launch-subagent

Pass

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill uses strong imperative language ("DO NOT", "NEVER EVER", "only ever use") to override default agent behavior regarding model selection and sub-agent orchestration.
  • [PROMPT_INJECTION]: The instructions mandate the use of specific, seemingly non-existent model names ("Fable 5 Max", "GPT 5.6 Sol Max Fast"). Forcing an agent to use invalid model identifiers can result in tool call failures or incorrect routing during sub-agent initialization.
  • [DATA_EXPOSURE]: The skill instructs the agent to pass specific skills like deepapi to sub-agents for search and scraping. While not inherently malicious, it directs the agent's data-gathering capabilities toward specific external tools based on the instructions provided in the brief.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 27, 2026, 03:51 AM
Security Audit — agent-trust-hub — launch-subagent