plan-review
Skill1 starsChanged 4 months agoMITView on GitHub
Review a saved plan with convergence-oriented discipline — verdict first, blockers capped, "I don't understand" converted to questions instead of findings. Use when asked to "review the plan", "sanity-check the plan", or for a fresh-eyes second opinion on a saved plan file before executing it; reads prior review entries so repeated rounds converge instead of re-litigating. Designed to avoid the fresh-eyes nit-list pattern where every review produces 6-8 items regardless of plan maturity.
47 other projects use a version of this.
Adopted by
Most recently taken up by
- antick/skillstheir own version, first seen Nov 16, 2026
- pantheon-org/tekhnetheir own version, first seen Nov 15, 2026
- ihorpar/agent-skillstheir own version, first seen Nov 15, 2026
- iuliandita/skillstheir own version, first seen Nov 15, 2026
- duncanmorris/agent-skillstheir own version, first seen Nov 13, 2026
- bas1l/claude-skillstheir own version, first seen Nov 11, 2026
Used with
Projects with it also use plan-execute, pr-comment 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.
longranger2Skillplan-execute
Use when the user says "/plan-execute", "plan execute", "implement plan", or "execute plan" and provides a finalized plan file path to carry out. Claude orchestrates, Codex writes code, Claude reviews, and Codex fixes issues until the quality bar is met.
Also in 3 projects that use this
BenjaminGSkillpr-comment
This skill should be used when reviewing someone else's pull request and you want drafted review comments posted to GitHub — either gate / gate-wf findings (tier-marked) or pr-challenge's reviewer questions (unmarked) — each drafted through humanizer and posted as a standalone inline comment at its file:line, after one confirmation. Triggers on "post these as review comments", "comment the gate findings on the PR", "draft and post my review", "leave review comments", after a gate / gate-wf run or a pr-challenge pass on a PR you're reviewing.
Also in 3 projects that use this
MicrosoftSkillspec-review
Independent, multi-model review of a design or spec **before** any code is written, for the microsoft/winappcli repo. Activate when a contributor asks to "review this spec", "review my design", "review this design doc", "validate this approach", "should we build this", "spec review", "design review", or "feature review". Fans out parallel sub-agents — each doing its OWN research against the real codebase and ecosystem rather than trusting the spec — covering necessity & scope, approach & alternatives, feasibility vs reality, risks & unknowns, DX & user impact, and a different-model-family cross-check. Emits a decision-oriented recommendation (proceed / proceed-with-changes / reconsider) to stdout. This is the PRE-CODE companion to the pr-review skill (which reviews code already written); use spec-review at the design/spec stage, not on an implemented diff. Does NOT write code or edit the spec.
Also in 3 projects that use this
AnthropicSkilloffice-hours
Set up and run Claude Code office hours, including the calendar event when the champion's own calendar connector allows it. Use when a champion wants a regular time for teammates to bring problems.
Also in 3 projects that use this
AWSSkillretro
Conduct a retrospective on a coding session.
Also in 5 projects that use this
BlockSkillplan
Draft a TDD, PRD, or lightweight plan in the conversation without committing it.
Also in 5 projects that use this
Other versions47 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 47
- ahmadrosid/nakama★ 826same name
- igerber/diff-diff★ 413same name
- Mathews-Tom/armory★ 340same name
- sd0xdev/sd0x-dev-flow★ 192same name
- sd0xdev/sd0x-harness★ 192same name
- YougLin-dev/Aha-Loop★ 181same name
- longranger2/claude-gpt-workflow★ 76same name
- ConnorGriffin/skills★ 21same name
- liewcf/agent-skills★ 16same name
- aws-solutions/konductor★ 12same name
37 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 repository21 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- ai-commentsSkill
- ai-draftSkill
- ai-nextSkill
- ai-postSkill
- call-traceSkill
- draft-prSkill
- dto-mappingSkill
- handoffSkill
- hum-commentsSkill
- jira-dcSkill
- meeting-summarySkill
- plan-from-issueSkill
- plan-nextSkill
- plan-saveSkill
- pr-commentsSkill
- pr-statsSkill
- review-sessionSkill
- skill-builderSkill
- validator-generatorSkill
- templateSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
16 rules. 0 are common, 12 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 40 edited versions we read keep these.
12
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 44
Check prior reviews: If a ## Reviews section exists, read every entry.
- Distinctive: line 45
Avoid repeating yourself: If a prior review already raised a concern and it was resolved (or explicitly scoped out), do not raise it again.
- Distinctive: line 52
Stop: Do not offer to make edits to the plan body.
- Very specific: line 72
After producing the review output above, append a short paragraph to the plan file under a ## Reviews heading at the bottom.
- Very specific: line 83
Use today's date from the conversation context (check the system reminder for currentDate).
- Distinctive: line 85
Keep each entry short on purpose.
- Very specific: line 86
Do not copy the full verdict/blockers/questions block into the Reviews section.
- Distinctive: line 92
Only flag things visible from the plan text itself:
- Distinctive: line 105
Anything the Reviews section shows was already raised: If prior review #2 asked about X and the plan wasn't changed, the user decided X is …
- Distinctive: line 114
If you can't confidently complete the Blocker sentence, it's a Question.
- Distinctive: line 128
Reviews 2–4: prefer converting gap concerns to Questions rather than Blockers.
- Distinctive: line 129
Reviews 5+: default to not raising gap concerns at all.
- Distinctive: line 151
If blockers is "none" and questions is "none" — verdict is execute.
- Distinctive: line 155
No preamble ("I've reviewed the plan...").
- Distinctive: line 156
No summary of what the plan does.
- Distinctive: line 158
No closing offer to make edits.
“Common” is measured: kept by at least 6 in 10 of the 40 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
Plan Review: Convergence-oriented fresh-eyes review
You are reviewing a plan that was written in a prior session. You do NOT have the session context the author had while writing it. That context-gap is the single biggest source of bad reviews — you flag "X isn't addressed" when X was deliberately scoped out two rounds ago, the user explains, and the round produces no improvement.
Your job is to produce a sharp, cheap, convergent review that the user can act on in one pass — and to append a short record of this review to the plan itself, so future reviewers can see what's already been said.
When to use this skill
- Gating a saved plan before
plan-nextstarts executing it — typically once, in a fresh session. - "Review the plan" / "sanity-check this plan" / a second opinion after major plan revisions.
When NOT to use this skill
- Reviewing code, diffs, or PRs — this reviews plan files; use a code-review flow for code.
- An unsaved, in-chat draft — persist it first (
plan-save); the review appends its record to the file. - Expecting fixes — this skill reports a verdict; editing the plan stays with the author.
Steps
- Find the plan file. If an argument was provided, use that path. Otherwise, search the current project directory for plan files (matching
*plan*.md,*PLAN*.md, or markdown inplans/). If exactly one is found and it's obviously the active plan, use it. If multiple, ask the user via AskUserQuestion, sorted by last-modified. - Read the entire plan file, including any
## Reviewssection at the bottom. - Check prior reviews. If a
## Reviewssection exists, read every entry. Two purposes:- Avoid repeating yourself. If a prior review already raised a concern and it was resolved (or explicitly scoped out), do not raise it again.
- Calibrate strictness. Count the entries. Apply the scaling in Rules → Round-N discipline.
- Produce the review in the output format below. Nothing else — no preamble, no recap of the plan, no "here's what I read."
- Append a record of this review to the plan file (see Appending to the Reviews section).
- Stop. Do not offer to make edits to the plan body. Do not continue into execution. The review and its record are the deliverables.
Output format
Exactly this structure, in this order:
**Verdict:** execute | don't execute
**Blockers:** [max 2, or "none"]
- [sharp, specific, cites plan section]
**Questions:** [max 2, or "none"]
- [genuine "I don't understand" items, phrased as questions]That is the entire output to the user. No additional sections. No polish list. No "minor suggestions." If you have nothing for a section, write "none" and move on.
Appending to the Reviews section
After producing the review output above, append a short paragraph to the plan file under a ## Reviews heading at the bottom. If the section doesn't exist, create it.
Format per entry:
### Review N — YYYY-MM-DD
[One short paragraph, 2-4 sentences. State the verdict. If blockers or
questions were raised, summarize them in one clause each. If none, say so.]- Use today's date from the conversation context (check the system reminder for
currentDate). - Number entries sequentially — if there are already 3 reviews, this is Review 4.
- Keep each entry short on purpose. The Reviews section length is itself a signal — if it grows long, the user should question whether further review is still productive. Fat paragraphs defeat that signal.
- Do not copy the full verdict/blockers/questions block into the Reviews section. Summarize.
Rules
What to flag
Only flag things visible from the plan text itself:
- Internal contradictions — phase 3 says X, phase 5 assumes NOT-X.
- Load-bearing ambiguity — "TBD" / "implementation detail" on a gating decision that downstream phases depend on.
- Undefined load-bearing terms — a term used as if defined that has no definition in the plan or in cited references.
- Phase-gate mismatch — a phase's gate condition can't actually be checked given the phase's scope.
What NOT to flag
- "X isn't addressed." If X isn't in the plan, it was almost certainly scoped out deliberately. Convert this to a Question ("Is X deliberately out of scope?") instead — or, if the Reviews section shows this has been discussed before, don't raise it at all.
- Polish, wording, "consider adding," section-reordering. These are never blockers and don't belong in the output.
- Hypothetical edge cases not in the current execution path.
- "You might want to verify..." — if it's important enough to verify, it's either a blocker or a question. If it's neither, cut it.
- Anything the Reviews section shows was already raised. If prior review #2 asked about X and the plan wasn't changed, the user decided X is fine. Don't re-raise.
Blocker vs. Question
Your default tempted-to-flag item is probably a Question, not a Blocker. Test:
- Blocker: "Executing this plan as written will produce a wrong / broken / incomplete result because ______." You can complete that sentence confidently from the plan text alone.
- Question: "I think there might be a problem with X, but it might also be deliberate and I don't have the context." → Question.
If you can't confidently complete the Blocker sentence, it's a Question. If the Question appears in a prior review, it's nothing — drop it.
Round-N discipline
First, separate your concern into one of two classes:
- Text-visible concern. Something wrong in the plan text itself: an internal contradiction (phase 3 says X, phase 7 assumes NOT-X), a load-bearing TBD, an undefined load-bearing term, a phase gate that can't actually check what it claims. These are caught by reading, not by guessing at coverage.
- Gap concern. "X isn't addressed" / "I'd expect to see Y" / "what about Z?" You're inferring that something should be in the plan and isn't. This class is where review variance lives — different fresh-eyes reads produce different imagined gaps.
Round-N discipline applies only to gap concerns. Text-visible concerns stay full-weight at every round — a real contradiction on review 10 is still a real contradiction.
Count entries in the ## Reviews section and apply to gap concerns:
- Reviews 0–1: normal weight. Gap concerns are fair game; fresh-eyes is most valuable here.
- Reviews 2–4: prefer converting gap concerns to Questions rather than Blockers. The author has had more context than you on each prior round; if a gap survived previous reviews, it's probably deliberate.
- Reviews 5+: default to not raising gap concerns at all. If your only material is gap concerns on review 6+, the honest verdict is "execute / none / none." Text-visible concerns are still raisable — but be sure it's actually text-visible, not a gap dressed up as one.
When in doubt, it's a Question
The hardest call in this skill is "is this a real issue or am I navel-gazing?" You won't always classify correctly. That's fine — the output format has a built-in safety valve:
- Questions are cheap. One sentence to answer, no execution halt, no re-plan.
- Blockers are expensive. They stop execution and demand fixes.
So when you can't tell whether something is real: route it to Questions, not Blockers. Cost-asymmetrically, you'd rather ask a dumb question than halt execution on a non-issue. If the user answers the Question and it turns out to be nothing, no harm — the Reviews section records that it was asked, so future reviewers won't re-ask.
Cap discipline
Max 2 blockers. Max 2 questions. Hard cap. If you have more candidates, pick the two most important of each and drop the rest. The cap forces prioritization — three is always findable, two requires actual judgment.
Verdict discipline
The verdict is committal. If you'd execute the plan as-is, say "execute" and mean it. Don't say "execute" and then list 4 blockers — pick one or the other.
- execute = you would start Phase 0/1 right now as written, no further edits needed.
- don't execute = at least one blocker needs resolving first.
If blockers is "none" and questions is "none" — verdict is execute. Period.
Format discipline
- No preamble ("I've reviewed the plan...").
- No summary of what the plan does.
- No "overall, the plan is solid" framing.
- No closing offer to make edits.
- Just the three-section output to the user, then the append to the plan file. Then stop.
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.

