pstack-codex

Pass

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The SKILL.md file contains instructions for the user to manually execute a Python installer script (poteto-session-mode.py install) to set up the compatibility layer.
  • [PERSISTENCE]: The installation script modifies the agent platform's local configuration file (~/.codex/hooks.json) to register a command that executes on every prompt submission (UserPromptSubmit). This ensures the adapter's logic persists across conversation sessions and is automatically loaded by the platform.
  • [INDIRECT_PROMPT_INJECTION]: The skill implements a runtime context injection mechanism that acts as an indirect prompt injection surface.
  • Ingestion points: The hook script reads the user's prompt from standard input (provided by the Codex platform) to detect activation commands like /poteto-mode.
  • Boundary markers: The injected context (e.g., 'Poteto Mode is active...') is appended to the prompt context without specific delimiters or safety warnings to distinguish it from user instructions.
  • Capability inventory: The hook script has file system access to read and write session state in ~/.codex/pstack-codex/session-modes/ and modifies platform configuration files.
  • Sanitization: The script uses shlex.quote to sanitize paths during command registration and validates session IDs using the uuid module.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 6, 2026, 11:51 PM