The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it interpolates arbitrary file content directly into sub-agent prompts without robust sanitization or delimiters that instruct the agent to ignore embedded instructions.
Ingestion points: The delegate.js script reads file contents from user-specified paths provided via the --files and --context flags.
Boundary markers: While the script uses markdown headers and code fences to wrap file content, it does not include explicit warnings to the sub-agent to disregard instructions found within those files.
Capability inventory: Sub-agents can process complex logic and potentially initiate further delegation or tool calls within their sessions.
Sanitization: No escaping, validation, or sanitization of the file content is performed before interpolation.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to the official Anthropic API (api.anthropic.com) to process delegated tasks. This is standard functionality for an LLM-based delegation tool and uses well-known technology service endpoints.

agent-delegate