The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses agent configuration files located in the user's home directory at ~/.claude/agents/. This allows it to identify and manage global agents defined at the system level, which is necessary for its stated purpose of auditing and coordinating subagents.
[INDIRECT_PROMPT_INJECTION]: The skill reads and analyzes existing agent definitions from various local and global directories, creating an ingestion surface for data that could influence its decision-making during audits or new agent creation.
Ingestion points: Reads existing agent files from agents/, .claude/agents/, and ~/.claude/agents/.
Boundary markers: The instructions do not specify any markers to delimit ingested file content from the skill's own instructions.
Capability inventory: Includes file write access to create new agent definitions and the ability to invoke the grilling skill for design validation.
Sanitization: No explicit sanitization or filtering of ingested agent content is mentioned.

subagent-creator