The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process session history, memory archives, and repository logs to identify repeated workflows. These data sources may contain untrusted content or malicious instructions designed to influence the agent's behavior during the 'close read' or skill authoring steps.
Ingestion points: The agent is instructed to use read_file on session logs, transcripts, and repository history archives.
Boundary markers: There are no defined delimiters or instructions to treat log content as data rather than instructions during the manual 'close read' phase of the mining procedure.
Capability inventory: The skill utilizes write_file to generate new skill files and the terminal to execute the scanner script.
Sanitization: No explicit sanitization of the log content is performed; the scanning script performs keyword matching on raw text blocks.
[COMMAND_EXECUTION]: The skill executes a local Python script scripts/scan_sessions.py to analyze archive directories. While the script is deterministic and uses only standard library modules, it requires read access to user-specified directories to perform its function.

skill-lifecycle-foundry