by meaning, not literal token), treat --interactive as implicitly set even if the flag was not literal. The keyword expresses the user's intent for review-before-post.
| Intent class | Match signal |
|---|---|
| Explicit review request | Phrases requesting review of the draft before publishing — "review first", "review before post", "review the draft", localized equivalents (e.g., the Korean phrase meaning "review needed" / "after review") |
| Interactive mode request | Words like interactive (any language transliteration), or phrases meaning "conversational mode" / "step-by-step ask" |
| "Important decision ask" intent | Phrases stating that important decisions / important parts must be asked — e.g., "ask important parts", "ask the key decisions", localized equivalents |
| Author-style review request | Phrases stating the user wants to see / review the artifact themselves — "let me review", "let me see the draft", "after I check" |
When auto-activation fires, the caller MUST emit a one-line acknowledgement in chat before the first ask:
Interactive mode auto-activated by caller args (intent: ) — drafts will be reviewed before POST.
| # | Don't | Do |
|---|---|---|
| 1 | Treat args keyword as flavor text and continue deterministic POST | Match against the intent-class table above. On any match, enable --interactive for the entire workflow |
| 2 | Activate interactive only for the matched step (e.g., only the Summary) | Once activated, applies to ALL POST steps in this consolidate run (Internal Review + Summary + any inline edits). One match → all-step gating |
| 3 | Silent activation without chat acknowledgement | Emit the one-liner above so the user can confirm the intent was caught |
| 4 | Map "important decision ask" only to scope-decision axes (Step 5 Axis B / Step 8 next-action) | Body content of Internal Review / Summary is an "important decision" too — every POSTed artifact is subject to review-before-post |
| 5 | Directly running a local/CLI review tool (e.g., CodeRabbit CLI) and skipping the sequential topics of the consolidate skill (collect -> internal -> classify) | Even when running local CLI tools, always route through the collect.md and internal.md steps to ensure that the Internal Code Review comments and AI Review Summary are always posted as a pair |
Interactive flow contract
When --interactive is on (literal or auto-activated), each artifact POST follows this contract:
- Draft — author the body into
.tmp/-draft.md(perfile-operations.md.tmp/policy) - Present — emit a chat summary of the draft (key sections, finding count, verdict line) + the
.tmp/path - Ask —
AskUserQuestionwith options:Approve as-is— POST immediatelyEdit (specify in Other)— capture user edits → apply via Edit → re-present → re-askReject — do not POST— skip POST for this artifact, record skip reason in chat
- POST — only after Approve. Use existing medium decision (deterministic) for the POST mechanics.
Edge cases
- Already POSTed before interactive auto-activation realized: see post.md "Damage control — Summary already posted in wrong medium" — content can be PATCHed in place. After
--interactiveauto-detection mid-flow, the caller asks the user whether to PATCH the already-posted artifacts with reviewed content. - Multiple artifacts in one consolidate run (Internal Review + Summary): ask each separately; do not bundle the two drafts into one ask.
--interactive+--inlinetogether: inline annotation bodies are part of the Internal Review draft; review them in the same ask.
Topics
| Topic | Description | Guide |
|---|---|---|
| classify | Step 4 Analyze and Classify (dual-label Type | Severity, PR scope cross-check) | classify.md |
| collect | Step 3 Collect AI Reviews + Step 3.6 superpowers framework load | collect.md |
| decide | Step 5 User Decision (Axis A/B) + Step 6 Fix or Reject | decide.md |
| internal | Step 3.5 Internal Code Review Fallback (medium: review POST w/ inline when line-specific findings exist, else issue comment) + Step 4.5 UI capture verification | internal.md |
| next | Step 8 Post-Summary Next-Action Ask | next.md |
| post | Step 7 Post Summary + Formal Review (unified vs separate) + Step 7.5 Status + Step 7.6 Deferred registration | post.md |
| pr | Workflow entrypoint — PR identify, skip conditions, Copilot availability pre-check (auto-fallback, no ask), Copilot sequential, worktree checkout, Rules | pr.md |
Topic Dependencies
pr (entry: identify → skip → worktree checkout)
└─→ git-repo/worktree (Step 2.7: check out PR branch into a worktree — all review runs against real files)
collect → internal (code-reviewer dispatch operates in the Step 2.7 worktree)
- Step 2.7 worktree checkout runs on every review (after skip conditions) so the code-reviewer reads real files, CONFLICTING is detected locally, and inline line numbers match HEAD.
Quick Reference
PR Review Workflow
Review CodeRabbit/Copilot feedback on a PR, decide what to act on, and post an AI Review Summary comment.
Entry: pr.md (Workflow index + Step 1, 2, 2.4, 2.5, 2.6, 2.7 + Rules)
Step execution order:
pr.mdStep 1 (Identify PR) + Step 2 (Skip Conditions) + Step 2.4 (Copilot availability pre-check — auto-fallback to Internal Review on unavailable, no ask) + Step 2.5 (Copilot sequential, multi-PR only — skipped on unavailable) + Step 2.6 (re-review trigger policy — first vs re-review) + Step 2.7 (checkout PR branch into a worktree — MANDATORY, all reviews)collect.mdStep 3 (Collect AI Reviews) + Step 3.6 (superpowers framework)internal.mdStep 3.5 (Internal Review Fallback, walkthrough only / on failure — medium decision: inline targets (Critical+Important line-specific;--inline= all) → single review POST (body = findings + comments[] = inline, re-review = new POST); no inline targets → issue comment (re-review = PATCH)) + Step 4.5 (UI capture verification)classify.mdStep 4 (Analyze and Classify)decide.mdStep 5 (Formal Review Decision — Axis B only) + Step 6 (Fix or Reject — only on explicit user instruction)post.mdStep 7 (Post Summary + Formal Review) + Step 7.5 (Status line) + Step 7.6 (Deferred registration)next.mdStep 8 (Post-Summary Next-Action Ask)
Related skills
Stores durable facts in a categorized, plain-markdown vault on disk, alongside your agent's built-in memory.
Join a video meeting as an AI bot with voice, avatar, and screenshare across four operating modes.
Find why your productivity system keeps failing, then apply the smallest fix — capacity math, bottleneck routing, durable local notes.
Run Git operations — commits, branches, merges, rebases, conflict resolution, and recovery — with safety rules enforced.
Fetch raw ad creative, app, ranking, and revenue data from AdMapix as structured JSON.
More from drumrobot
Browse all skillsRun a 4-stage research → plan → review → TDD-implement workflow for code changes with optional PR capture.
Manage Claude Code sessions: lookup, search, repair, migration, compression, and cleanup in one command surface.
Manage Git worktrees, ghq clones, SourceGit registration, and multi-account credentials with safety checks.
Diagnose agent mistakes, harden the prompt that caused them, and finish the original task.
Convert plans and code changes into GitHub issues and PRs, with auth handling, dependency tracking, and pre-publish scans.
Convert functionality into Claude Code automations (Agent, Skill, Rule, Hook, Command)