self-improve
Skill17 starsChanged 7 months agoMITView on GitHub
Recursive self-improvement loop. Reviews and improves the agent's own skills, rules, identity files, hooks, and scripts based on recent usage patterns. Runs nightly via launchd or manually with /self-improve.
39 other projects use a version of this.
Adopted by
Most recently taken up by
- savannahfeder/agentboxtheir own version, first seen Nov 15, 2026
- Hrutin-Suriya-Kotni/until-hiredtheir own version, first seen Nov 15, 2026
- 5uck1ess/devkittheir own version, first seen Nov 15, 2026
- Yeachan-Heo/oh-my-claudecodetheir own version, first seen Nov 9, 2026
- naman10parikh/pennytheir own version, first seen Nov 8, 2026
- DYNO-SOFT/self-improving-looptheir own version, first seen Nov 8, 2026
Used with
Projects with it also use ecosystem-reference, fractal-delegation 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.
naman10parikhSkillecosystem-reference
Quick-access directory of Claude Code ecosystem repos, skills, plugins, MCPs, and patterns. Use when improving the harness.
Also in 5 projects that use this
naman10parikhSkillfractal-delegation
Also in 5 projects that use this
roronoazoroshao369Skillmemory-compression
Also in 5 projects that use this
aaronjmarsSkillaeon-update
Pull framework updates from the upstream Aeon repo into this instance - 3-way merges canon's new commits into a PR, never clobbering operator config.
Also in 4 projects that use this
aaronjmarsSkillauto-workflow
Two-mode aeon.yml workflow builder - analyze inspects URLs and emits a tiered, signal-verified skill-enablement plan plus an aeon.yml diff; enable flips slugs to enabled:true and opens a PR.
Also in 4 projects that use this
- +5 more from naman10parikh
- +2 more from aaronjmars
Other versions39 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 39
- Yeachan-Heo/oh-my-claudecode★ 41ksame name
- Yeachan-Heo/oh-my-claude-sisyphus★ 41ksame name
- zereight/gitlab-mcp★ 2.1ksame name
- aaronjmars/aeon★ 794same name
- aeonfun/aeon★ 793same name
- savannahfeder/agentbox★ 419same name
- forefy/.context★ 154same name
- eai-org/agent-toolkit★ 94same name
- nitinjain999/platform-skills★ 71same name
- aaronjmars/miroshark-aeon★ 20same name
29 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 repository11 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- daily-standupSkill
- distill-sessionSkill
- heartbeatSkill
- humanizerSkill
- memory-updateSkill
- playwright-cliSkill
- promoteSkill
- researchSkill
- search-memorySkill
- setupSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 4 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 33 edited versions we read keep these.
4
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 60
One change per concern: Don't combine unrelated improvements into one edit.
- Distinctive: line 63
Be conservative: If unsure whether a change is an improvement, skip it.
- Distinctive: line 64
Never delete skills or rules: Only add, refine, or consolidate.
- Very specific: line 65
Test hooks after changing them: If you modify .claude/settings.json or a script, verify the hook still fires correctly.
- Distinctive: line 69
After completing the loop, append to the daily log:
“Common” is measured: kept by at least 6 in 10 of the 33 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
Recursive Self-Improvement
Review the agent's own operational infrastructure and make improvements based on evidence from recent usage.
Signal Gathering
Collect evidence before changing anything:
- Daily logs (last 7 days) — look for:
- Tasks that took multiple attempts or unusually long
- Patterns the agent struggled with repeatedly
- Gaps where no skill existed for a common task
- Workflows that felt clunky or required workarounds
- User overrides — look for:
- Places the user corrected the agent's approach
- Redirections ("no, do it this way instead")
- Explicit feedback on what worked or didn't
- Preferences the agent missed or forgot
- Runtime logs — check
.claude/runtime/for:- Hook failures or errors in
launchd/*.log - Watchdog incidents in
incidents/*.log - Stale paths or broken references in scripts
- Hook failures or errors in
Improvement Scope
Review and improve these in order:
1. Skills (.claude/skills/*/SKILL.md)
- Clarity — are instructions unambiguous? Would a fresh session follow them correctly?
- Completeness — do they cover edge cases that came up in logs?
- Gaps — should a new skill exist for a recurring pattern?
- Redundancy — do any skills overlap and cause confusion?
2. Rules (.claude/rules/*.md)
- Relevance — are all rules still applicable to how the agent is actually used?
- Contradictions — do any rules conflict with each other or with skills?
- Missing rules — did the logs reveal patterns that should be codified?
- Specificity — are rules concrete enough to act on, or vague platitudes?
3. Identity files (CLAUDE.md, identity/SOUL.md)
- Alignment — does the described role match how the agent is actually used?
- Completeness — are there capabilities or constraints missing from the description?
- Imports — are all
@references valid and loading correctly?
4. Hooks & scripts (.claude/settings.json, .claude/scripts/*.sh)
- Health — do all hooks reference scripts that exist and are executable?
- Error handling — do scripts fail silently or report errors?
- Stale paths — are there hardcoded paths that may have moved?
Execution Rules
- Auto-apply all changes. The morning standup will surface what changed.
- One change per concern. Don't combine unrelated improvements into one edit.
- Document every change. Append a summary to
daily-logs/YYYY-MM-DD.mdunder## [HH:MM] Self-Improvementwith what was changed and why. - Preserve intent. Improve clarity and coverage, don't change the agent's personality or role.
- Be conservative. If unsure whether a change is an improvement, skip it. False positives erode trust.
- Never delete skills or rules. Only add, refine, or consolidate. Flag candidates for removal in the daily log for the user to decide.
- Test hooks after changing them. If you modify
.claude/settings.jsonor a script, verify the hook still fires correctly.
Output
After completing the loop, append to the daily log:
## [HH:MM] Self-Improvement
### Changes made
- [file]: [what changed] — [why, citing evidence from logs]
### Skipped (needs human input)
- [observation]: [why it wasn't auto-fixed]
### Health check
- Skills: X reviewed, Y improved
- Rules: X reviewed, Y improved
- Hooks: all passing / [issues found]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.

