recoup-internal-consulting-account-reality
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from external sources (Gmail threads, Slack messages, and meeting transcripts) which could contain malicious instructions designed to influence the agent's behavior.
- Ingestion points: The agent loads data from
integrations/gmail/threads/,integrations/slack/AGENTS.md, andintegrations/granola/transcript files. - Boundary markers: The skill provides a specific defense by instructing subagents to "cite source + quote; do NOT interpret" and to verify figures against "verbatim transcripts."
- Capability inventory: The skill has read access to sensitive integration data and write access to the
AGENTS.mdfiles within the workspace. - Sanitization: The instructions do not specify any technical sanitization or filtering of external content before it is processed by the model, relying instead on the instruction to quote verbatim.
Audit Metadata