coo-advisor
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill utilizes several data ingestion points that create a potential surface for indirect prompt injection attacks.
- Ingestion points: The agent is instructed to read from a platform-specific context file (
~/.claude/company-context.md) and process various external JSON files (kpi_data.json,processes.json, andcapacity_data.json) to perform its analysis. - Boundary markers: The instructions do not define clear delimiters or specify that the agent should ignore instructions embedded within the data, which could lead to the agent following malicious commands hidden in the ingested business metrics.
- Capability inventory: The skill includes and executes three local Python scripts (
capacity_modeler.py,operational_kpi_tracker.py,process_efficiency_scorer.py) that facilitate the processing and analysis of this external data. - Sanitization: There are no evident validation or sanitization routines for the natural language content found within the input files before they are presented to the agent for decision-making.
Audit Metadata