mirror of
https://github.com/microsoft/SkillOpt.git
synced 2026-07-07 00:15:39 +08:00
f9db99853b
Restructure into plugins/{claude-code,codex,copilot}/ — one engine, three thin
shells, all calling the shared plugins/run-sleep.sh -> python -m skillopt_sleep.
- claude-code/: existing plugin moved here; runner delegates to the shared
launcher (fixes repo-root resolution after the move).
- codex/: ~/.codex/prompts/sleep.md custom prompt + ~/.agents/skills SKILL.md +
install.sh + AGENTS.md hint — Codex's documented, stable extension surfaces.
- copilot/: a stdlib-only MCP server (mcp_server.py) exposing sleep_* tools,
plus mcp-config.example.json and a copilot-instructions snippet. Verified end
to end (initialize -> tools/list -> tools/call returns real engine output).
- plugins/README.md overview table; main README News + a dedicated SkillOpt-Sleep
section; pyproject lists skillopt_sleep as a first-class package.
Decoupling emphasized throughout: open-source tool (skillopt_sleep/) with zero
dependency on the research package. 29 tests pass; all three shells resolve.
Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
75 lines
3.2 KiB
Markdown
75 lines
3.2 KiB
Markdown
# SkillOpt-Sleep — plugins for Claude Code, Codex, and Copilot
|
|
|
|
One engine, three thin shells. **SkillOpt-Sleep** gives a local coding agent a
|
|
nightly **sleep cycle**: it reviews your past sessions offline, replays your
|
|
recurring tasks on your own API budget, and consolidates what it learns into
|
|
**validated** long-term memory and skills — behind a held-out gate, staged for
|
|
your review. Your agent gets better the more you use it, with no model-weight
|
|
training.
|
|
|
|
It synthesizes three ideas: **SkillOpt** (validation-gated bounded text
|
|
optimization — the research in this repo), **Claude Dreams** (offline memory
|
|
consolidation; input never mutated; review-then-adopt), and the **agent sleep**
|
|
literature (short-term experience → long-term competence).
|
|
|
|
> **This is an open-source tool, decoupled from the research code.** The engine
|
|
> lives in the top-level [`skillopt_sleep/`](../skillopt_sleep) package and has
|
|
> **zero dependency** on the paper's `skillopt/` experiment package (the
|
|
> validation gate is vendored). You can ship/use it without the research stack.
|
|
|
|
## The three integrations
|
|
|
|
| Platform | Folder | Mechanism | Status |
|
|
|---|---|---|---|
|
|
| **Claude Code** | [`claude-code/`](claude-code) | `.claude-plugin` + `/sleep` command + skill + hooks | full, installable |
|
|
| **Codex** | [`codex/`](codex) | `~/.codex/prompts/sleep.md` + `~/.agents/skills` + `AGENTS.md` | full |
|
|
| **Copilot** | [`copilot/`](copilot) | MCP server (`sleep_*` tools) + `copilot-instructions` | full (MCP) |
|
|
|
|
All three call the **same** [`plugins/run-sleep.sh`](run-sleep.sh) → `python -m
|
|
skillopt_sleep`, so behaviour is identical everywhere. Per-platform setup is in
|
|
each folder's README.
|
|
|
|
## Quick start (Claude Code)
|
|
|
|
```bash
|
|
git clone <repo-url> && cd SkillOpt-Sleep
|
|
# Claude Code:
|
|
/plugin marketplace add ./plugins/claude-code
|
|
/plugin install skillopt-sleep@skillopt-sleep
|
|
/sleep status
|
|
```
|
|
Codex: `bash plugins/codex/install.sh`.
|
|
Copilot: register `plugins/copilot/mcp_server.py` as an MCP server.
|
|
|
|
## What one "night" does
|
|
|
|
```
|
|
harvest ~/.claude (or session) transcripts → mine recurring tasks → replay offline
|
|
→ consolidate (reflect → bounded edit → GATE on real held-out tasks)
|
|
→ stage proposal → (you) adopt
|
|
```
|
|
|
|
Nothing live changes until you adopt; every adopt backs up first.
|
|
|
|
## Controls (work on all platforms)
|
|
|
|
`--gate on|off` · `--rollouts-k K` (multi-rollout contrastive reflection) ·
|
|
`--budget-tokens/--budget-minutes` · `--preferences "..."` · separate
|
|
optimizer/target models (`--optimizer-model` / `--target-model`) · slow-update
|
|
long-term memory. Full guide:
|
|
[`../docs/sleep/CONTROLLABLE_DREAMING.md`](../docs/sleep/CONTROLLABLE_DREAMING.md).
|
|
|
|
## Does it actually work?
|
|
|
|
Validated on the public
|
|
[gbrain-evals](https://github.com/garrytan/gbrain-evals) `skillopt-v1` benchmark
|
|
with **real models on both Claude and Codex**: deficient skills go **0.00 →
|
|
1.00** on held-out sets (all 4 seeds incl. a real tool-use loop), cross-model
|
|
transfer is positive, and the gate blocks regressions. Full results:
|
|
[`../docs/sleep/FINAL_REPORT.md`](../docs/sleep/FINAL_REPORT.md).
|
|
|
|
Deterministic proof (no API key):
|
|
```bash
|
|
python -m skillopt_sleep.experiments.run_experiment --persona researcher --assert-improves
|
|
```
|