The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs extremely forceful language ("ABSOLUTELY MUST", "not negotiable", "not optional") and instructs the agent to ignore its own internal reasoning ("rationalizing") that might suggest skipping a skill. It also uses "Instruction Priority" markers to explicitly override default system behavior. These patterns are characteristic of behavior-shaping instructions but are mitigated by a clearly defined hierarchy that places explicit user instructions at the highest priority level.
[COMMAND_EXECUTION]: The skill and its reference files describe tool mappings across different platforms (Claude Code, Codex, and Gemini CLI) and instruct the agent on when to invoke these tools. However, the files contain no shell scripts, direct command executions, or automated tasks that would execute without explicit model invocation based on user tasks.

using-superpowers