adaptogen-run

Pass

Audited by Gen Agent Trust Hub on Jul 14, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements a workflow that ingests untrusted data from target skills and tasks, interpolating it into prompts for further execution, which constitutes an indirect prompt injection surface.\n
  • Ingestion points: External data is read from target skills and task anchors as described in SKILL.md.\n
  • Boundary markers: No specific delimiters or security instructions are used to isolate the interpolated task content from the subagent's instructions.\n
  • Capability inventory: The skill utilizes the Agent tool to spawn sub-processes and the Skill tool to transition state. In 'leaf-agent mode', the agent executes the target procedure directly in its own context, which bypasses subagent isolation.\n
  • Sanitization: The instructions explicitly require information to be 'pasted into the prompt directly' without escaping or filtering of the external content.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 14, 2026, 12:03 AM
Security Audit — agent-trust-hub — adaptogen-run