The Agent Skills Directory

[SAFE]: The orchestrator skill provides a structured way to manage project knowledge and execution. It coordinates local sub-skills and maintains state using standard markdown files without any identified malicious patterns or safety violations.- [COMMAND_EXECUTION]: The skill uses ls -1 .cockpit/ to detect the project's directory structure. This is a common and safe operation for a file-based management tool.- [DATA_EXFILTRATION]: Although the skill reads project files such as objective.md and plan.md, it does so within the local file system scope. No network capabilities or patterns indicating data exfiltration were found.- [PROMPT_INJECTION]: The skill processes project-specific data and user-provided inbox items, creating a potential surface for indirect prompt injection. However, since this ingestion is limited to the skill's primary function and no dangerous capability chains (like arbitrary command execution from remote sources) are present, it is considered safe.

pk-harness