agentleFS
Sign inStart free
YehudaFrankelYehudaFrankel/clankbrain/.claude/skills/evolve/SKILL.md

evolve

Skill16 starsChanged 7 months agoMITView on GitHub

Reviews accumulated lessons.md entries, identifies repeated patterns, and proposes or creates new reusable skills. Turns hard-won session knowledge into formalized, reusable workflow. Triggers on /evolve or \"evolve lessons into skills\" or \"cluster patterns into skills\".

41 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use instinct-export, instinct-import 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.
  • TheKREATORSSkill

    instinct-export

    Export instincts from project/global scope to a file

    Also in 4 projects that use this

  • TheKREATORSSkill

    instinct-import

    Import instincts from file or URL into project/global scope

    Also in 4 projects that use this

  • AWSSkill

    map

    Map verified service capabilities against customer-provided control framework and threat model. Produces mapping-results.json, controls-catalog.md, and framework-mapping.md. Supports pluggable frameworks: CCMv4, NIST 800-53, CIS Benchmarks, ISO 27001, etc. Split into 4 focused sub-skills for better accuracy; this skill orchestrates them.

    Also in 5 projects that use this

  • diego-tobalinaSkill

    build-fix

    Fix build and compilation errors systematically. Run when build fails.

    Also in 3 projects that use this

  • launch52-aiSkill

    skill-create

    Create and validate Claude Code skills with proper structure, checklists, and validation scripts. Use when creating a new skill, auditing existing skills, or ensuring skills meet quality standards. Generates complete skill scaffolds with SKILL.md, check scripts, guides, and templates.

    Also in 3 projects that use this

  • qtzx06Skill

    eval

    Evaluate trained YOLO model accuracy with mAP metrics and per-class error analysis. Identifies failure cases for targeted improvement. Use after training.

    Also in 5 projects that use this

  • +1 more from TheKREATORS

The file

Show its rules and text

We split this skill into the rules it gives an agent

22 rules. 0 are common, 15 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 36 edited versions we read keep these.

  • 15

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Allowed Tools: Read, Edit, Write, Glob, Bash

    line 14
  2. Distinctive:

    Before making any changes, show a summary of what would happen:

    line 21
  3. Distinctive:

    Ask: "Apply these changes?

    line 24
  4. Distinctive:

    Only proceed on explicit yes.

    line 25
  5. Distinctive:

    Look for entries with the same root cause or "Apply when" condition

    line 32
  6. Distinctive:

    Create it? (yes/no)

    line 44
  7. Very specific:

    Run this phase every time /evolve is called, after Phase 2.

    line 57
  8. Very specific:

    Read tasks/skill_scores.md.

    line 60
  9. Very specific:

    Read the skill's SKILL.md at .claude/skills/[name]/SKILL.md

    line 72
  10. Distinctive:

    Read ALL "What Failed" entries for this skill from skill_scores.md

    line 73
  11. Distinctive:

    Quality gate: Check each entry contains a step number, what was produced, and what was needed.

    line 74
  12. Distinctive:

    Do not guess at what failed.

    line 79
  13. Distinctive:

    Add an improvement note at the bottom of the changed step:

    line 82
  14. Very specific:

    Update Skill Patched column in skill_scores.md for all patched rows: change - to today's date (e.g.

    line 86
  15. Very specific:

    Update the skill_scores.md row: change Skill Patched from - to [date] (e.g.

    line 96
  16. Distinctive:

    When /evolve detects that a new decision supersedes, explains, or exemplifies an existing one, flag it:

    line 106
  17. Very specific:

    Add the appropriate Relates to: field when writing to decisions.md.

    line 111
  18. Distinctive:

    Allowed Tools: Read, Edit, Write, Glob, Grep, Bash

    line 123
  19. Distinctive:

    Only create skills for patterns with clear, repeatable trigger conditions

    line 138
  20. Very specific:

    Run /learn before /evolve to ensure latest session patterns are captured

    line 140
  21. Distinctive:

    After creating skills, update MEMORY.md index if your project uses one

    line 141
  22. Distinctive:

    Default to preview mode — never patch skills without explicit confirmation

    line 142

“Common” is measured: kept by at least 6 in 10 of the 36 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Skill: evolve

Trigger: /evolve or "evolve lessons into skills" or "cluster patterns into skills"

Description: Reviews accumulated lessons.md entries, identifies repeated patterns, and proposes or creates new reusable skills. Turns hard-won session knowledge into formalized, reusable workflow.

Allowed Tools: Read, Edit, Write, Glob, Bash

Steps

  1. Preview mode (default): Before making any changes, show a summary of what would happen:
    • Skills that would be patched (name, step number, what would change)
    • New skills that would be created (pattern, frequency) Ask: "Apply these changes? [y/N]" Only proceed on explicit yes.
  2. Read memory files:
    • .claude/memory/lessons.md
    • .claude/memory/decisions.md (if exists)
  3. Identify repeated patterns:
    • Look for entries with the same root cause or "Apply when" condition
    • Flag patterns that appear 2+ times across sessions
  4. Check for existing skill coverage:
    • List all skills in .claude/skills/
    • For each repeated pattern, check if an existing skill already covers it

  5. For uncovered repeated patterns, propose:

    Pattern: [description] Appeared: N times Suggested skill name: [name] Create it? (yes/no)

  6. On confirmation, create the skill:
    • Create .claude/skills/[name]/SKILL.md
    • Include: trigger phrase, description, allowed tools, step-by-step instructions, notes
    • Make steps specific to your actual stack and patterns
  7. Report: "Created N new skills: [list]"

Phase 3: Compound Learning Loop — Self-Improving Skills

Run this phase every time /evolve is called, after Phase 2.

Step A — Find failing skills

Read tasks/skill_scores.md. Find all rows where:

  • Correction Needed = Y
  • Skill Patched = - (not yet patched by /evolve)

Why Skill Patched not Code Fixed: Code Fixed records how the immediate code issue was resolved at the time (manual/-). Skill Patched records whether the SKILL.md itself was updated to prevent recurrence. A row can have Code Fixed = manual and Skill Patched = - — meaning the code was fixed in the session, but the skill hasn't been improved yet. This column is the gate.

Group by skill name. Only proceed with skills that have 2 or more Y entries with Skill Patched = -. Single failures are noise. Patterns are signal.

Log skipped skills: "Skipped [skill] — only 1 failure logged. Need 2+ to patch."

Step B — For each qualifying skill

  1. Read the skill's SKILL.md at .claude/skills/[name]/SKILL.md
  2. Read ALL "What Failed" entries for this skill from skill_scores.md
  3. Quality gate: Check each entry contains a step number, what was produced, and what was needed. If any entry is vague (no step number, no specific output):
    • Skip this skill entirely
    • Log: "Skipped [skill] — insufficient failure data in row [date]. Re-run /learn with structured format before patching."
    • Do not guess at what failed.
  4. Identify the step number cited most often across the Y entries
  5. Rewrite that specific step to prevent the failure — be concrete, not generic

  6. Add an improvement note at the bottom of the changed step:

    [YYYY-MM-DD] Improved: [one line — what changed and why]

  7. Update Skill Patched column in skill_scores.md for all patched rows: change - to today's date (e.g. 2026-04-06)
  8. Log to tasks/skill_improvements.md: | [date] | [skill] | Step [N] | [what failed before] | [what was fixed] | skill_scores.md [dates] |
  9. Save all updated files

Step C — Log the improvement

Append to tasks/skill_improvements.md:

| [date] | [skill] | Step N | [what failed before] | [what was fixed] | skill_scores row [date] |

Update the skill_scores.md row: change Skill Patched from - to [date] (e.g. 2026-04-06). This is what allows /evolve-check to classify the skill as 🔵 PATCHED on the next run.

Step D — Report

Compound Learning: patched N skills

  • [skill name]: Step [N] — [one line summary of what changed]

Relationship Types for Decisions

When /evolve detects that a new decision supersedes, explains, or exemplifies an existing one, flag it:

  • supersedes — new decision replaces an old one
  • explains — new decision clarifies why an old one was made
  • example-of — new decision is a concrete instance of a general pattern

Add the appropriate Relates to: field when writing to decisions.md.

Skill File Format

Skill: [name]

Trigger: [exact phrase(s) that invoke this skill]

Description: [one sentence — what it does and when to use it]

Allowed Tools: Read, Edit, Write, Glob, Grep, Bash

Steps

  1. ...
  2. ...

Notes

  • ...

Notes

  • Only create skills for patterns with clear, repeatable trigger conditions
  • Generic advice ("always test") doesn't make a good skill — specific workflows do
  • Run /learn before /evolve to ensure latest session patterns are captured
  • After creating skills, update MEMORY.md index if your project uses one
  • Default to preview mode — never patch skills without explicit confirmation

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free