The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs strong, coercive language in SKILL.md to force compliance and override the agent's default decision-making process. Phrases such as 'YOU DO NOT HAVE A CHOICE', 'This is not negotiable', and 'You cannot rationalize your way out of this' are designed to suppress agent autonomy.
[PROMPT_INJECTION]: The skill includes a 'Red Flags' section that explicitly instructs the agent to ignore its own internal reasoning (e.g., 'I need more context first', 'The skill is overkill') in favor of mandatory skill invocation.
[PROMPT_INJECTION]: The skill identifies a significant indirect prompt injection surface by instructing the agent to load and prioritize behavior-modifying instructions from various external files and directories.
Ingestion points: Project-level instruction files (CLAUDE.md, AGENTS.md, GEMINI.md, TODO.md) and user-local skill directories (~/.claude/skills/, ~/.agents/skills/, ~/.copilot/skills/).
Boundary markers: None identified in the provided instructions; the skill explicitly states these files 'override default behavior'.
Capability inventory: The reference files map these instructions to powerful capabilities including shell execution (Bash, run_shell_command), file modification (Edit, apply_patch), and subagent dispatch (Agent, invoke_agent).
Sanitization: No sanitization or verification protocols are defined for the instructions ingested from these sources.

using-superpowers