The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided content from build-notes.md and claude-prompts.md to generate new instructions. While this creates a potential ingestion point for untrusted data, the skill is designed specifically to translate these notes into structured templates, and no exploitable capabilities or dangerous interpolations were found.
Ingestion points: Reads from workspace/playground/<build>/build-notes.md and workspace/playground/<build>/claude-prompts.md.
Boundary markers: The skill does not use specific delimiters for external content but operates within a highly structured template format.
Capability inventory: Limited to Read and Write tools within the workspace playground directory.
Sanitization: The skill relies on structured output patterns defined in references/prompt-patterns.md to bound the generated content.
[EXTERNAL_DOWNLOADS]: The skill references an external resource for the Tone.js library in its generated prompts.
Evidence: The prompt templates in references/prompt-patterns.md instruct the user to load https://unpkg.com/tone@15.5/build/Tone.js.
Source: The resource is hosted on unpkg.com, which is a well-known and established CDN for NPM packages.

claude-prompt-writer