The Agent Skills Directory

[SAFE]: The skill provides cross-platform tool mapping and meta-instructions for tool-calling consistency across Claude Code, Copilot CLI, Codex, and Gemini CLI environments.
[PROMPT_INJECTION]: The instructions employ aggressive and imperative language ("ABSOLUTELY MUST", "not negotiable", "not optional") and claim that skills "override default system behavior." These are characteristic patterns of prompt injection designed to bypass system constraints. However, the risk is mitigated by an explicit instruction priority that keeps user-provided instructions at the highest priority level, ensuring the user remains in control.
[PROMPT_INJECTION]: The skill defines a system for processing external skill files, creating an attack surface for indirect prompt injection. Ingestion points: Untrusted content enters via the Skill tool as described in SKILL.md. Boundary markers: The mapping documentation in references/codex-tools.md suggests using XML tags for agent instructions to delimit content. Capability inventory: The framework maps high-privilege tools including Bash, Write, Edit, and Task (subagent dispatch) across platforms. Sanitization: No explicit sanitization or validation of the external skill content is described in the provided files.

using-superpowers