The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/role-x.py utilizes the Python subprocess module to execute Git commands such as git rev-parse and git diff. These commands are used to identify the current branch and recently modified files, which serve as signals for the lens selection algorithm. These operations are restricted to the local workspace and are used for legitimate context-gathering purposes.
[DATA_EXFILTRATION]: The skill implements a telemetry capture system that records routing decisions and metadata about user prompts (like word counts and SHA256 digests) to a local file at ~/.config/broomva/role/events.jsonl. An optional 'sanitized prompt capture' feature can be enabled by the user to log specific keywords. This data is stored locally and used by the suggest command to help the user identify patterns for new lens creation; no data is transmitted to external servers.
[PROMPT_INJECTION]: The skill is designed to manage agent instructions via 'lenses'. While it dynamically injects content from roles/*.md files into the agent's working context, it employs several security measures: it specifically discourages 'persona theater' in favor of substantive context, it uses an internal _safe_inline function to sanitize strings (stripping control characters and square brackets) before they are included in the agent's prompt, and it clearly labels the injected blocks to maintain context boundaries.

role-x