Oc Guard Skill

Safe OpenClaw config planning/apply workflow with bilingual execution receipts.

Security Scan

The skill's files, instructions, and runtime behavior are consistent with a local OpenClaw config guard: it invokes a bundled Python CLI, uses the OpenClaw/OpenCode CLIs, and only accesses local paths and optional environment overrides.

Purpose & Capability

Name/description (oc-guard: safe config plan/apply) match the included files and required binaries. The script implements plan/apply, path allowlists, validation, backup/rollback, and bilingual receipts. Required binaries (python3, openclaw, opencode) are appropriate for the stated purpose.

Instruction Scope

SKILL.md directs agents to run the bundled CLI and enforce plan/apply/gating rules; it does not ask the agent to read unrelated system files or external endpoints. The repository docs and AGENTS.md warn not to publish secrets and call out diagnostic files in /tmp; these behaviors align with the script.

Install Mechanism

There is no install spec (instruction-only with a bundled script). That is low-risk and coherent: the repo expects the script to be made executable and linked into ~/.local/bin optionally. No external download URLs or archive extracts are present.

Credentials

The skill declares no required env vars but supports a set of optional environment overrides (OPENCLAW_HOME, OCGUARD_*, etc.). The script will create a local receipt secret file (~/.openclaw/.ocguard_receipt_secret) if OCGUARD_RECEIPT_SECRET is not provided. This is proportionate to producing signed receipts but users should be aware of the local secret file and the optional env overrides.

Persistence & Privilege

The skill does not request always:true and is user-invocable only — normal defaults. It writes a local receipt secret and log files under ~/.openclaw and /tmp and may create backups under ~/.openclaw/config-backups. These are scoped to the user's environment and are expected for a config guard; users should confirm they are comfortable with these local files being created.

Assessment

This skill appears to do what it claims: a local Python CLI that gates OpenClaw config changes via plan/apply, performs backups, and emits signed bilingual receipts. Before installing or allowing autonomous invocation, check: 1) verify the openclaw and opencode binaries on your system are the official tools the script will call (the script runs them via subprocess), 2) review scripts/oc-guard.py (it will create a receipt secret file at ~/.openclaw/.ocguard_receipt_secret and logs in /tmp) and confirm you are comfortable with those local files, 3) run the provided smoke checks (python3 -m py_compile scripts/oc-guard.py and scripts/oc-guard.py plan --proposal templates/proposal.template.json) and inspect outputs, and 4) ensure /tmp diagnostic files (e.g., /tmp/oc-guard-last-opencode-output.txt) are handled as they may contain local diagnostic output. The skill does not request external credentials and does not contain hardcoded remote endpoints, but it does invoke external CLIs — only proceed if those CLIs are trusted. If you want extra assurance, run the script in a sandbox or inspect the remainder of the script not included in the truncated listing.

Purpose

All config-changing requests must go through oc-guard. Do not directly edit ~/.openclaw/openclaw.json. When possible, invoke the bundled CLI at {baseDir}/scripts/oc-guard.py.

Hard Rules

1. Use oc-guard plan before apply.
2. High-risk changes require oc-guard apply --confirm.
3. Always return execution receipt first.
4. If command is not executed, return 【模型说明-未执行】.
5. Never claim success without a real oc-guard receipt.

Common Commands

• {baseDir}/scripts/oc-guard.py --help
• {baseDir}/scripts/oc-guard.py plan "<requirement>"
• {baseDir}/scripts/oc-guard.py apply --confirm "<requirement>"
• {baseDir}/scripts/oc-guard.py plan --proposal <file>
• {baseDir}/scripts/oc-guard.py apply --confirm --proposal <file>