The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands using the ao CLI utility (e.g., ao harvest, ao dedup, ao metrics) to perform file system operations, deduplication, and metadata reporting.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of ingesting and promoting untrusted data. Ingestion points: The ao harvest command recursively scans .agents/ directories within user-specified roots like ~/gt/ and ~/projects/ as described in SKILL.md. Boundary markers: There is no evidence of boundary markers, delimiters, or instructions to ignore embedded instructions within the extracted artifacts. Capability inventory: Artifacts are promoted to a global hub (~/.agents/learnings/) which is then intended for injection into future agent sessions via tools like ao inject, potentially propagating malicious instructions. Sanitization: No sanitization or validation of the content of the 'learnings' or 'patterns' is performed beyond a confidence score calculation.

harvest