ea-intro-planner

Pass

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted external data from user-provided papers and abstracts, creating a potential surface for indirect prompt injection.\n
  • Ingestion points: Phase 1 (Source Analysis) accepts PDFs, pasted text, and URLs as input for analysis.\n
  • Boundary markers: Absent. User-provided content is inserted directly into the subagent prompt without explicit isolation or instructions to ignore embedded commands.\n
  • Capability inventory: The skill can write data to intro_workspace.json and spawn subagents to process text.\n
  • Sanitization: Potential risks are mitigated by a strict JSON schema requirement for subagent output, which prevents untrusted data from altering the agent's logic flow.\n- [COMMAND_EXECUTION]: The skill involves writing state information to a local file named intro_workspace.json. This is a standard and benign use of file-system capabilities for maintaining session context across different phases of the workflow.\n- [NO_CODE]: The skill is composed entirely of instructional markdown and does not package any scripts or binary executables.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 21, 2026, 07:38 AM
Security Audit — agent-trust-hub — ea-intro-planner