Safe Exec.Bak

Safe command execution for OpenClaw Agents with automatic danger pattern detection, risk assessment, user approval workflow, and audit logging. Use when agen...

Key things to check before installing: 1) Resolve metadata mismatches: SKILL.md declares 'jq' and a git install; the registry listed none — ensure the actual install process will provide jq and uses the GitHub URL you reviewed. 2) Understand how "automatic monitoring" is implemented: the shipped code shows CLI wrappers and pending files, not a background daemon or shell hooks. Ask/inspect any platform integration that claims to "monitor all shell commands" — that step could add persistent hooks or cron jobs. 3) Audit the core scripts (especially safe-exec.sh and the request creation code). The approval executor uses eval on command text read from JSON files; ensure pending files cannot be modified by untrusted parties (file permissions) and that commands are sanitized. 4) Be cautious about agent-mode auto-bypass: OPENCLAW_AGENT_CALL or SAFE_EXEC_AUTO_CONFIRM cause confirmation to be skipped (non-interactive). If agents are untrusted or multi-tenant, this can let agents run commands without human approval for LOW/MEDIUM risk. Prefer keeping auto-confirm disabled unless you fully trust the agents. 5) Verify pending and audit directories permissions (only user-writable) and rotate/backup audit logs. Don't install system-wide or enable automatic shell hooking without a code review. 6) If you plan to install from the GitHub URL, compare the remote repo contents with the archive reviewed here to ensure no post-publish changes. If you want, I can: (a) point out exact lines in safe-exec.sh where eval/command execution happens, (b) list needed hardening changes (e.g., avoid eval, use safe exec arrays, check file ownership), or (c) produce a short checklist to use when reviewing the remote repo prior to cloning.

Type: OpenClaw Skill Name: safe-exec-bak Version: 1.0.0 The safe-exec skill is a legitimate security utility designed to provide a safety layer for OpenClaw agents by intercepting and auditing dangerous shell commands. It implements a risk assessment engine (in safe-exec.sh) that categorizes commands into risk levels (CRITICAL, HIGH, MEDIUM, LOW) using regex patterns and manages a local approval workflow. All executions are recorded in a local audit log (~/.openclaw/safe-exec-audit.log). While the tool includes features like an 'Agent Mode' that bypasses interactive prompts and a 'Context-aware' mechanism to downgrade risk levels based on specific user keywords, these are documented as intentional features for automation and power-user control. The package contains extensive documentation and changelogs (CHANGELOG.md, CLAWDHUB_SECURITY_RESPONSE.md) detailing the removal of previous monitoring and external network components in response to security reviews, demonstrating a clear focus on local security and transparency.