The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the uv tool to execute local Python scripts (transcript-miner.py and transcript-inspect.py) for analyzing project transcripts. These scripts are bundled with the skill and run within the agent's local environment to discover and summarize session history.
[INDIRECT_PROMPT_INJECTION]: As the skill processes content from past transcripts, it interacts with potentially untrusted data. This surface is well-mitigated by the implementation of 'bounded retrieval' strategies (limiting session counts and excerpt sizes) and a strict redaction framework in transcript_common.py that sanitizes external data before it is presented to the agent context.

agents-introspection