accelint-onboard-openspec
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the codebase to automate the generation of
openspec/config.yamland pre-fill interview answers. - Ingestion points: The agent reads ARCHITECTURE.md and README.md in Phase 0, and scans numerous configuration files (package.json, tsconfig.json, vite.config.*, etc.) across the directory tree during Phase 3 codebase inference.
- Boundary markers: The instructions lack specific delimiters or instructions for the agent to disregard potential instructions embedded within the codebase artifacts it analyzes.
- Capability inventory: The skill utilizes parallel discovery agents to traverse the filesystem, read multiple files, and write the final configuration file to the project directory.
- Sanitization: The skill performs technical validation of YAML syntax (checking for tabs, quoting special characters, and indentation) and requires a human preview/confirmation before writing, but it does not explicitly sanitize the semantic content extracted from project files before processing it.
Audit Metadata