The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses imperative, high-pressure language to override the agent's internal reasoning and default behavior.
Evidence in SKILL.md: 'YOU ABSOLUTELY MUST invoke the skill', 'This is not negotiable. This is not optional. You cannot rationalize your way out of this.'
The 'Red Flags' section explicitly instructs the agent to ignore its own internal logic or justifications for not using a tool (e.g., 'This is just a simple question', 'I need more context first', 'Let me explore the codebase first').
It forces a specific sequence of operations ('before ANY response including clarifying questions') which bypasses the agent's autonomy in determining the best response strategy.
[INDIRECT_PROMPT_INJECTION]: The skill defines a behavior hierarchy that prioritizes instructions from external workspace files, which could contain malicious directives if the environment is compromised.
Ingestion points: CLAUDE.md, GEMINI.md, and AGENTS.md are identified as high-priority instruction sources in SKILL.md.
Boundary markers: No explicit delimiters or warnings to ignore embedded instructions are provided for these ingested files.
Capability inventory: The skill establishes tool mappings for file system access (Write, Edit), shell execution (Bash), and subagent dispatch (Task) across the reference files (references/codex-tools.md, references/copilot-tools.md, references/gemini-tools.md).
Sanitization: No sanitization, validation, or filtering of the content within these external files is described.

using-superpowers