adk-review
Category router for code review work - reviewing a pull request, reviewing local uncommitted changes, addressing review feedback, or capturing a session handoff so the next reviewer or session can continue without information loss. Picks one of adk-review-pr, adk-review-local, adk-review-feedback, adk-review-handoff.
ADK Review (Category Router)
Routes any "look at code that already exists, judge it, capture state" intent to the right review task. Activate one of the listed task skills below; do not review directly from this router.
When to use this category
- A PR or merge request needs a structured review with severity-tiered findings.
- Local uncommitted (or branch) changes need a self-review before pushing or committing.
- Reviewer comments need to be triaged and addressed in code with traceable replies.
- A long-running task needs to be paused or handed off without losing context.
When NOT to use this category
- Implementing new code ->
adk-build - Reviewing documentation only ->
adk-docs-review - Auditing an entire repo or website ->
adk-audit - Drafting commit/PR text from a clean review ->
adk-publish-commit
Shared workflow
Every task in this category honors the same six steps. The task skills tailor the content of each step.
- Confirm intent - identify target (PR URL, branch, file set, comment thread, session); set scope and severity bar.
- Gather - read the diff, related code, related tests, recent history, and prior comments. Use repo evidence over guessing.
- Assess - produce findings ordered by severity (Blocker / Critical / Should Have / May Have / Nitpick / Question). Separate verified issues from open questions.
- Resolve - either author replies / fixes (review-feedback) or capture the snapshot (handoff). PR/local review just reports.
- Validate - if changes were made, run repo-native checks; if findings were posted, verify they are visible.
- Report - findings-first summary; explicit "not validated" markers; recommended next step.
Task selection
| Task skill | Use when | Do NOT use when |
|---|---|---|
adk-review-pr | A specific PR URL or remote MR is the target; produce findings for posting back | The change is local and not yet pushed - use review-local |
adk-review-local | Uncommitted changes, staged changes, or current branch vs. base needs a self-review before push | The change is already on a remote PR - use review-pr |
adk-review-feedback | Existing review comments must be triaged and addressed in code with traceable replies | The reviewer comments are not yet posted - just iterate locally |
adk-review-handoff | A session must be paused, switched, or handed off; capture state, decisions, blockers, git status | Writing project documentation - use docs-write |
Severity ladder
All review-style outputs use the same labels:
| Label | Meaning |
|---|---|
Blocker | Must fix before merge / before continuing. Bug, security hole, broken contract. |
Critical | Strongly recommended fix; would normally block release. |
Should Have | Improvement that meaningfully raises quality; deferring needs justification. |
May Have | Optional polish; no harm in skipping. |
Nitpick | Style or taste; never block on this alone. |
Question | Reviewer is unsure; needs clarification, not necessarily a fix. |
Lead with the highest severity. Never bury a Blocker under nitpicks.
Activation
Once you have picked a task, load adk-review-<task> and follow it. Each task skill is fully standalone - everything it needs is in its own folder.
Anti-patterns
- Reviewing inside this router. Always hand off to a task skill.
- Mixing severity levels in one bullet ("nit/blocker"). Pick one.
- Posting findings without verifying the diff context.
- Marking something "fixed" in feedback without running validation.
Clarifying questions (default-ask)
When running without --auto, the skill asks these questions in order, one at a time. Under --auto, the skill picks the safest option for each (see references/clarifying-questions.md) and reports the choices.
- Where do the changes live: a remote PR, local working tree, or already-pushed branch? — How to pick: Remote PR → adk-review-pr. Local uncommitted → adk-review-local. Already-pushed but no PR yet → adk-review-local first, then publish.
- Are there existing reviewer comments to address? — How to pick: Yes → adk-review-feedback. No → adk-review-pr or adk-review-local.
- Is the goal a session handoff (capture state for the next agent/human)? — How to pick: Yes → adk-review-handoff.
Default report: Routed task + why.
Detailed report (on request or --verbose): Decision table covering all four review modes.
Artifact: review-routing-decision — Inline message.
Artifact path: (none)
Clarifying questions (default-ask)
When running without --auto, the skill asks these questions in order, one at a time. Under --auto, the skill picks the safest option for each (see references/clarifying-questions.md) and reports the choices.
- Where do the changes live: a remote PR, local working tree, or already-pushed branch? — How to pick: Remote PR → adk-review-pr. Local uncommitted → adk-review-local. Already-pushed but no PR yet → adk-review-local first, then publish.
- Are there existing reviewer comments to address? — How to pick: Yes → adk-review-feedback. No → adk-review-pr or adk-review-local.
- Is the goal a session handoff (capture state for the next agent/human)? — How to pick: Yes → adk-review-handoff.
Default vs detailed output
Default report: Routed task + why.
Detailed report (on request or --verbose): Decision table covering all four review modes.
Artifact: review-routing-decision — Inline message.
Artifact path: (none)
<!-- adk:references:start -->References shipped with this skill
These files live in references/ next to this SKILL.md. Read them when the skill activates; they are inlined here so the skill is fully self-contained (no cross-skill or shared sources).
| File | Purpose |
|---|---|
references/anti-patterns.md | Things to avoid when running this skill. |
references/artifact-format.md | The deliverable's format and where it lives (.temp/ contract). |
references/clarifying-questions.md | The default-ask questions for this skill, with how-to-pick rubrics. |
references/constitution.md | Non-negotiable rules and working/communication discipline. |
references/interaction-contract.md | Default-ask, explained-options, --auto contract every skill must follow. |
references/output-format.md | Default vs detailed report shapes; severity labels; verbosity rules. |
references/persona.md | The agent persona that drives this skill. |