The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted GitHub PR and issue comments to build the 'Clone' model. An attacker could potentially influence the generated agent's behavior by submitting specifically crafted comments in a repository being analyzed.\n
Ingestion points: scripts/collect-review-activity.py fetches comments from GitHub repository endpoints.\n
Boundary markers: The skill instructions do not specify sanitization or boundary markers for the ingested external text.\n
Capability inventory: The agent has the capability to write local files, including new agent skill definitions (SKILL.md) and behavioral models (MODEL.md, VOICE.md).\n
Sanitization: No explicit sanitization of the comment bodies is performed before processing.\n- [COMMAND_EXECUTION]: The skill relies on local command execution to interact with the repository and GitHub.\n
The scripts/collect-review-activity.py script uses subprocess.run to execute the GitHub CLI (gh) for data collection.\n
The workflow.md instructions guide the agent to use git commands (git log, git blame, git shortlog) to analyze repository ownership and history.\n- [DYNAMIC_EXECUTION]: The skill generates a new AI agent skill and supporting configuration files at the end of the training process.\n
The training process results in the creation of a new SKILL.md, MODEL.md, and VOICE.md in the ~/.agents/skills/cr-clone-<name>/ directory.\n- [PERSISTENCE]: The skill establishes persistence by creating a new, independent agent skill directory on the host system.\n
This allows the generated 'Clone' agent to be invoked across different sessions independently of the trainer skill.

cr-clone-trainer