The Agent Skills Directory

[PROMPT_INJECTION]: The 'Always-on mode' section provides a block of instructional text containing strong imperative markers (e.g., 'MANDATORY', 'MUST', 'Do NOT skip') intended for the user to add to the agent's global configuration file. These instructions are designed to override the agent's default decision-making process regarding tool usage and enforce the skill's procedure.
[PERSISTENCE]: The install.sh script establishes a persistence mechanism by creating symbolic links within various agent skill directories (e.g., ~/.claude/skills/, ~/.codex/skills/, ~/.agents/skills/). Furthermore, the documentation provides specific instructions to modify global agent instruction files (CLAUDE.md, AGENTS.md, etc.) to ensure the skill remains active and prioritized across all agent sessions.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a broad ingestion surface as it is instructed to sweep and read the SKILL.md frontmatter and descriptions from all installed skills in the system. This allows instructions embedded in other, potentially untrusted, skills to influence the agent's execution roster.
Ingestion points: Reads metadata from ~/.claude/skills/, ~/.codex/skills/, and project-level .claude/skills/ directories.
Boundary markers: No specific boundary markers or 'ignore' instructions are used when interpolating ingested skill descriptions into the selection process.
Capability inventory: The skill has the capability to read any file within the skill directories and can trigger the loading of multiple other skills via the agent's native tools.
Sanitization: The content read from the inventory is not sanitized before being processed by the agent to determine the skill roster.

use