ccf-experiment-designer

Pass

Audited by Gen Agent Trust Hub on Aug 13, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from the user, specifically 'ideas', 'drafts', and 'supplied real results' as noted in SKILL.md. This represents a potential attack surface where an adversary could embed malicious instructions within a paper draft to influence the agent's behavior or subsequent tool calls. The provided instructions do not include explicit boundary markers or sanitization steps to isolate this external data from the core system prompt.
  • [SAFE]: The skill uses localized file references (e.g., ../ccf-humanization/references/) to maintain consistency across a suite of related tools, which is a standard pattern for modular agent skills. No instances of obfuscation, hardcoded credentials, unauthorized remote code execution, or privilege escalation patterns were identified.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 13, 2026, 08:26 AM
Security Audit — agent-trust-hub — ccf-experiment-designer