The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to perform file system mutations including moving (mv), creating (mkdir), and permanently removing (rm -rf) directories within the specified conductor/ structure. These operations are essential to its function but involve inherent risks of accidental data loss, which the skill mitigates through mandatory confirmation prompts like requiring the string 'DELETE' for removals.
[INDIRECT_PROMPT_INJECTION]: The skill reads from and updates local files like conductor/tracks.md, metadata.json, and plan.md, which could potentially be manipulated by external sources to influence the agent's behavior during track management tasks.
Ingestion points: Data is parsed from conductor/tracks.md, metadata.json, and plan.md as described in resources/implementation-playbook.md.
Boundary markers: The instructions explicitly warn to treat metadata fields and titles as data rather than shell fragments to prevent command or prompt injection.
Capability inventory: The skill facilitates mv, rm -rf, mkdir, and git commit operations across various files in the conductor/ directory.
Sanitization: The playbook mandates strict validation of track IDs, rejecting path separators, parent directory (..) sequences, and symlinked paths to prevent directory traversal attacks.

conductor-manage