How do I install the Self Evolve skill?

Run npx skillmds@latest add lstr-1/self-evolve in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Self Evolve safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Self Evolve?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Self Evolve free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Self Evolve?

LSTR-1 (@lstr-1) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Self Evolve by lstr-1 — Agent Skill

Name: Self Evolve
Author: LSTR-1

Self Evolve

Reflects on a completed task's outcome and distills reusable atomic rules into persistent state (skills and memory), governed by the self-evolution policy. Invoke after a task passes its deterministic grader, on "/reflect", or as the Evolve stage of the metacognitive loop. Trigger phrases "reflect on this task", "/reflect", "learn from this", "distill a rule".

LSTR-1 a7ad392 2.1 KB Updated Sep 17, 2026

File contents

self-evolve

GDPevo reflection loop. Converts a passed task's outcome into persistent, provenance-tracked state, without destructive mutation.

When to Use

After a task has been graded and passed its deterministic grader.
When a reusable rule was exercised and should persist across future tasks.
On /reflect to close the Evolve stage of the metacognitive loop.

When NOT to Use

Do NOT use when the task failed its grader (evolving from a failed task propagates error).
Do NOT use for one-time or temporary instructions.

Instructions

Score the task — obtain grader_score from the deterministic grader (self-test). If no grader ran, do NOT distil; report the gap.
Select supervision type per 40-evolution-policy.md: reflect (default), fewshot (canonical only), self (exploratory).
Distil atomic rules — extract the smallest reusable rule(s) the task exercised. Record provenance: task_slug, grader_score, supervision_type.
Propose, do not auto-commit — write the proposal to .trae/memory/evolution-ledger.md with status: proposed. Apply to a skill or memory only on approval.
Record cost — tokens, turns, and tool calls, alongside accuracy (cost is a first-class metric).

Output Contract

One ledger entry appended with full schema: timestamp, task_slug, supervision_type, grader_score, distilled_rule, provenance, status.
An explicit proposal (skill update / memory update / no-op) with the reason.

Constraints

No destructive or irreversible evolution without explicit approval.
No secrets or personal data in any distilled artifact (GDPR Art. 28).
MUST NOT evolve away from 00-identity.md or 20-output-spec.md.

---name: self-evolvedescription: Reflects on a completed task's outcome and distills reusable atomic rules into persistent state (skills and memory), governed by the self-evolution policy. Invoke after a task passes its deterministic grader, on "/reflect", or as the Evolve stage of the metacognitive loop. Trigger phrases "reflect on this task", "/reflect", "learn from this", "distill a rule".---# self-evolveGDPevo reflection loop. Converts a passed task's outcome into persistent, provenance-tracked state, without destructive mutation.## When to Use- After a task has been graded and passed its deterministic grader.- When a reusable rule was exercised and should persist across future tasks.- On `/reflect` to close the Evolve stage of the metacognitive loop.## When NOT to Use- Do NOT use when the task failed its grader (evolving from a failed task propagates error).- Do NOT use for one-time or temporary instructions.## Instructions1. **Score the task** — obtain `grader_score` from the deterministic grader (`self-test`). If no grader ran, do NOT distil; report the gap.2. **Select supervision type** per `40-evolution-policy.md`: `reflect` (default), `fewshot` (canonical only), `self` (exploratory).3. **Distil atomic rules** — extract the smallest reusable rule(s) the task exercised. Record provenance: `task_slug`, `grader_score`, `supervision_type`.4. **Propose, do not auto-commit** — write the proposal to `.trae/memory/evolution-ledger.md` with `status: proposed`. Apply to a skill or memory only on approval.5. **Record cost** — tokens, turns, and tool calls, alongside accuracy (cost is a first-class metric).## Output Contract- One ledger entry appended with full schema: `timestamp`, `task_slug`, `supervision_type`, `grader_score`, `distilled_rule`, `provenance`, `status`.- An explicit proposal (skill update / memory update / no-op) with the reason.## Constraints- No destructive or irreversible evolution without explicit approval.- No secrets or personal data in any distilled artifact (GDPR Art. 28).- MUST NOT evolve away from `00-identity.md` or `20-output-spec.md`.

Frequently asked questions