claude-project-manifest

Pass

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and summarization of project files, creating an attack surface where malicious instructions embedded in those files could influence the agent during manifest generation.
  • Ingestion points: Project files are read and processed to generate the manifest.yaml (documented in SKILL.md and references/workflow-integration.md).
  • Boundary markers: None explicitly defined for the generated summaries of file content.
  • Capability inventory: The provided scripts perform file reads and writes (references/workflow-integration.md).
  • Sanitization: No specific natural language sanitization is mentioned for the content extracted from project files.
  • [EXTERNAL_DOWNLOADS]: The skill refers to standard GitHub actions (actions/checkout, actions/setup-python) and the PyPI package 'pyyaml' in its workflow integration guides as part of suggested development workflows.
  • [SAFE]: The skill demonstrates security awareness by explicitly instructing users to avoid storing secrets in the manifest files (e.g., in assets/templates/manifest.yaml and references/annotation-guidelines.md).
  • [SAFE]: Utility scripts provided in references/workflow-integration.md use yaml.safe_load(), preventing potential code execution vulnerabilities associated with untrusted YAML parsing.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 6, 2026, 04:22 PM