The Agent Skills Directory

[PROMPT_INJECTION]: Automated detection of strings such as 'ignore previous instructions' and 'you are now' in HEARTBEAT.md, SKILL-v2.3-backup.md, and references/security-patterns.md are false positives. These files utilize the phrases as defensive patterns for the agent to identify and block within external data, not as instructions to override its own behavior.
[COMMAND_EXECUTION]: The skill includes local shell scripts (scripts/security-audit.sh, scripts/wal_hook.sh) used for security auditing and implementing the 'Write-Ahead Logging' (WAL) protocol. It also references standard CLI tools like grep, tail, and cron for workspace management. No malicious or remote command execution patterns were found.
[SAFE]: The skill incorporates extensive security documentation and hardening rules, specifically advising against implementing instructions from external sources, connecting to untrusted AI networks, and leaking context to shared channels.

proactive-agent