recoup-internal-consulting-lead-temperature
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes a local Python script located at
integrations/granola/_work/pull_transcript.pyto retrieve verbatim meeting transcripts from Granola.\n- [EXTERNAL_DOWNLOADS]: Fetches data from Granola's public API (public-api.granola.ai) to obtain transcription details and verbatim call records for analysis.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from Gmail threads and Granola call transcripts, which could potentially contain malicious instructions designed to manipulate the lead scoring process.\n - Ingestion points: The skill ingests Gmail email bodies and Granola verbatim transcripts from the local
integrations/directory and external APIs into the agent's context (SKILL.md, Step 1).\n - Boundary markers: The instructions do not define explicit boundary markers or delimiters to isolate untrusted email and transcript content from the skill's logic.\n
- Capability inventory: The skill possesses the capability to write records to the
AGENTS.mdfile, sync computed scores to Attio via its API, perform network operations via the Granola API, and execute local Python scripts.\n - Sanitization: The instructions emphasize manual citation and verification of primary sources but do not provide technical sanitization, filtering, or escaping for the external text being analyzed.
Audit Metadata