The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes bash to run local analysis scripts included in the package, such as find-skill-sessions.py and version-anchor.py. The version-anchor.py script executes git via subprocess.run to resolve skill version history based on hashes of the skill bodies found in session logs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests historical session data which presents an attack surface for instructions embedded in previous conversation logs.
Ingestion points: scripts/find-skill-sessions.py and scripts/version-anchor.py read raw .jsonl session log files from the agent's local storage.
Boundary markers: The skill instructions and scripts search for specific <skill name="..."> injection markers to identify and isolate skill content within the log files.
Capability inventory: The skill performs file reads across session and relay data directories (scripts/find-skill-sessions.py, scripts/relay-enrich.py) and executes local shell commands for indexing and git history resolution (scripts/version-anchor.py).
Sanitization: The scripts use regex to validate skill names ([A-Za-z0-9_-]+) and rely on json.loads to parse structured data from the logs, which provides a degree of validation against malformed content.

skill-call-extract