recoup-internal-consulting-followup-sequencer
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted client-provided data, specifically 'stakes in their words,' to personalize follow-up email templates. This creates a vulnerability surface where instructions embedded in client input could potentially influence the agent's behavior during the drafting process.
- Ingestion points: Deal specifics and client stakes (SKILL.md).
- Boundary markers: None specified for the interpolated data.
- Capability inventory: Local script execution (create_draft.py) for email staging and file system operations.
- Sanitization: No explicit sanitization or validation of the ingested client data is mentioned.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local script
integrations/gmail/_work/create_draft.pyto stage emails in Gmail. The instruction includes a safety requirement for a typed confirmation and an explicit flag (--send) before any email is actually transmitted, which provides a control mechanism for this capability.
Audit Metadata