The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted repository data, including roadmaps, commits, and project files, which creates an attack surface for indirect prompt injection. Malicious instructions embedded in the analyzed code or documentation could potentially influence the agent's behavior during the facilitation process.\n
Ingestion points: Repository files analyzed during the 'Forensic Retrospective' phase (SKILL.md Step 1 and references/retreat-protocol.md).\n
Boundary markers: No delimiters or warnings are used to differentiate ingested repository data from the agent's core instructions.\n
Capability inventory: The skill is authorized to use bash, view_file, write_to_file, and replace_file_content to manage the project environment.\n
Sanitization: There is no evidence of sanitization or filtering of the content read from repository files.\n- [COMMAND_EXECUTION]: The skill uses bash to execute shell commands such as git log for churn analysis and knip (via bunx) for identifying dead code. It also includes instructions to uninstall unused packages, which involves destructive command execution as part of its debt-purge functionality.\n- [DYNAMIC_EXECUTION]: The skill includes a script scripts/wisdom-updater.ts that imports and runs code from a relative path (../../coach/scripts/wisdom-updater) outside its own directory. This creates a functional dependency on the 'coach' skill and involves executing code from another location in the file system.

periodic-retreat