self-critique
Skill2 starsChanged 4 days agoMITView on GitHub
Honest retrospective with a per-dimension score on the result, skills, agents, and workflow used. Rate work from 1 to 5 across dimensions (correctness, completeness, robustness, simplicity, verification, process, tools) — what could be done better, which assumptions should have been questioned, which alternatives existed — with concrete proposals for improvement. Use after quality gate according to hub settings.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- jaskirat1616/grok-skillstheir own version, first seen Nov 1, 2026
- goolars/Claude-Code-Stoptheir own version, first seen Oct 25, 2026
- jfkconstruct/claude-skillstheir own version, first seen Oct 11, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- jaskirat1616/grok-skills★ 22same name
- goolars/Claude-Code-Stop★ 4same name
- jfkconstruct/claude-skills★ 0same name
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 repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- acceptance-criteriaSkill
- decision-recordSkill
- hub-managerSkill
- memorySkill
- orchestratorSkill
- quality-gateSkill
- report-pdfSkill
- source-verificationSkill
- structured-handoffSkill
- task-intakeSkill
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 0 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
0
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 14
After the quality gate of any non-trivial task (setting "at completion", default) or every phase ("at every phase").
- Very specific: line 19
If self-critique is disabled in Settings (quality.selfCritique: off).
“Common” is measured: kept by at least 6 in 10 of the 3 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
Self-Critique and Improvement
Critically dissects both the outcome and how it was achieved, so the next iteration is better. A focused, actionable review.
When to use
- After the quality gate of any non-trivial task (setting "at completion", default) or every phase ("at every phase"). Reported by
route.mjs. - Whenever a gate failed, attempts were repeated, or user corrections were required.
When not to use
- If self-critique is disabled in Settings (
quality.selfCritique: off). - On small tasks with clean outcomes: the final check suffices.
Point-by-point evaluation
Rate each dimension from 1 to 5, stating the rationale and what is needed to reach a 5. Ground scores in evidence (tests, checks, sources); when in doubt, choose the lower score.
| Dimension | Question |
| Correctness | Does it do what it should, without errors? |
| Completeness | Does it satisfy all acceptance criteria? |
| Robustness | Does it handle edge cases, errors, Mac and Windows? |
| Simplicity | Is it the minimal, readable solution? |
| Verification | How thoroughly is it proven (tests, commands)? |
| Process | Was execution efficient and well-ordered? |
| Tooling | Did used skills, agents, and workflows help? |
Scale: 5 excellent · 4 good · 3 sufficient · 2 weak · 1 inadequate. Any score under 4 must produce at least one weakness; proposals target lowest scores. No overall average: an average hides weaknesses.
Procedure
- Outcome: what is weak, incomplete, or merely acceptable? Actively seek defects rather than confirmation.
- Assumptions: which assumptions went unverified? Which, if false, alters the result?
- Alternatives: was there a simpler, more robust approach? Why was it not chosen?
- Process: where was time or context wasted (redundant reads, superfluous agents, retries)? Where was deeper analysis needed?
- Tooling: for each skill, agent, or workflow, what was missing, ambiguous, or misleading?
- Proposals: at most 3 actionable improvements, prioritizing lowest scores (e.g. "add check X to quality-gate").
- Application, per
quality.improvementssetting:- propose (default): present proposals to user;
- apply to own skills: edit only
skills/own/,agents/own/, orworkflows/, run doctor and tests; never touch external items.
- A single experience is a hypothesis for improvement, not a universal rule: clarify this when proposing.
- Record (if enabled):
node "<hub>/scripts/evaluate.mjs" record '<json>'withrequest,size,workflow,skills,agents,gate,retries,routing, andscoresacross the seven dimensions (correctness,completeness,robustness,simplicity,verification,process,tools).
Expected output
Evaluation:
Correctness 4/5 — <rationale> → for 5: <what is needed>
Completeness …/5 · Robustness …/5 · Simplicity …/5 · Verification …/5 · Process …/5 · Tooling …/5
Weaknesses of deliverable: …
Assumptions to verify: …
Better alternatives: …
Process improvements: …
Skills/agents/workflow feedback: <name> — <issue>
Proposals (max 3): 1. … — where — how to verify
Applied: … | Proposed to user: …Attention
- Self-critique does not reopen user decisions without new facts.
- Honesty first: if the outcome has limitations, state them clearly even if the gate passed.
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.

