recoup-release-plan-rollout
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill fetches artist data, demographics, and marketing insights from external API endpoints (api.recoupable.dev). This data is incorporated into the master RELEASE.md and creative briefs without explicit boundary markers or sanitization logic, creating a surface where malicious content in the research data could influence the agent's behavior during the plan or doc stages.
- Ingestion points: Data retrieved from /api/research/* endpoints (profile, metrics, insights, audience) via curl commands in references/endpoints.md.
- Boundary markers: Absent; the agent is instructed to synthesis results directly into markdown files.
- Capability inventory: File write operations to releases/ workspace; execution of deterministic Python scripts for timeline generation and validation.
- Sanitization: No explicit filtering or escaping instructions for data returned from the research API.
- [DATA_EXFILTRATION]: The skill manages a workspace containing highly sensitive information, including team contacts (names, emails, phones in Section 15 of references/release-template.md), detailed campaign budgets (Section 16), and master audio asset links (Section 2.3). While instructions for generating 'Shareable Exports' include scrubbing sections marked [INTERNAL], the risk remains that the agent could inadvertently expose these details if instructions are bypassed or misinterpreted.
Audit Metadata