The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for delegating tasks to specialized subagents that process data from potentially untrusted sources, such as web searches and internal codebases.
Ingestion points: External data is ingested via the librarian agent using tools like webfetch and Web Search, while internal data is processed by the explore agent searching the codebase. Evidence found in references/subagent-types.md.
Boundary markers: The skill mandates a 7-section delegation structure to define task scope, though it does not explicitly specify data-level delimiters. Evidence found in SKILL.md and references/delegation-patterns.md.
Capability inventory: Subagents are granted access to tools such as Grep, Read, Glob, webfetch, and the GitHub CLI. Evidence found in references/delegation-patterns.md.
Sanitization: The documentation provides explicit guidance on verifying delegated output and using tool whitelists to minimize risk. Evidence found in SKILL.md and references/common-pitfalls.md.
[COMMAND_EXECUTION]: The instructions and templates include the use of shell-based tools for code analysis and file system navigation.
Evidence: Examples in references/common-pitfalls.md and references/delegation-patterns.md demonstrate the use of grep, glob, and ast_grep_search within subagent prompts.

subagents-creator