The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs extremely strong, imperative language ("ABSOLUTELY MUST", "not negotiable", "not optional") to override default AI behavior and enforce a "skills-first" workflow. This represents a meta-instructional injection intended to govern how the agent processes all subsequent tasks. A critical safety mitigation is present: the skill explicitly defines a priority hierarchy where user instructions always take precedence over the skill's own rules.
[COMMAND_EXECUTION]: The documentation provides explicit mappings for shell execution and subagent tools across multiple platforms (Claude Code, Copilot CLI, Gemini CLI, and Codex). It references tools like Bash, run_shell_command, and spawn_agent. The skill includes environment detection logic using read-only git commands (git rev-parse) to determine if the agent is operating within a linked worktree or a detached HEAD state.
[EXTERNAL_DOWNLOADS]: The skill defines tool equivalents for network-enabled operations, including WebFetch, WebSearch, and web_fetch. While it documents these capabilities for use in the agent's workflow, it does not specify any automated remote code execution or unauthorized external resource loading within its own instructions.

using-superpowers