init-claude
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes multiple untrusted data sources to generate an agent operating manual, creating a surface for potential instruction injection.
- Ingestion points: The skill reads from a wide range of project documentation files, including those in
docs/technical-specs/,docs/coding-standard/,docs/task-breakdown/,docs/deployment-plan/, anddocs/business/(SKILL.md). - Boundary markers: The skill does not employ explicit delimiters or instructions to ignore potentially malicious content embedded within these source documents during the distillation process.
- Capability inventory: The skill performs extensive file system read operations across the project structure and writes the resulting synthesized manual (
CLAUDE.mdorAGENTS.md) to the repository root. - Sanitization: The skill distills content based on "real values" found in the docs but lacks specific sanitization or validation mechanisms to prevent the inclusion of malicious instructions from the source material into the final operating manual.
Audit Metadata