Liquefy OpenClaw

Data handling (v0.1.1): All input values are passed through an inline vault-scan gate (ported from tools/liquefy_redact.py) before reaching the Liquefy CLI. The gate strips API keys, tokens, credentials, PII, and card numbers, replacing them with typed placeholders. No matched values are logged. Workspace files are processed locally; nothing is transmitted externally.

Use this skill when the user wants to:

• scan or pack an OpenClaw workspace with Liquefy
• understand why Liquefy denied a path
• search or restore from an existing Liquefy vault
• run OpenClaw with rollback, context budgeting, replay blocking, or restore-on-failure
• install or inspect native Liquefy integration inside OpenClaw
• vault OpenClaw session/state directories
• protect OpenClaw state or pulled history before risky actions

Start Here

Default to the local Liquefy CLI:

• liquefy openclaw --json
• liquefy openclaw run --json
• liquefy search
• liquefy restore
• liquefy safe-run --json
• liquefy context-gate compile --json
• liquefy context-gate history --json
• python tools/openclaw_tracevault.py
• python tools/liquefy_openclaw_plugin.py hook install|uninstall|status
• liquefy state-guard ...
• liquefy history-guard ...

Optional packaged plugin path:

• if the user already has the OpenClaw plugin installed, liquefy_scan and liquefy_pack_apply are available
• do not assume those plugin tools exist in a standalone skill install

Preflight

Before using repo-relative tools or plugin commands:

• verify liquefy is installed and callable
• verify repo-relative tools/ paths exist before using python tools/...
• verify optional plugin tools exist before using them
• verify LIQUEFY_SECRET is set before any secure pack/apply flow
• prefer --json for commands that support it when machine-readable results matter

If a required command, repo path, or dependency is missing:

• stop and explain the missing dependency
• do not guess alternate commands
• do not pretend plugin or repo-only surfaces are available

Compatibility note:

• this standalone skill is aligned to the Liquefy OpenClaw command surface currently shipped in this repo
• check liquefy version --json and liquefy openclaw --version --json before relying on newer flags
• if the optional OpenClaw plugin is used, it expects Liquefy OpenClaw CLI >= 1.1.0

Hard Rules

1. Prefer read-only scan before any write/apply action.
2. Do not ask for --include-secrets or risky allow-category overrides unless the user explicitly wants denied paths included.
3. If the user asks for protection on the next OpenClaw run, do not use raw openclaw run; prefer liquefy openclaw run or liquefy safe-run.
4. If context-gate reports blocked=true, stop and explain block_reason instead of trying to route around it.
5. If secure pack/apply is requested, ensure LIQUEFY_SECRET exists before enabling --secure.
6. Do not pretend the standalone skill install also installed the plugin package. Treat plugin tools as optional.
7. Do not run pack, apply, restore, hook install/uninstall, or other write-affecting operations unless the user explicitly requested that action after review.

References

• For workflow selection and commands, read references/workflows.md.
• For install/update truth, standalone vs plugin split, and trust posture, read references/integration.md.
• For JSON contracts and stable machine-readable fields, read references/contracts.md.

Load only the reference file needed for the current request.

Quick Routing

• Workspace audit or pack: use liquefy openclaw by default; only use plugin tools if the user already has the plugin enabled.
• Policy denial or override question: use liquefy openclaw --explain ... and report the matched category/reason.
• Next run must be protected: use liquefy openclaw run or liquefy safe-run, not raw OpenClaw.
• User wants automatic OpenClaw archival: use native hook install/status/uninstall via liquefy_openclaw_plugin.py.
• User wants specific agent session vaulting: use openclaw_tracevault.py.
• User needs restore/search on an existing vault: use liquefy search / liquefy restore.
• User wants filesystem/state protection outside packing: use state-guard or history-guard.

Example Intents

• "Scan my OpenClaw workspace and tell me what Liquefy would pack."
• "Why did Liquefy deny credentials/api.pem?"
• "Run OpenClaw with replay blocking and rollback."
• "Show the context gate history for this workspace."
• "Pack this workspace with ratio profile after review."
• "Install Liquefy hooks into OpenClaw and show status."
• "Restore this OpenClaw vault and search it for a trace id."
• "Protect wallet-state.json before the next run."