dsh-continual-evolve

中文 | English

Continual self-evolution for DeepSeek Harness: a versioned, auditable, rollback-safe harness state layer — prompt notes, memories, skills, subagent specs — refined from session trajectories.

The model proposes, the code guarantees. Every mechanical safety property — schema validation, atomic writes, snapshots, versioning, audit trail, acceptance decisions — is enforced in code, never by prompt discipline.

Why

Agents accumulate reusable experience (repeated failures, durable facts, reusable procedures) and forget it next session. This plugin turns that experience into first-class state:

• Local scope per session; global scope across sessions with merge semantics — plus mechanical promotion guards so only portable, substantial, non-duplicate knowledge reaches global
• Deterministic rollback: inverse edits generated from applied results — no LLM re-guessing
• Benchmark loop: candidate refinements are evaluated against frozen cases by a separate scorer before acceptance (rubric encrypted at rest)

How it works

1. Sediment — the model creates entries via evolve_add, or the automatic review gate proposes them from the session trajectory (turn-interval + compaction checkpoints).
2. Guard — code-enforced validation: edit schema, blast-radius/scope coherence, and the promotion policy (project-scoped markers, thin content, near-duplicate detection keep the global store clean).
3. Approve — global writes require explicit human approval; local-fate proposals are consulted before they land.
4. Apply & inject — atomic apply with snapshot + audit event. Prompt notes and delegation specs inject into the system prompt (capped, relevance-ranked, zero tokens when empty); memories/skills appear as a capped directory index.
5. Validate & roll back — benchmarks score candidates against frozen cases; rejected candidates roll back deterministically.

Install

# from npm (installs and activates — ships its own bundle patch)
dsh plugin add dsh-continual-evolve

# or from source (first GitHub installs require approving the allowBuilds step)
dsh plugin add ZK-Andy/dsh-continual-evolve

Restart dsh web after installing or updating.

Usage

Commands (in-session):

Model tools: evolve_list / add / update / delete / rollback.

Injection shape: prompt notes and delegation specs inject with content (≤6/kind × 180 chars, relevance-ranked). Memories and skills appear as a directory index ([kind:id] title, capped at 15 lines with a fold counter) — full text via evolve_list. Empty store = zero injected tokens.

Configuration

Example profile patch:

- id: continual-evolve
  config:
    autoReview: true
    reviewIntervalTurns: 6

Development

pnpm install && pnpm build   # deps + tsc -> lib/
pnpm test                    # vitest (527 tests)
pnpm test:coverage           # v8 coverage, thresholds enforced in CI
pnpm lint                    # oxlint src test

Project layout:

├── src/                   # engine, tools, commands, gate, fate, benchmark, usage…
├── test/                  # vitest suites (33 files)
├── lib/                   # build output (tsc)
├── docs/
│   ├── design.md          # full design doc (hardening matrix)
│   ├── FAQ.md             # real failure/fix records
│   ├── gap-analysis.md    # vs prime-agent /refine + penguin-harness
│   ├── experiment-bootstrap.md
│   ├── archive/           # closed point-in-time reports
│   └── research/          # penguin report + prime-agent annotated source
├── examples/README.md     # seed benchmark cases
└── .agents/               # AI collaboration layer (AGENTS.md, skills, ADR notes)

Docs & provenance

• Design: docs/design.md · Pitfalls: docs/FAQ.md · Gap analysis: docs/gap-analysis.md · D2 experiment: docs/experiment-bootstrap.md
• Lineage: penguin-harness (concept; Apache-2.0) — report in docs/research/penguin-harness-self-evolution.md; prime-agent /refine (engineering shape; MIT) — annotated reference source in docs/research/prime-agent-refinement.ts. This package is an original implementation on the DSH plugin surface.

License

MIT