The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Script scripts/install.sh downloads a remote shell script from https://plannotator.ai/install.sh and executes it directly by piping the curl output into bash. This pattern executes unverified remote code with the user's local privileges.
[COMMAND_EXECUTION]: Setup scripts (setup-claude.sh, setup-codex.sh, setup-gemini.sh, setup-opencode.sh) perform automated modifications to local configuration files for various AI tools (e.g., ~/.claude/settings.json, ~/.codex/config.toml) and install persistent executable hooks (~/.codex/hooks/omu-notify.py, ~/.gemini/hooks/omu-plannotator.sh).
[EXTERNAL_DOWNLOADS]: scripts/install.sh performs global installations of NPM packages (agent-browser, playwriter) and uses bunx or npx to fetch and run setup code from external sources like oh-my-opencode and oh-my-ag without version pinning.
[PROMPT_INJECTION]: The skill architecture is susceptible to indirect prompt injection via the plan.md file processed by the planning loop. Ingestion point: plan.md. Boundary markers: Absent. Capability inventory: High, including shell execution and configuration modification. Sanitization: Absent. Instructions embedded in the plan could manipulate the agent's logic during execution transitions.

omu