# Openclaw Mem Dream Lite

> Operate experimental Dream Lite planning, governed refresh canaries, rollback, and Director rehearsals. Use only for explicitly reviewed derived card maintenance.

- Skill: `phenomenoner/openclaw-mem-dream-lite` (Agent Skill)
- Install (CLI): `npx skillmds@latest add phenomenoner/openclaw-mem-dream-lite`
- Raw SKILL.md: https://api.skillmd.com/api/skills/phenomenoner/openclaw-mem-dream-lite/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: phenomenoner (https://skillmd.com/u/phenomenoner)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/phenomenoner/openclaw-mem-dream-lite

---


# Dream Lite Lab

## Authority

- Keep `apply plan` and `apply verify` zero-write.
- Restrict `apply run` to one governor-approved `refresh_card` with witness, snapshots, TTL, rolling write caps, receipt, and rollback.
- Keep `compile_new_card` proposal-only.
- Treat Director observe, stage, checkpoint, and apply outputs as untrusted rehearsal artifacts.
- Never let Director rehearsal canonize authority files or claim live mutation.

```bash
openclaw-mem dream-lite apply plan --json
openclaw-mem dream-lite apply verify --receipt <receipt> --json
openclaw-mem dream-lite apply rollback --receipt <receipt> --json
openclaw-mem dream-lite director observe --json
openclaw-mem dream-lite director checkpoint --json
```

Human notifications should summarize suggestions, recommended handling with reasons, choices, and the run window. Keep ordinary no-op and rehearsal-only runs silent.

## Verify

```bash
openclaw-mem dream-lite apply plan --json
python -m pytest tests/test_cli_surface_lock.py -q
git diff --check
```