The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as git logs and project files, which could contain malicious instructions designed to mislead the diagnostic process.
Ingestion points: Git history (log, status, diff), .planning/ directory artifacts, and general file system state.
Boundary markers: None identified.
Capability inventory: Bash for command execution, Read and Write for report generation.
Sanitization: Includes a strict rule to redact sensitive data like API keys, tokens, and absolute paths from its output.
[SAFE]: The skill explicitly enforces a read-only policy for project source files and requires grounding all findings in evidence, which limits the potential for unauthorized modifications or fabricated reports during the forensic process.

gsd-forensics