cogny
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted task descriptions and playbooks from an external queue, instructing the agent to follow them step-by-step, which creates an indirect prompt injection surface.
- Ingestion points: External data enters the agent context via the
mcp__cogny__get_next_tasktool, specifically through theskill_contentfield containing the analysis playbook (SKILL.md). - Boundary markers: Absent. The prompt lacks explicit delimiters or structural encapsulation to isolate the untrusted
skill_contenttext from the main system guidelines. - Capability inventory: The agent has a broad set of capabilities, including powerful local tools (
Bash,Read,Write,WebFetch) and ad platform integration tools (mcp__cogny__linkedin_ads__*,mcp__cogny__google_ads__*, etc.). - Sanitization: Absent. The instruction set contains no filtering, verification, or sanitization constraints to prevent the agent from obeying rogue instructions embedded within the task payload.
Audit Metadata