recoup-internal-consulting-nightly-janitor
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes internal Python scripts to maintain workspace indexes and perform automated quality evaluations.
- Evidence:
python integrations/granola/_work/build_transcript_indexes.pyandpython evals/janitor/score_run.py. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (emails and transcripts) which could contain embedded instructions intended to manipulate the reconciler's behavior.
- Ingestion points: Reads
business/ops/nightly-digests/<date>.mdcontaining raw captured transcripts and emails (Step 0). - Boundary markers: The skill uses 'evidence-gating' to ground actions in primary sources but lacks explicit delimiters or instructions to ignore embedded commands within the ingested text.
- Capability inventory: The skill has permissions to move folders (
pipeline/toclients/), write files (AGENTS.md,_board.md), execute local scripts, and perform Git commits. - Sanitization: Relies on manual human review ('staged for confirmation') for CRM writes and high-risk funnel moves rather than automated input sanitization.
Audit Metadata