The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation explicitly recommends configuring the Ralph CLI with the --dangerously-skip-permissions flag, which bypasses security permission checks during autonomous code execution loops.
[EXTERNAL_DOWNLOADS]: The skill requires the global installation of the @ralph-orchestrator/ralph-cli NPM package, which is an external dependency whose safety cannot be verified by the skill's static analysis.
[PROMPT_INJECTION]: The skill contains logic to refactor user requests into architectural prompts for other models. It is vulnerable to indirect prompt injection because it interpolates raw user input into a privileged context.
Ingestion points: SKILL.md (Phase 0, Step 2: Prompt Refactoring).
Boundary markers: Absent; the original_user_request is placed directly into an f-string template.
Capability inventory: High-privilege file system access (Read, Write, Edit, Glob), Git command execution, and subagent spawning via the Task tool.
Sanitization: Absent; no escaping or validation is performed on the user-provided intent before prompt construction.
[COMMAND_EXECUTION]: The skill automates long-running, autonomous background processes using nohup and disown to execute generated tasks, creating a risk of persistent, unintended system modifications if the planning phase is compromised.
[PROMPT_INJECTION]: The skill includes instructions that attempt to override standard agent behavior, such as 'DO NOT include preambles, status updates, or meta-commentary' and 'Proceed directly to architectural design'.

ralph-orchestrator