write-blueprint
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface since it dynamically ingests files from the project codebase to explore structure and discover tooling.
- Ingestion points: Ingestion occurs in
SKILL.md(Step 1) where the coordinator and parallel subagents read codebase files, entry points, configuration data, and CI pipelines. - Boundary markers: Absent. There are no explicit boundaries or delimiters defined to instruct the subagents to ignore potential natural language prompt injections embedded inside source code comments or configurations.
- Capability inventory: The skill writes plan artifacts to the filesystem (
docs/plans/), dispatches subagents via Claude Code's Agent tool (SKILL.md), and executesgit committo persist files. - Sanitization: Absent. Codebase contents and tool definitions are analyzed directly without sanitization or escaping mechanisms.
Audit Metadata