The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes a mandatory behavioral directive requiring the agent to announce completion of the guide in its next response, which functions as an override of standard conversational behavior.
[COMMAND_EXECUTION]: The skill provides a bash script skills-search that executes grep and find commands on the local filesystem. The script's logging mechanism uses an echo command with double-quoted variables ($query), which allows for command substitution if the search patterns contain backticks or $() constructs.
[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves searching for, reading, and following instructions in external SKILL.md files. This architecture creates an ingestion surface for potentially untrusted instructions without defined boundary markers or sanitization logic.
Ingestion points: Local files located within the ~/.claude/skills/ directory and its subfolders.
Boundary markers: The skill does not define specific delimiters or 'ignore embedded instructions' warnings for the external content it loads.
Capability inventory: The agent is granted capabilities to execute the skills-search shell script, use the TodoWrite tool for task management, and dispatch subagents for historical context synthesis.
Sanitization: The skill does not implement any validation, filtering, or escaping for the content of the referenced skill files before the agent is instructed to follow them.

Getting Started with Skills