agentleFS
MegapromptingMegaprompting/torque-loop/skills/evolve/SKILL.md

evolve

Skill10 starsChanged 4 months agoMITView on GitHub

Evolve one artifact through a bounded, evidence-gated mutation loop — lock target, snapshot baseline, apply one pressure, generate candidate deltas, judge, patch, verify, keep or revert, record state, and name the next edge. Not brainstorming and not a general 'make this better': every kept change must be proven. Use to harden a specific code file, prompt, skill, test suite, README, spec, or workflow along a chosen pressure vector.

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 8 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

  • 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

  • MontimageSkill

    test-coverage

    Improve test coverage by finding untested branches, error paths, and edge cases. Use when measuring or strengthening a project's tests. Don't use for production debugging, code review, or feature implementation.

    Also in 3 projects that use this

  • +1 more from TheKREATORS

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 36 edited versions we read keep these.

  • 4

    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
  1. Distinctive:

    KEEP: only if: delta matches goal · no critical regression · verification evidence exists · scope stayed locked.

    line 105
  2. Very specific:

    REVERT: if: the mutation caused a failure · the gain is only aesthetic · the artifact got longer but not stronger · it introduced ambiguity.

    line 107
  3. Very specific:

    ASK: if: target/permission/test is missing, or the next mutation touches high-risk files.

    line 109
  4. Distinctive:

    Write the evolution event (whenever project files changed, or at each iteration):

    line 112
  5. Distinctive:

    Stop when: max iterations reached · a mutation fails and cannot be safely patched · verification cannot be performed · the next best move …

    line 129
  6. Distinctive:

    Never end in "recommendations." Every run must leave behind one of: a patch, a stronger artifact, a reverted failed mutation, a state …

    line 167

“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

Ratchet Evolve

Mutate. Test. Keep the delta. Serialize the next edge.

You are running a bounded artifact evolution loop. Your job is not to brainstorm improvements and not to "make this better." Your job is to produce evidence-gated deltas: one target, one pressure, small mutations, a proof gate, and a durable record.

LOCK → SNAPSHOT → PRESSURE → MUTATE → JUDGE → APPLY → VERIFY → KEEP/REVERT/ASK → RECORD → NEXT EDGE

The hard constraint, always:

No proof → no keep.
No keep → no progress claim.
No state → no loop continuity.

Invocation

/ratchet:evolve <target> --goal "<improvement>" [--iterations 2] [--test "<cmd>"] [--mode code|prompt|docs|workflow|auto] [--write]

Parse from $ARGUMENTS: the target artifact, the --goal, --iterations (default 2), an optional --test command, --mode (default auto), and --write (default false — without it, propose patches but do not modify files). If the target or goal is missing, go straight to ASK.

The deterministic helpers live in the ratchet-evolve CLI (call via Bash; fall back to node "<plugin root>/bin/ratchet-evolve" if not on PATH). In Claude Code, <plugin root> is $CLAUDE_PLUGIN_ROOT; in Codex local development, use this repo root or the installed plugin path shown by codex plugin list --json.

The loop

1. LOCK

State, before any edit: the exact target, the desired delta, the proof that the delta worked, and the forbidden scope. Scope drift is the primary failure mode — name what you will not touch.

2. SNAPSHOT

Capture the baseline. No baseline = rewriting, not evolving.

ratchet-evolve snapshot <target> --goal "<goal>" --mode <mode>

Records path, hash, git state, mode, byte/line count. Read the file for real before mutating.

3. PRESSURE

Choose one primary pressure and at most one secondary. Name the pressure you reject and why (usually novelty, because it tempts a rewrite instead of a stronger delta). ratchet-evolve pressure <mode> suggests a starting vector.

4. MUTATE

Generate candidate deltas, not rewrites. Each candidate: name · exact change · why it should improve the artifact · expected measurable effect · risk · rollback cost. No mutation may exceed the locked scope — if the best move would, stop and ASK.

5. JUDGE

Score candidates and pick exactly one. No menu ending.

ratchet-evolve score mutation '[{"name":"...","impact":5,"evidence":4,"reversibility":5,"goalFit":5,"risk":2,"complexity":2}, ...]'

Score = Impact × Evidence × Reversibility × Goal-Fit − Risk − Complexity. State the chosen mutation and the most tempting rejected one, with the reason it lost.

6. APPLY

Apply the smallest safe delta (only if --write; otherwise present the patch). No unrelated cleanup, no aesthetic rewrite. Show it as:

CHANGE
Before: <old section>
After:  <new section>
Reason: <one sentence>

7. VERIFY

Gather evidence — never self-grade.

ratchet-evolve verify <target> --test "<cmd>" --mode <mode>

With a test command it runs and reports pass/fail. Without one, it returns manual checks you must actually perform. For prompts/docs, run the embarrassment harness: can a lazy model escape the instruction? can a user misread the next action? does it still force a concrete output, prevent self-praise, and leave durable state?

8. KEEP / REVERT / ASK

End every iteration in exactly one state.

  • KEEP only if: delta matches goal · no critical regression · verification evidence exists · scope stayed locked.
  • REVERT if: the mutation caused a failure · the gain is only aesthetic · the artifact got longer but not stronger · it introduced ambiguity. (If --write, actually undo it.)
  • ASK if: target/permission/test is missing, or the next mutation touches high-risk files.

9. RECORD

Write the evolution event (whenever project files changed, or at each iteration):

ratchet-evolve log append '{"target":"...","goal":"...","iteration":1,"pressure":{"primary":"testability","secondary":"specificity"},"chosenMutation":"...","filesTouched":["..."],"verification":{"commands":[],"manualChecks":["..."],"result":"pass"},"verdict":"KEEP","remainingRisks":["..."],"nextEdge":"..."}'

Appends to <project>/.ratchet/evolve-log.jsonl.

10. NEXT EDGE

Name the next mutation: small, specific, testable, attached to the same artifact. It is persisted with the event and readable next session via ratchet-evolve next. Check ratchet-evolve status for the running kept/reverted/asked tally.

Copy-paste shapes live in templates/: evolve-event.json (the log-append payload), evolve-manual-checks.md (manual evidence when there is no test command), and evolve-report.md (the output contract below).

Stop rules

Stop when: max iterations reached · a mutation fails and cannot be safely patched · verification cannot be performed · the next best move requires changing the locked target · improvement becomes aesthetic instead of functional · a risky write needs approval. Defaults: maxIterations 2 · maxFilesTouched 3 · maxMutationSize small · allowDestructiveEdits false.

Output contract — return exactly this

RATCHET EVOLVE REPORT

TARGET: <path>
GOAL: <one line>
BASELINE: <hash · mode · git state>

PRESSURE
Primary: <p>
Secondary: <p or none>
Rejected: <p> — <why>

ITERATION 1
Mutation candidates: <named list with scores>
Chosen mutation: <name> — <why it won>
Patch: <CHANGE block, or "proposed (no --write)">
Verification: <commands + result, or manual checks + result>
Verdict: KEEP | REVERT | ASK

ITERATION 2
...

FINAL STATE
Kept changes: <list or none>
Rejected changes: <list or none>
Tests/evals run: <list>
Remaining defects: <list or none>
Next edge: <the next small mutation>
State written: .ratchet/evolve-log.jsonl (<n> events)

Never end in "recommendations." Every run must leave behind one of: a patch, a stronger artifact, a reverted failed mutation, a state record, or a blocking question. This is a mutation loop with a gate — it is never a general "make this better" command.

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.