skill-improver
Skill318 starsChanged 14 days agoMITView on GitHub
Captures lessons from completed tasks and turns them into concrete improvements. Triggers proactively after high-ceremony tasks (build failures, multi-hour sessions, production incidents) and on-demand when the user asks for "post-mortem", "reflect", "improve the skill", "lessons learned", "what went wrong". Covers skills, repo artifacts (scripts, CI, configs), and process changes. For small/obvious fixes, edits directly. For structural changes, produces a plan using the manual-planning format. Use this whenever a task surfaced gaps in tooling, documentation, or automation.
19 other projects use a version of this.
Adopted by
Most recently taken up by
- svg153/skillstheir own version, first seen Nov 1, 2026
- davidanthonyn/claude-skillstheir own version, first seen Oct 20, 2026
- air-gapped/skillstheir own version, first seen Oct 19, 2026
- juandelossantos/another-agent-skillstheir own version, first seen Oct 17, 2026
- AlvaroVFon/skillstheir own version, first seen Oct 13, 2026
- ilamanov/skillstheir own version, first seen Oct 13, 2026
Other versions19 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 19
- Galaxy-Dawn/claude-scholar★ 5.9ksame name
- Heyvhuang/ship-faster★ 337same name
- ribeirogab/zeno-agent★ 9same name
- air-gapped/skills★ 6same name
- exiao/meta-skills★ 4same name
- AlvaroVFon/skills★ 2same name
- karozi/awesome-product-management-skills★ 2same name
- LiozShor/claude-code-skills★ 2same name
- prPMDev/my-portfolio-kit★ 2same name
- BenSchZA/skills★ 1same name
9 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 repository49 other files it gives its agents.
- .agents/skills/unsafe-checker/AGENTS.mdAGENTS.md
- benchmarks/CLAUDE.mdCLAUDE.md
- CLAUDE.mdCLAUDE.md
- e2e/CLAUDE.mdCLAUDE.md
- src/CLAUDE.mdCLAUDE.md
- src-tauri/CLAUDE.mdCLAUDE.md
- website/CLAUDE.mdCLAUDE.md
- website/llms-full.txtllms.txt
- website/llms.txtllms.txt
- ci-gate-diagnosisSkill
- coding-guidelinesSkill
- domain-webSkill
- exploration-modeSkill
- flathub-maintenanceSkill
- m01-ownershipSkill
- m02-resourceSkill
- m03-mutabilitySkill
- m04-zero-costSkill
- m05-type-drivenSkill
- m06-error-handlingSkill
- m07-concurrencySkill
- m09-domainSkill
- m10-performanceSkill
- m11-ecosystemSkill
- m12-lifecycleSkill
- m13-domain-errorSkill
- m14-mental-modelSkill
- m15-anti-patternSkill
- manual-planningSkill
- runbooksSkill
- rust-call-graphSkill
- rust-code-navigatorSkill
- rust-dailySkill
- rust-deps-visualizerSkill
- rust-learnerSkill
- rust-refactor-helperSkill
- rust-skill-creatorSkill
- rust-symbol-analyzerSkill
- rust-trait-explorerSkill
- security-stanceSkill
- seo-auditSkill
- seo-performance-reviewSkill
- solidjsSkill
- sync-lockfilesSkill
- tauri-agent-devSkill
- tauri-v2Skill
- tiptapSkill
- todo-managerSkill
- unsafe-checkerSkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 10 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 19 edited versions we read keep these.
10
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 35
If the session didn't use any skill, focus on repo/process improvements only.
- Distinctive: line 83
The user can always override the plan/execute decision: If the user says "just fix it directly" for something that would normally need a …
- Distinctive: line 93
Create a plan using the manual-planning skill's format.
- Distinctive: line 101
After applying fixes (directly or via plan execution):
- Distinctive: line 102
Run format/lint commands if the changed files are in checked languages
- Very specific: line 133
If a plan was created, add: Plan: docs/plan-name.md — awaiting approval.
- Distinctive: line 137
Don't over-improve: If the session went smoothly and no concrete gaps were found, say so and stop.
- Distinctive: line 138
One session, one reflection: Don't chain multiple reflection cycles.
- Distinctive: line 140
Respect the user's override: If the user says "just fix it" for something structural, apply directly.
- Distinctive: line 144
Use manual-planning for any change that needs a plan (structural changes)
- Distinctive: line 145
Use todo-manager if the reflection surfaces new TODO items for the backlog
“Common” is measured: kept by at least 6 in 10 of the 19 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
Skill Improver
Turn every hard-earned lesson into a reusable improvement. After completing a non-trivial task (especially one that hit friction), run this skill to catalog what happened, classify the gaps, and apply fixes — either directly for obvious issues, or via a structured plan for larger changes.
When to Run
Proactive (run without being asked):
- A build, deploy, or CI job failed and was debugged
- A task took more than 3 rounds of trial-and-error
- A session lasted more than one hour and involved multiple skill uses
- The user explicitly expressed frustration or surprise at a gap
On-demand (run when the user says):
- "Reflect on this session" / "post-mortem" / "what went wrong"
- "Improve the skill" / "update the skill with these learnings"
- "Lessons learned" / "what should we change"
Scope: Reflect on ALL skills used during the session, plus any repo artifacts and process steps involved. If multiple skills were loaded, review each one. If the session didn't use any skill, focus on repo/process improvements only.
Workflow
Step 1 — Reflect
Catalog observations in three buckets, with concrete examples:
What went well:
- Patterns, tools, or decisions that saved time or prevented errors
- Skill instructions that proved particularly useful
- Diagnoses or fixes that were reached quickly
What went wrong:
- Mistakes, dead ends, or incorrect assumptions
- Skill instructions that were missing, wrong, or misleading
- Manual workarounds that had to be invented on the fly
What was missing:
- Scripts, checks, or tooling that would have caught the issue earlier
- Documentation or guardrails that didn't exist
- Process steps that were skipped or not automated
Step 2 — Classify
Group findings into three categories. For each finding, identify the concrete artifact that needs to change (file path, skill section, CI workflow, etc.).
Skill changes — improvements to an existing skill:
- Missing failure entries in a diagnosis table
- Wrong fix instructions
- Undocumented pitfalls or edge cases
- Better log access or debugging workflows
Repo artifact changes — new or modified files in the repository:
- Utility scripts (e.g.,
check-node-sources.mjs) - CI workflow steps or pre-release checks
- Configuration or lockfile validation
Process changes — workflow or convention changes:
- Release checklist items
- Pre-commit or pre-tag validation steps
- Cross-skill integration (e.g., pre-release skill should validate lockfiles)
Step 3 — Apply
If no improvements are identified: state that explicitly and stop. Do not invent changes.
The user can always override the plan/execute decision. If the user says "just fix it directly" for something that would normally need a plan, apply the changes directly instead. If the user says "make a plan" for something small, create one.
For obvious fixes (typos, missing warnings, incomplete instructions): Apply directly to the target file using the Edit tool. These are changes where:
- The fix is a single addition or correction to an existing section
- There's no design decision to make
- The change can't break anything else
- Example: adding a missing failure entry to a diagnosis table, or a warning about a known pitfall
For structural changes (new sections, new scripts, reorganized workflows): Create a plan using the manual-planning skill's format. These are changes where:
- The fix requires creating a new file or reorganizing existing content
- Multiple files are affected
- The user should review the approach before implementation
- Example: adding a new utility script, restructuring a skill's diagnosis workflow, adding pre-release steps
Step 4 — Verify
After applying fixes (directly or via plan execution):
- Run format/lint commands if the changed files are in checked languages
- For scripts, run a syntax check and a quick smoke test
- Re-read the changed skill sections to ensure they don't contradict other parts of the skill or adjacent skills
- Summarize what was changed and why
Output Format
At the end, present using this template:
## Session Reflection
### What Went Well
- [item]
### What Went Wrong
- [item]
### What Was Missing
- [item]
## Changes Made
| File | Change |
|------|--------|
| path/to/file | what changed and why |
<!-- If no changes were made: -->
No improvements identified.If a plan was created, add: **Plan:** docs/plan-name.md — awaiting approval.
Gotchas
- Don't over-improve. If the session went smoothly and no concrete gaps were found, say so and stop. Inventing minor tweaks dilutes the skill.
- One session, one reflection. Don't chain multiple reflection cycles. If the user wants another, they'll ask.
- Obvious fix ≠ trivial preference. "I'd word this differently" is a preference, not a fix. Only apply direct edits when the current text is objectively wrong or missing.
- Respect the user's override. If the user says "just fix it" for something structural, apply directly. If they say "make a plan" for something small, create one. Their call always wins.
Integration with Other Skills
- Use manual-planning for any change that needs a plan (structural changes)
- Use todo-manager if the reflection surfaces new TODO items for the backlog
- The output format mirrors what you'd see in a good commit message: what, why, and where
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.

