The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions by ingesting and processing various forms of external, potentially untrusted data, creating a surface for indirect prompt injection.
Ingestion points: The skill reads repository instructions, project requirements, specifications, and diagnostic outputs such as logs, network traces, and error messages (referenced in SKILL.md, implement.md, and diagnose.md).
Boundary markers: Instructions rely on document-level separation but do not mandate explicit prompt-level delimiters or 'ignore' directives when interpolating external content into the agent's context.
Capability inventory: The skill possesses extensive capabilities, including reading/writing to the local filesystem (for code and project records), executing subprocesses (for running tests and harnesses), and performing network operations (via remote tracker updates and subagent coordination).
Sanitization: The skill includes a dedicated 'Redact' phase in diagnose.md that explicitly requires the agent to replace secrets and sensitive headers in command outputs and captured artifacts with <REDACTED> before processing them.

iterate