SkillOpt-Sleep OpenClaw Reference Adaptation
A reference-only port of the SkillOpt-Sleep nightly self-evolution cycle to OpenClaw — for study and porting, not a ready-to-run integration.
Data boundary disclosure is thorough (DeepSeek/Ollama outbound content scope, non-redaction assumption, manual adoption gate, auto_adopt=false, backups), with guidance on TLS, secret management and isolated testing; deducted for plaintext credential loading from ~/.openclaw/.env, contributor machine absolute paths, and redact_secrets=true being only a sample config value rather than an enforced guarantee.
Scripts are self-consistent, return diagnosable errors like [BACKEND_ERROR], and offer dry-run and mock-backend alternatives; deducted because the README explicitly states the checkout is not directly runnable (Python 3.10 syntax gaps, stale backend factory signature), key paths cannot be statically confirmed, and edge-case handling is thin.
SKILL.md/README clearly declare the audience (porters, not end users), non-fit boundaries, trigger commands (/sleep subcommands) and configuration caveats; DeepSeek is mainland-China reachable so no environment-block issue; deducted because triggering depends on an unported command handler and fixtures are single-environment examples only.
Docs are well layered (SKILL.md → README → shared docs), MIT license explicit, maintenance responsibility (official Microsoft repo) clear, and known limitations honestly disclosed; deducted for no per-directory versioning/changelog, leftover contributor-environment assumptions, and slash_sleep.py targeting an old manifest shape.
The claimed value is reference material only; the README states it is not plug-and-play and output requires substantial adaptation. Sample config values are not enforced and cost estimates come from a stale environment measurement; marginal value is limited to study/porting use, so direct-usability points were heavily deducted.
The core repository has CI and a test suite, but none covers this plugin path; tests/*. here are example fixtures, not executable assertions, and token/cost figures derive from a single unverifiable dry-run with no third-party reproduction of this skill.
- The directory is explicitly declared non-runnable; use it only as porting reference, do not execute or schedule unchanged.
- Absolute paths and limits in the sample config are not enforced; replace and review all before porting.
- Credentials are loaded from a plaintext .env; prefer your own secret-management mechanism.
- Outbound prompts are not guaranteed redacted; inspect task content, TLS and provider retention before using real data.
- This skill has no dedicated test coverage or versioning; add isolated end-to-end tests before adoption.
What does this skill do, and when should you use it?
This is the reference adaptation in microsoft/SkillOpt's plugins/openclaw directory, showing one way to wire the shared skillopt_sleep nightly self-evolution cycle (harvest → mine → replay → consolidate, all behind a held-out validation gate) to a custom DeepSeek Chat Completions backend. The SKILL.md is explicit that it is a contributed reference, not a supported plug-and-play integration: scripts preserve assumptions from the contributor's machine, and parts of the wrapper have not been ported to current shared-engine interfaces. It suits maintainers planning an OpenClaw port of SkillOpt-Sleep, not anyone seeking a deployable skill today.
Provides skillopt_sleep_openclaw.py (a DeepSeek backend prototype with an Ollama embedding helper not wired into the cycle), run_sleep.py (a custom cycle wrapper with environment-specific paths and a backend-registration shim), slash_sleep.py (an experimental command helper for an older staging-manifest shape), run_sleep_cron.sh (a machine-specific category runner, not a portable cron installer), a sample config., and example task fixtures. The SKILL.md itself is a status and adaptation guide enumerating six porting gaps: replacing hardcoded paths, updating the backend factory to the current get_backend contract, migrating to the current staging manifest and adopt flow, converting OpenClaw transcripts into a supported session format, rebuilding scheduling around the adapted wrapper, and adding isolated end-to-end tests.
- A maintainer wants to port the SkillOpt-Sleep nightly self-evolution cycle to OpenClaw and needs an existing attempt as a starting point.
- An engineer wants to study how to write a custom DeepSeek Chat Completions backend for skillopt_sleep.
- A team is deciding whether the porting effort is worthwhile and needs an explicit gap list against the current supported interfaces.
- A contributor plans to turn this reference into a portable integration and must update the reference README, plugin integration reference, and Sleep docs together.
What are this skill's strengths and limitations?
- Honestly labeled as a reference rather than a product, with the directory README as the authoritative adaptation guide.
- Enumerates six concrete porting gaps and the current backend call contract, giving porters a clear roadmap.
- Shows real working examples of a custom DeepSeek backend and the staging/adopt flow.
- Includes explicit data and credential boundary warnings (what is sent to the remote endpoint, secret handling, TLS and retention checks).
- Not a runnable integration: paths, backend protocol, and staging logic are not migrated to current shared-engine interfaces.
- run_sleep_cron.sh depends on the contributor's original machine paths; config. limits are not guaranteed or enforced.
- No end-to-end tests for this directory; fixture scores and old cost estimates establish nothing about other deployments.
- The shared scheduler will not invoke this custom backend or its category fixtures; DeepSeek support exists only in this prototype.
How do you install this skill?
The directory ships inside the microsoft/SkillOpt repository at plugins/openclaw/ and cannot be installed as a standalone working skill. The collection is installable via pip install skillopt (Python 3.10+, MIT license), but the README documents no separate installation path for this reference directory. The SKILL.md explicitly instructs: do not run or schedule the files unchanged.
How do you use this skill?
Treat it as reading material: start with the directory's README.md (the authoritative status and adaptation guide), then consult the plugin integration reference (../README.md) and the SkillOpt-Sleep docs (docs/sleep/README.md). To actually exercise SkillOpt-Sleep itself, the source recommends the supported shared CLI: python -m skillopt_sleep --backend mock. Concrete adaptation steps beyond the six documented gaps are not provided.