review
Skill1 starsChanged 5 months agoMITView on GitHub
Multi-lens project review. Default tier runs 12 Sonnet lenses on the codebase in parallel. Optional tier adds UI / browser / breaking-change. Invoke with /review for a full pass, or "/review security" for a single lens.
192 other projects use a version of this.
Adopted by
Most recently taken up by
- dzhng/skillstheir own version, first seen Nov 15, 2026
- muxy-app/muxytheir own version, first seen Nov 15, 2026
- macro-inc/macrotheir own version, first seen Nov 15, 2026
- yagizdo/quivertheir own version, first seen Nov 15, 2026
- markmhendrickson/neotomatheir own version, first seen Nov 15, 2026
- felipeam86/pipe-skillstheir own version, first seen Nov 15, 2026
Used with
Projects with it also use close-spec, design-with-images and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
dzhngSkillclose-spec
When a spec under specs/<feature>/ is done shipping (a write-spec build, or any planned task that produced a spec), archive it to specs/done/ and rewrite it from a build-plan into a durable rationale — the why, the principles, the invariants — pointing back to the real code for the how. Use when implementation has landed and the plan no longer matches what shipped, or the user says a feature/spec is finished. Pairs with [write-spec](../write-spec/SKILL.md) (the plan this closes) and [review](../review/SKILL.md) (run before closing).
Also in 5 projects that use this
dzhngSkilldesign-with-images
Explore UI and visual design with image generation, then iterate the real implementation against the user-selected concept. Use when brainstorming visual directions, offering mockup variants, or implementing an approved generated design; do not stop at the mockups or the first code approximation.
Also in 5 projects that use this
LambenthanSkilldaily-arxiv
每日从 arXiv 拉取新论文,过滤相关性,auto-ingest 高优先级论文,检测 SOTA 更新
Also in 5 projects that use this
LambenthanSkillexp-eval
实验判决门:Review LLM 独立评判实验结果 → 4 种判决路径 → 自动更新 claims confidence、ideas status、graph edges
Also in 5 projects that use this
- +3 more from dzhng
- +5 more from Lambenthan
Other versions192 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 192
- garrytan/gstack★ 142ksame name
- ansible/ansible★ 71ksame name
- docling-project/docling★ 71ksame name
- traefik/traefik★ 66ksame name
- ClickHouse/ClickHouse★ 51ksame name
- temporalio/temporal★ 24ksame name
- ccch1mneyyy/dsh-Go★ 24ksame name
- tisfeng/Easydict★ 15ksame name
- juicedata/juicefs★ 15ksame name
- elie222/inbox-zero★ 13ksame name
182 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository7 other files it gives its agents.
- build-planSkill
- executeSkill
- github-syncSkill
- initSkill
- iterateSkill
- saveSkill
- skill-forgeSkill
The file
We split this skill into the rules it gives an agent
8 rules. 0 are common, 2 are distinctive, and 6 are very specific to this project.
0
Common
Most of the 157 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 47
If ambiguous → AskUserQuestion: "Full pass (default 12 lenses), one lens, or full + optional tier (adds UI / browser / breaking-change)?"
- Distinctive: line 49
Read the matching reference:
- Very specific: line 53
All lenses share patterns documented in references/review-lens-framework.md.
- Very specific: line 58
Create artifacts/reviews/ if it doesn't exist.
- Very specific: line 59
Check fresh-findings cache (per framework Phase A): for each lens about to run, check db_age_hours '<lens>' 'findings' 'lens'.
- Distinctive: line 80
No hard concurrency limit (Sonnet subagents are unbounded).
- Very specific: line 93
Read findings from DB. Build a brand-new file artifacts/reviews/review-synthesis-<N>.md where N is the next integer (never overwrite a …
- Very specific: line 188
Before completing, read and follow ../../references/cross-cutting-rules.md.
“Common” is measured: kept by at least 6 in 10 of the 157 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Review
YOUR ROLE: ORCHESTRATOR ONLY. You dispatch Sonnet subagents; you do NOT read code and produce review findings inline. The lenses derive their value from being independent Sonnet workers running in parallel — if you do the review in the main thread, all of that collapses: lenses lose independence, context blows up, and the synthesis loses its multi-perspective signal.
The ONLY work the main thread does is: Phase 1.5 SAST pre-scan (run scanners directly), Phase 3 dispatch and wait, Phase 5 synthesis from returned findings. If you catch yourself reading source files for the review itself in Phase 3, stop and dispatch the Agent.
Multi-lens code + plan review. Sonnet subagents run review lenses in parallel; the main thread aggregates findings into a synthesis at artifacts/reviews/review-synthesis-N.md.
Two paths share the orchestration:
- Full — 12 default-tier lenses in parallel; user can opt in to 3 optional-tier lenses
- Single lens — one named lens (
/review security,/review tests, etc.)
Inputs
| Input | Source | Required |
project-context.md | Project root | Yes |
project-plan.md | Project root | Recommended (for drift / completeness lenses) |
features.md | Project root | Recommended |
| The full codebase | Project root | Yes |
Outputs
- Per-lens findings:
db_upsert '<lens>' 'findings' 'lens' "$CONTENT" - Synthesis:
artifacts/reviews/review-synthesis-N.md— never overwrites a prior synthesis; N is the next sequential integer
Instructions
Phase 0: Detect mode
| Trigger | Mode |
/review (no args) | full default-tier |
/review full or "full review" | full default-tier + ask about optional tier |
/review <lens-name> | single lens (e.g., /review security, /review tests) |
/review --include browser style | not supported (no flags) — user picks via AskUserQuestion |
| "review the auth code" or area-scoped | single lens that fits OR ask via AskUserQuestion which lens |
If ambiguous → AskUserQuestion: "Full pass (default 12 lenses), one lens, or full + optional tier (adds UI / browser / breaking-change)?"
Read the matching reference:
- Full or full+optional →
references/default-tier.md(andoptional-tier.mdif applicable) - Single lens →
references/default-tier.mdorreferences/optional-tier.md(whichever lists the named lens)
All lenses share patterns documented in references/review-lens-framework.md.
Phase 1: Verify inputs
- Confirm
project-context.mdexists. If missing, stop and tell the user to run/initfirst. - Create
artifacts/reviews/if it doesn't exist. - Check fresh-findings cache (per framework Phase A): for each lens about to run, check
db_age_hours '<lens>' 'findings' 'lens'. If <24h and unchanged, offer to reuse.
Phase 1.5: SAST pre-scan
Read references/sast-prescan.md for the full protocol. Run local linters and scanners in parallel — gitleaks (secrets), the available language linters (biome / oxlint for JS/TS, ruff for Python, cargo clippy for Rust), plus the heavier cross-language scanners semgrep (semantic SAST) and trivy (dependency CVEs / IaC / secrets). Assemble $SAST_SUMMARY (≤5000 chars; keep CRITICAL/HIGH + all gitleaks/trivy-secret findings, drop INFO/LOW).
This summary is injected into every lens prompt at Phase 3 so Sonnet reviewers cross-reference real findings, not just impressions. Skip cleanly if any tool is unavailable — lens reviews still run regardless.
Phase 3: Fan out lenses (Sonnet subagents)
STOP — this is the dispatch phase. You spawn subagents; you do NOT review code yourself.
For each lens in the chosen scope:
- Spawn
Agentwithsubagent_type: "general-purpose". - Pass the lens prompt assembled from:
- The lens spec (the lens's section in
references/default-tier.mdorreferences/optional-tier.md) - Deep references (when the spec lists them): paths under
references/lenses/<lens>/*.md— pass these paths to the subagent and instruct it to read them on demand agents/review-lens.mdtemplate (filled with lens name, scope, finding format)$SAST_SUMMARYfrom Phase 1.5 (machine-verified findings to cross-reference)- Project root + paths to inputs
- The lens spec (the lens's section in
- No hard concurrency limit (Sonnet subagents are unbounded). Single-lens mode spawns one.
- Subagents return findings as response text. The MAIN thread writes them to DB:
Subagents do NOT have access todb_upsert '<lens>' 'findings' 'lens' "$AGENT_RESPONSE"references/db.sh. Per cross-cutting anti-pattern A2.
Phase 4: Wait for completion
Confirm all subagents returned. If any failed (timeout, crash, empty), note in synthesis but don't block.
Phase 5: Synthesis
Read findings from DB. Build a brand-new file artifacts/reviews/review-synthesis-<N>.md where N is the next integer (never overwrite a prior synthesis):
N=$(ls artifacts/reviews/review-synthesis-*.md 2>/dev/null | sed -E 's/.*-([0-9]+)\.md/\1/' | sort -n | tail -1)
NEXT=$(( ${N:-0} + 1 ))Structure:
# Review Synthesis <NEXT>
Generated: <date>
Project: <name>
Scope: <full default | full + optional | single lens: X>
## Summary
- Total findings: N (after dedup)
- By severity: CRITICAL: X, HIGH: Y, MEDIUM: Z, LOW: W
- SAST findings included: N (gitleaks/linters)
- Lenses run: <list>
- Lenses skipped (cached fresh): <list>
- Lenses failed: <list with reason>
## SAST Findings
[Pulled from Phase 1.5 $SAST_SUMMARY. Machine-verified secret scans + linter output.]
## Cross-Lens Patterns
[Patterns spanning >1 lens — highest priority]
## HIGH Confidence Findings
[Flagged by multiple lenses OR by SAST + a lens]
## MEDIUM Confidence Findings
[Flagged by one lens with strong evidence]
## Notable LOW Confidence Findings
[Single-lens findings at CRITICAL/HIGH severity — may be false positives but too important to drop]
## Per-Lens Summary
[One-paragraph per lens: finding count + headline issue]
## Recommendations
[Prioritized action list: fix first, fix later, ignore.]Phase 6: Present + handoff
Show the user:
- Total finding count + severity distribution
- Top 3 cross-lens patterns
- Lenses that found zero issues (suspicious — usually means prompt was too narrow or model "all-clear" hallucination)
Offer:
"Review complete. {N} findings (CRITICAL: X, HIGH: Y). Run
/execute fixto apply the prioritized fixes, or/saveto snapshot and continue."
References (on-demand)
references/review-lens-framework.md— shared output pattern, fresh-findings check, severity tiersreferences/default-tier.md— 12 default lenses with per-lens checks + finding categoriesreferences/optional-tier.md— 3 optional lenses (UI, browser, breaking-change)
Agent prompts (on-demand)
agents/review-lens.md— Sonnet subagent prompt template (filled per lens)
Examples
User: /review
→ Phase 0: full default-tier. Phase 1.5 runs gitleaks + linters.
Phase 3 spawns 12 Sonnet subagents in parallel. Phase 4 waits. Phase 5
writes review-synthesis-3.md. Phase 6 presents top items.User: /review security
→ Single lens mode. Phase 3 spawns one Sonnet subagent for security-review.
Synthesis still uses the standard template — just one lens populated.User: full review + check the UI too
→ Full + optional UI. Phase 3 spawns 13 subagents (12 default + ui).User: review the auth code
→ Phase 0 ambiguous, AskUserQuestion → user picks security + tests.
Two-lens fan-out scoped to auth/ directory.Before completing, read and follow ../../references/cross-cutting-rules.md.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

