spec-driven-workflow
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill establishes a multi-phase lifecycle where the agent's behavior is directed by the content of local markdown artifacts.
- Ingestion points: The agent reads and acts upon instructions, requirements, and tasks defined in
workflow/BACKLOG.md,SPEC.md,PLAN.md, andTASKS.md. - Boundary markers: Absent. The instructions and templates do not provide delimiters or warnings to isolate potentially untrusted content within these files from the agent's execution logic.
- Capability inventory: The agent uses its standard toolset (file system operations, terminal access) to execute the steps mapped out in these documents.
- Sanitization: Absent. There is no requirement for the agent to validate or sanitize the content of the workflow files before proceeding with implementation or deletion tasks.
- [COMMAND_EXECUTION]: The skill includes instructions for managing the file system lifecycle, specifically the deletion of directories.
- Evidence: The 'Verification & Cleanup' section in
SKILL.mddirects the agent to 'delete the completed workflow// subdirectory' once the user confirms completion. - Risk: This capability relies on the agent correctly scoping the deletion to the
workflow/directory to prevent accidental data loss or path traversal if a malicious or malformed<change-slug>is processed.
Audit Metadata