Skip to main content
On this page

adk-review

Review, audit, look-at, sanity-check, check any review-able target. Triggers on: GitHub PR URL (specialized review-pr — most common), . or local path (review-code-changes against the working tree), markdown file or Confluence URL (review-doc), comment-thread URL (review-comments), repo path with —audit (audit-repo), PR URL with —audit (audit-pr). Read-only by default. Produces severity-tiered findings (blocker / critical / should / may / nit) with path:line + ≤15-word evidence quotes from the actual file. Six dimensions in order: correctness → tests → security → performance → readability → consistency. Loads adk-agent-code-reviewer always, adk-agent-security-reviewer when diff touches auth/input/crypto/deps. Under —fix: applies accepted findings locally + pushes after confirm (never force, never merges, never protected branches). Under -i: walks each finding. Under —plan: read-only review-and-recommend; no edits. Refuses single-pass for diffs >5000 LOC.

Source

skills/adk-review/SKILL.md

Frontmatter

YAML
name: adk-reviewdescription: |  Review, audit, look-at, sanity-check, check any review-able target. Triggers on: GitHub PR URL (specialized review-pr — most common), `.` or local path (review-code-changes against the working tree), markdown file or Confluence URL (review-doc), comment-thread URL (review-comments), repo path with --audit (audit-repo), PR URL with --audit (audit-pr). Read-only by default. Produces severity-tiered findings (blocker / critical / should / may / nit) with `path:line` + ≤15-word evidence quotes from the actual file. Six dimensions in order: correctness → tests → security → performance → readability → consistency. Loads adk-agent-code-reviewer always, adk-agent-security-reviewer when diff touches auth/input/crypto/deps. Under --fix: applies accepted findings locally + pushes after confirm (never force, never merges, never protected branches). Under -i: walks each finding. Under --plan: read-only review-and-recommend; no edits. Refuses single-pass for diffs >5000 LOC.allowed-tools: [Read, Edit, Write, Grep, Glob, Bash, WebFetch, Agent]argument-hint: "<target> [-i|--interactive] [--fix] [--audit] [--plan|--act] [--severity blocker|critical|should] [--detailed] [--deep]"metadata:  category: code  kind: task  layer: 1  paths: ["**/*.{ts,tsx,js,jsx,py,go,rs,java,rb,php,cs,kt,swift,c,cpp,h,hpp,sh,sql,yaml,yml,json,toml,md}"]  model: sonnet  effort: medium  user-invocable: true  disable-model-invocation: false  needs_mcp_required: []  needs_mcp_optional: [adk-mcp-github, adk-mcp-atlassian]  needs_meta_info: [workspaces, repos]  forks_emitted: [severity-bar, dimensions, auto-post-policy, confidence-threshold, nit-tolerance, mode, model-depth]

Workflow body

adk-review — review any target (lightweight)

Polymorphic on target. Read-only by default; --fix extends to apply + push (never force, never merge, never to protected branches).

Lightweight by design — no worktree, no embeddings, no SCIP. For a deep PR review with code-context retrieval (worktree + LanceDB embeddings + SCIP + feature-flag tracing), use /adk-pr-review.

--detailed means “gather more local/programmatic context where available” (for this lightweight skill, use the code-index light path more aggressively). --deep means “use the stronger model profile” per shared/model-depth.md. For large PRs, prefer /adk-pr-review; for ambiguous but small reviews, --deep is enough.

Hybrid path — repo-bound (<repo>/.temp/adk/review/<task>/) when reviewing local code (., --audit, a doc in the cwd repo); global ($ADK_DATA_HOME/reviews/<task>/) when reviewing a remote PR URL from outside the relevant repo. Path resolved per shared/paths.md.

References (loaded as needed)

Aspect File
Target dispatch (which sub-flow) references/dispatch.md
Workflow (Phase 0–4) references/workflow.md
Fork IDs references/forks.md
Hard rules + refusals references/rules.md
Sub-flow detail references/<sub-flow>.md — authored on first real use

Cross-skill dependencies

  • Personas: shared/personas/{code-reviewer,security-reviewer,test-engineer}.md
  • Plan/Act mode: shared/plan-act-mode.md
  • Model depth: shared/model-depth.md
  • Code index (light path): shared/guidelines/code-index.md — similar() over each diff hunk to surface adjacent code the diff alone doesn’t show. Heavy PR review with full embedding + reranking is /adk-pr-review, not this skill.
  • Constitution: shared/constitution.md
  • Advisor + question-first: shared/advisor.md, shared/question-first.md, shared/narration.md

References shipped