Permission Guard
v1.0.0分层权限守卫系统。基于 Claude Code Permission System 设计，包含危险命令拦截、规则优先级链、自动模式白名单和拒绝追踪。
⭐ 0· 117·0 current·0 all-time
by@sinnzen
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
medium confidencePurpose & Capability
Name and description match the actual contents: an instruction-only permission-checking system for command execution. The skill requests no binaries, credentials, or installs, which is proportional to a rules-and-checks design.
Instruction Scope
SKILL.md stays on-topic (command parsing, blacklists/whitelists, secret scanning, denial tracking, shadow-rule detection, hooks). It references reading/writing a user config (~/.openclaw/permissions.json) and logging/denial tracking — expected for a guard. However, the instructions include auto-mode rule updates and post-execution updates to rules (updateAutoModeRules), which give the agent discretion to change persistent policy; this is a behavior-level risk (not an incoherence) because it can silently broaden allowed actions if misused.
Install Mechanism
No install spec and no code files — instruction-only. Nothing will be written to disk by an installer step. Risk from installation mechanism is minimal.
Credentials
The skill declares no required environment variables, credentials, or external config paths. It proposes local secret-scanning regexes (e.g., sk-... patterns) which are reasonable for detecting sensitive tokens but do not by themselves request or exfiltrate secrets.
Persistence & Privilege
The skill suggests persisting user config and denial tracking under ~/.openclaw/permissions.json and updating auto-mode rules after decisions. Combined with the platform default (model invocation allowed), automatic rule updates could lead to gradual permission creep if an agent is allowed to act autonomously. There is no 'always: true' or system-wide config edits requested, and it does not ask to modify other skills' configs.
Assessment
This skill is internally consistent with being a local permission guard and doesn't request credentials or install anything. Before enabling it, consider: 1) Review and control the config file (~/.openclaw/permissions.json) format and contents so auto-whitelisting won't silently approve dangerous commands. 2) Disable or restrict any feature that lets the agent auto-update rules unless you require and audit that behavior. 3) Avoid using 'bypass' mode and keep defaultMode conservative (e.g., 'ask' or 'default'). 4) If you allow autonomous agent invocation, add audits or require explicit confirmations for saving rule changes. 5) Test the guard in a safe environment to ensure the rejection/allow rules behave as you expect. These mitigations reduce the operational risk even though the skill itself appears coherent.Like a lobster shell, security has layers — review code before you run it.
latestvk979n4zbnmtkwx2ev9vbr48zjs840h1k
License
MIT-0
Free to use, modify, and redistribute. No attribution required.