The Agent Skills Directory

[PROMPT_INJECTION]: The skill's workflow involves ingesting data from external sources (via WebSearch) and local codebase files to generate instructions and personas for new subagents. This creates a potential surface for indirect prompt injection where untrusted external content could influence the behavioral guidelines of the generated agents.
Ingestion points: Data enters the agent's context through the WebSearch tool and by reading local project files (e.g., CLAUDE.md, README.md).
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings when processing the retrieved external data.
Capability inventory: The skill is designed to write new agent definition files to the .claude/agents/ directory, which are subsequently loaded into the agent's execution environment.
Sanitization: There are no explicit instructions to sanitize, escape, or validate the content retrieved from external searches before it is synthesized into subagent personas.

writing-agents