The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled Python script (scripts/mine_sessions.py) to extract and process session data. This script is part of the skill's package and operates only on local data.
[INDIRECT_PROMPT_INJECTION]: The skill processes historical chat logs (events.jsonl and session-store.db) which may contain untrusted content from previous agent interactions. * Ingestion points: The skill reads session logs from the local directory ~/.copilot/session-state/ and the SQLite database ~/.copilot/session-store.db. * Boundary markers: It utilizes internal event metadata (such as type == 'user.message') and regex-based filtering to distinguish user-typed input from machine-generated traffic. * Capability inventory: The skill executes local Python scripts and performs read/write operations on the local filesystem within the ~/.copilot directory. * Sanitization: It implements an extensive redact function that masks secrets, API keys (e.g., AWS, GitHub, Slack tokens), emails, and URLs using regular expressions before data is used for analysis or displayed in the chat interface.

prompt-mirror