plan

Pass

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: LOWPROMPT_INJECTION
Full Analysis
  • Indirect Prompt Injection (LOW): The skill is designed to ingest and process untrusted data from the repository (e.g., README.md, docs/) to generate plan content. This creates a surface where embedded instructions could influence the agent's behavior. \n
  • Ingestion points: The skill instructions in SKILL.md prompt the agent to read README.md, docs/, CONTRIBUTING.md, and other files for context. \n
  • Boundary markers: No explicit delimiters are used to separate untrusted repository content from the agent's instructions. \n
  • Capability inventory: The skill can read, write, and delete files within the ~/.codex/plans directory. \n
  • Sanitization: The skill implements a strict regex (^[a-z0-9]+(-[a-z0-9]+)*$) for filenames to prevent path traversal and ensures metadata is single-line to prevent YAML injection. \n- Data Exposure (LOW): The create_plan.py script allows reading any local file through the --body-file argument. While restricted to the agent's permissions, a prompt injection attack could trick the agent into copying sensitive configuration files or keys into a plan file within the localized plans directory.
Audit Metadata
Risk Level
LOW
Analyzed
Feb 16, 2026, 07:38 AM