Dsh Auto Evolve — AI skill for Claude Code

dsh-auto-evolve

A **self-evolving plugin** for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) (`dsh`). It observes how the agent runs, proposes improvements to *its own* assets via the LLM, validates each proposal inside a sandboxed trial agent, and applies only verified mutations — with a versioned ledger and automatic rollback on regression.

[中文](README.zh.md)

**The idea.** Instead of shipping a static skill set, the plugin owns a small *genome* of evolvable assets (skills, tool-result post-processors, prompt sections, guard policies). Every runtime signal (tool failures, repeated calls, request errors) is collected; when a threshold crosses, a proposal cycle drafts a mutation, a sandboxed agent replays the failing episode with and without the change, and the change is applied only if the trial shows measurable improvement. Every mutation is versioned in a durable ledger and can be rolled back.

How it works

Observe ──▶ Propose ──▶ Validate ──▶ Apply ──▶ (observe again, for regressions)
  │             │            │            │
  tools/result  ctx.llm      sandboxed    ctx.skills.register + ledger
  request-error (stream)     sub-agent    disposer kept for rollback

The evolution loop

Layer	Module	What it does
Observe	src/observe	Listens on tools/result and agent/request-error; records deduplicated signals (tool failures, no-progress repeats, request errors) into the durable observations table; fires onTrigger when a threshold crosses.
Propose	src/propose	A bounded cycle snapshots the genome + recent observations, calls ctx.llm.stream() with a strict prompt, and validates the model output against a closed mutation vocabulary (add / patch / retire over skill / post-processor / prompt-section / guard-policy). Anything that fails parsing or schema validation is discarded — never applied. Mutations whose content fingerprint already matches a pending `ca