The Agent Skills Directory

[SAFE]: The skill implements a mandatory 'Sensitive-data boundary' that explicitly prohibits reading, echoing, or summarizing secrets, credentials (.env, .key, .pem), and environment variables.
[COMMAND_EXECUTION]: Utilizes dynamic context injection (using the ! syntax) to gather benign repository metadata such as recent commits and working tree status at load time for orientation.
[SAFE]: Data access is strictly scoped to the current project, with specific prohibitions against raw database access or cross-project memory leakage.
[SAFE]: File writing is restricted to a specific, opt-in directory for scout reports, ensuring no unauthorized persistence or modification of the repository occurs.

recall-scout