Local Approvals
v1.0.0Local approval system for managing agent permissions. Use CLI to approve/deny requests, view history, and manage auto-approved categories.
⭐ 3· 1.6k·3 current·3 all-time
by@shaiss
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description match the actual behavior: a local approvals CLI that reads/writes JSON state in ~/.openclaw/skills/local-approvals. Minor mismatch: _meta.json mentions "multi-channel notifications" and a remote repo/homepage, but the included code contains no notification or network logic and requires no credentials.
Instruction Scope
SKILL.md instructs running the local cli.py and references local state files under the user's home directory. The runtime instructions and the code operate only on local files and CLI args; they do not reference external endpoints, unrelated env vars, or system paths outside the skill's directory and the user's home.
Install Mechanism
No install spec or downloads are present; code is included in the bundle and runs with python. No external packages, URLs, or archive extraction are used.
Credentials
The skill does not request environment variables, credentials, or external config paths. It only reads/writes JSON under ~/.openclaw/skills/local-approvals, which is proportionate to a local approval system.
Persistence & Privilege
The skill persists state to ~/.openclaw/skills/local-approvals/state.json and pending.json (expected for this functionality). This is normal, but note that persisted 'auto_approve' entries can cause future operations to be auto-approved if the `--learn` option is used.
Assessment
What to consider before installing:
- This is a local-only CLI that stores state in ~/.openclaw/skills/local-approvals (state.json and pending.json). Back up these files if you need an audit trail before testing.
- The code does not request credentials or make external network calls, so it is coherent with a local approvals tool. Still inspect the included files yourself for any unexpected changes before running.
- Be cautious with the "--learn"/auto-learn feature: approving and auto-learning a category will allow that category to be auto-approved in the future without review. Use reset to revoke learned categories if needed.
- Metadata mismatches: _meta.json mentions multi-channel notifications and a GitHub homepage/repository that the code does not implement; verify origin/authenticity of the skill (check the repository or publisher) if provenance matters to you.
- Minor technical notes: schemas include JavaScript-style comments (//) which are not valid JSON if you try to parse them as strict JSON; this is a documentation artifact not used at runtime. Also the code uses status/decision naming like 'decided' / 'denied' / 'approved' while schemas mention 'rejected' — this is inconsistent but not a security issue; it may cause confusion in integrations.
- If you plan to allow agents to invoke skills autonomously, remember that this skill can be used to gate actions; ensure your agent's approval-requesting behavior and the human review process are aligned to avoid accidental auto-approvals.
If you want higher assurance, ask the publisher for the canonical repository link and verify the published files match that source.Like a lobster shell, security has layers — review code before you run it.
latestvk978tpsb5jg99hqy147498j7v580k5fz
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
