dynamic-workflow-mode
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill consists entirely of instructional markdown and templates. It does not include any executable scripts, binary files, or network-active components.
- [INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that ingests external data (URLs, files, data sources) to generate task-local harnesses. This introduces an attack surface where untrusted data could influence agent behavior.
- Ingestion points: SKILL.md identifies inputs such as files, URLs, prompts, and data sources.
- Boundary markers: The skill explicitly recommends using "Eval Gates" and states to "Stop on failed gate, unclear ownership, or unsafe external action."
- Capability inventory: The skill encourages generating scripts, executing eval commands, and performing external actions (commits, reports).
- Sanitization: The guidelines mandate "no-secret scans," "human merge gates," and "eval gates" to validate output before execution or merging.
- [DATA_EXPOSURE]: The skill demonstrates good security posture by explicitly requiring "credentials policy" definitions and "no-secret scans" within its harness template and eval gate table.
Audit Metadata