The Agent Skills Directory

[SAFE]: The skill is strictly educational documentation for a rule-based monitoring system. It promotes security best practices by providing examples of how to detect and warn against dangerous operations such as privilege escalation (sudo), recursive file deletion (rm -rf), and the accidental inclusion of sensitive information (.env files, API keys) in code. No evidence of data exfiltration, remote code execution, or malicious obfuscation was found.
[PROMPT_INJECTION]: The skill describes an architecture that ingests untrusted data (shell commands, file contents, and user prompts) to trigger rule-based messages. While this creates a surface for indirect prompt injection—where external data might trigger a rule to misguide the agent—the skill itself serves a defensive purpose and lacks any malicious injection payloads. Evidence chain: 1. Ingestion points: command, file_path, new_text, old_text, content, and user_prompt fields as defined in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: bash tool monitoring, file tool monitoring, and stop event interception. 4. Sanitization: Absent (uses regex matching).

Writing Hookify Rules