The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it interpolates untrusted user input into sub-agent task calls.
Ingestion points: The user's prompt is captured via the /best-practices command in SKILL.md.
Boundary markers: User input is enclosed in double quotes within Task tool calls (e.g., Task task-intent-analyzer("[user's prompt]")) but lacks explicit delimiters or instructions for sub-agents to ignore embedded commands.
Capability inventory: Sub-agents like codebase-context-builder have access to file system tools (Read, Glob, Grep) and the Bash tool for exploration.
Sanitization: There is no evidence of input validation or sanitization before the user's prompt is passed to the sub-agents.
[EXTERNAL_DOWNLOADS]: The skill provides links to official documentation and technical resources to guide the user. These include the official Claude Code documentation (code.claude.com), Anthropic's prompt engineering guides (docs.anthropic.com), and a technical reference for prompt patterns on GitHub (github.com/Dicklesworthstone/meta_skill). These are documented as informational resources and do not facilitate the download or execution of malicious external scripts.

best-practices