The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard system utilities including find, ls, awk, mv, rm, du, gzip, and jq to manage local files. These commands are scoped to the research session directory and are intended for legitimate file maintenance tasks like listing file sizes, moving items to the trash, and compressing large log files.
[INDIRECT_PROMPT_INJECTION]: The skill scans local file systems, which exposes an attack surface where malicious file names could attempt to influence the agent's behavior.
Ingestion points: File names and metadata are ingested via find and ls commands in SKILL.md.
Boundary markers: The instructions implement a mandatory "Analysis" step where the agent must present a cleanup plan to the user, providing a manual review boundary.
Capability inventory: The agent can move or delete files (mv, rm) and execute shell commands to verify file integrity.
Sanitization: Risk is mitigated by a hardcoded PROTECTED_PATTERNS list that explicitly prevents the deletion of critical research data, and by requiring double confirmation from the user before executing any destructive actions.

Cleaning Up Research Sessions