prose-flow-improver
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied documents to improve prose, creating a surface for indirect prompt injection where malicious instructions embedded in the input text could attempt to influence the agent's behavior during the multi-pass revision workflow.
- Ingestion points: As described in
SKILL.md, the skill ingests user prose for diagnosis and rewriting in the "Rewrite Workflow" and "Agent Orchestration" sections. - Boundary markers: The instructions define a "Parent Shape Brief" and specific "Worker Prompt Requirements" to constrain sub-agents, though these are instructional rather than technical delimiters.
- Capability inventory: The skill orchestrates multiple sub-agents (workers) to perform chunk-level revisions and potentially write to draft output files.
- Sanitization: Safety relies on the "Parent-Only Merge and Rhythm Audit" step, where the parent agent is instructed to manually verify that no constraints, facts, or commands were dropped or altered during the process.
Audit Metadata