Openclaw Safe Uninstall

OpenClaw Safe Uninstall

Overview

This skill performs high-risk local OpenClaw uninstall work through a strict sequence:

1. scan
2. plan
3. explicit human approval
4. apply
5. verify

Do not jump straight to deletion.

Use the deterministic helper script in scripts/openclaw_uninstall.py instead of ad-hoc rm -rf unless the user explicitly wants a manual one-off cleanup.

Hard boundaries

• Never auto-delete anything under .codex/skills or .agents/skills.
• Never auto-delete anything under protected AI tooling roots such as .claude/skills, .claude/agents, .claude/commands, .gemini/commands, .opencode/skills, or .config/opencode/skills.
• Never treat openclaw in a path string as sufficient deletion evidence by itself.
• Known companion traces such as AutoClaw must be reported as manual_review, not auto-deleted.
• Require explicit user intent before apply.
• Require all three apply flags:
  • --yes
  • --acknowledge-risk
  • --confirm "REMOVE OPENCLAW FROM THIS MACHINE"
• If root-owned artifacts remain, report exact manual commands rather than claiming success.

Required workflow

Step 1: Scan

Run:

python3 scripts/openclaw_uninstall.py scan --json

Review:

• detected state directories
• service registrations
• CLI/package installs
• shell injection
• manual review artifacts
• excluded paths

Step 2: Build the uninstall plan

Run:

python3 scripts/openclaw_uninstall.py plan --json

Explain:

• what will be deleted automatically
• what requires privilege
• what is intentionally excluded
• what remains manual review only

Step 3: Apply only with explicit approval

Run:

python3 scripts/openclaw_uninstall.py apply \
  --yes \
  --acknowledge-risk \
  --confirm "REMOVE OPENCLAW FROM THIS MACHINE" \
  --json

If the user wants rehearsal first, add --dry-run.

Step 4: Verify

Run:

python3 scripts/openclaw_uninstall.py verify --json

Summarize:

• removed artifacts
• remaining privileged artifacts
• remaining manual review artifacts
• exact next-step commands if anything remains

Platform notes

• macOS:
  • detect LaunchAgents like ai.openclaw.gateway, ai.openclaw.<profile>, and legacy com.openclaw.*
  • detect shell completion injection
  • detect optional app bundle paths
• Linux:
  • detect systemd --user units such as openclaw-gateway[-<profile>].service
  • detect legacy unit names like clawdbot-gateway and moltbot-gateway
  • detect wrapper installs and global package paths
• Windows:
  • detect Scheduled Task naming conventions
  • detect Startup-folder fallback launchers
  • detect gateway.cmd and node.cmd in the state dir
• Neighboring AI tooling ecosystems:
  • protect Claude Code subagent/custom-command directories
  • protect Gemini CLI custom-command directories
  • protect OpenCode skill directories and compatibility skill roots

References

• references/research-notes.md
• references/artifact-matrix.md
• references/safety-model.md

Validation

Before you claim the uninstall plan is correct, run:

python3 -m unittest discover -s tests -v