ralph
Warn
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPERSISTENCE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides installation instructions that download extensions and templates from untrusted GitHub repositories, specifically 'github.com/Q00/ouroboros' and 'github.com/supercent-io/skills-template'. It also utilizes 'npx' to fetch and execute external code during the installation process.
- [COMMAND_EXECUTION]: The provided 'scripts/setup-codex-hook.sh' script executes shell commands and 'python3' logic to programmatically modify the user's local '~/.codex/config.toml' file and write new prompt files to the filesystem.
- [PERSISTENCE]: The skill establishes behavioral persistence by injecting instructions into the AI agent's core configuration ('developer_instructions' in config.toml). Furthermore, it implements an autonomous execution loop ('ralph mode') that is designed to persist across agent turns until specific verification criteria are met.
- [NO_CODE]: The Claude Code hook configuration references several functional scripts, including 'session-start.mjs', 'keyword-detector.mjs', and 'drift-monitor.mjs', which are missing from the provided skill package and cannot be verified.
Audit Metadata