docs-ai-prd
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external data described as "evidence, constraints, and dependencies" to generate PRDs and specifications. This creates a surface where malicious instructions embedded in user-supplied evidence could potentially influence the agent's output, although this is a standard risk for document-generation skills.
- Ingestion points:
SKILL.mdworkflow step 2 ("Gather evidence, constraints, and dependencies"). - Boundary markers: The skill recommends using explicit "must NOT" lists and binary acceptance criteria, though it lacks specific prompts to ignore instructions within ingested evidence.
- Capability inventory: File writing (via agent), network operations (via
validate_sources.py), and command execution. - Sanitization: None explicitly defined for external evidence processing.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute local Python scripts for maintenance tasks.
- Evidence:
SKILL.mdandreferences/operational-guide.mdmention runningpython3 scripts/validate_sources.pyandpython3 scripts/append_learning.py(referenced but not provided). - [EXTERNAL_DOWNLOADS]: The
scripts/validate_sources.pyutility performs network operations to verify the status of URLs referenced in the project's documentation. - Evidence:
scripts/validate_sources.pyusesurllib.requestto sendHEADandGETrequests to external domains. The target URLs identified indata/sources.jsonprimarily belong to trusted organizations or well-known services (Anthropic, GitHub, NIST, ISO, etc.), which is considered safe under the trust scope rules.
Audit Metadata