build
Pass
Audited by Gen Agent Trust Hub on Jun 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill detects and executes local project scripts defined in
package.json(such astest,lint, andtypecheck). This is used as a feedback loop to verify that implementation phases are completed correctly within the project's own environment. - [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it reads and follows instructions found in plan files (
.specs/plans/<slug>.md). If these plan files contain malicious instructions, they could potentially influence the agent's behavior during the code implementation phase. - Ingestion points:
.specs/plans/<slug>.md(markdown plan files) - Boundary markers: None explicitly defined for the ingested data content.
- Capability inventory: File system read/write, git command execution, and execution of local scripts via
package.json. - Sanitization: The skill does not explicitly sanitize the natural language instructions found within the plan files before processing them.
Audit Metadata