kevinslin/agstack/dev-review
Claude Code and agent skill by kevinslin/agstack
- community
Checking that the SKILL.md file is still in the repository.
Review code, docs, specs, architecture, UX, or design docs.
Install
npx vskill@latest install kevinslin/agstack/dev-reviewSKILL.md
dev.review
Workflow
- Identify the review type from the user's request and artifact.
- Examples: code, docs, design-doc, spec, simplify-spec, architecture, ux, skills, integrator, deslop, dead-code.
- Use
simplify-specwhen the user asks to radically simplify a spec or its proposed implementation, narrow the solution, or find an 80/20 approach. - Use
specfor correctness, approved-scope completeness, ownership, safety, and implementation readiness; leave design minimization tosimplify-spec. - If ambiguous, ask one clarifying question before reviewing.
- Load the matching workflow from
./references/workflow-[review-type].md.
- If the workflow file does not exist, ask the user for the prompt to add and pause the review.
- For skills reviews, use
$scfor the local skill-authoring contract. - For docs reviews, use
./references/workflow-docs.md; when reviewing OpenClaw docs and$openclaw-docsis available, apply its guidance as domain-specific context. - For
integrator, default input artifacts are outputs fromag-learnand adjacent retrospectives. - For code reviews that require flow docs, use
$specy.
- Route every top-level review request through the
trigger:loopshortcut fromdev.shortcuts.
- If the user request already contains
trigger:..., resolve it throughdev.shortcuts. - If no shortcut trigger is present, invoke
trigger:loopwith the resolved review instruction, for exampletrigger:loop review the current diff with $dev.review. - Give the loop reviewer the review type, workflow file, artifact paths, current diff, and review scope.
- Routing a plain review request through
trigger:loopdoes not grant edit
authorization. In that case, require the reviewer pass and parent classification, but stop after reporting any blocker or major findings; do not start a fixer pass without user authorization.
- An explicit
trigger:loopinvocation, explicit review-and-fix request, or
edit task authorizes the shortcut's scoped fixer passes. Follow the full reviewer/classifier/fixer loop for those requests.
- When this skill is already running inside a
trigger:loopreviewer pass, apply the workflow directly to the material and produce the review instead of nesting another loop. - When this skill is already running as a read-only reviewer subagent for
trigger:spec, apply the selected workflow directly; do not redispatch the parent shortcut or start a nested review/fixer loop.
- For PR or CI-backed review loops, keep going until the remote exit condition is met.
- Completion is remote-state based, not patch based: current head SHA is known, relevant CI is green, unresolved non-outdated review threads are zero, and actionable comments are addressed or explicitly routed to the user.
- Before saying the loop is finished, run a final PR gate query and report head SHA, failing/pending check count, unresolved thread count, and actionable comment count.
- If any required check is failed/pending or any actionable review item remains, the loop is not finished; continue fixing or report the exact blocker.
Output
- Lead with findings ordered by severity (blocker/major/minor) or by impact if severity is unclear.
- Prefer concrete, actionable feedback over generic commentary.
- Call out assumptions, risks, and unclear ownership/abstractions.
- Propose simplifications when the selected workflow owns simplification.
- Keep the review concise; avoid restating large sections of the input.
- For code reviews, include both the
Simplicity AuditandTest Audit
checklists required by ./references/workflow-code.md; never check an item without inspected evidence or report a clean review with either audit incomplete.
- Give every added or modified test a
keep,merge, ordeletedisposition.
Treat tests without distinct production-behavior evidence as major findings; remove or consolidate them when edits are authorized, or report them without changing files during a review-only request.
- Treat materially avoidable duplicate ownership, parallel implementations,
and implementation-coupled test machinery as major when a concrete smaller design preserves approved capabilities, security, and supported compatibility.
- If blocker or major findings remain in a review-only request, report them as
unresolved and state that no fixer pass ran because edits were not authorized.
Workflows
./references/workflow-code.mdfor code review../references/workflow-docs.mdfor developer documentation, user guides, API references, CLI references, quickstarts, READMEs, and troubleshooting docs../references/workflow-design-doc.mdfor design doc review../references/workflow-spec.mdfor product, implementation, or test spec review../references/workflow-simplify-spec.mdfor reducing a spec to the smallest complete implementation that satisfies its current goal../references/workflow-architecture.mdfor architecture and system-boundary review../references/workflow-ux.mdfor UX review../references/workflow-skills.mdfor reviewingSKILL.mdfiles and bundled skill resources../references/workflow-integrator.mdfor integrating learnings into skill/code/project changes../references/workflow-deslop.mdfor anti-slop code review focused on excess complexity, patch size, and unnecessary helper extraction../references/workflow-dead-code.mdfor dead-code review that accounts for every new class, function, method, variable, constant, option, config field, and other named artifact introduced by a PR.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingactive/dev.review/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/kevinslin/agstack/dev-review)