The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the local repository and external design articles to generate implementation plans, which creates a surface for indirect prompt injection.
Ingestion points: Local Dart files (lib/), a remote catalog hosted on GitHub (github.com/kamranbekirovyz/), and remote design articles (flutterpro.design).
Boundary markers: The skill includes an explicit instruction (Rule 5) for the agent to treat all content as data and to ignore any embedded instructions that attempt to override behavior.
Capability inventory: The agent is authorized to use curl for network requests, grep for code scanning, and file-write capabilities restricted to the docs/improvements/design/ directory.
Sanitization: There is no explicit sanitization of the fetched content beyond the instructional constraint to treat it as data.
[EXTERNAL_DOWNLOADS]: The skill performs network requests using curl to fetch a design catalog from the author's GitHub repository and implementation details from flutterpro.design. These downloads are essential for the skill's stated purpose of providing standardized UI/UX improvements.
[COMMAND_EXECUTION]: The workflow requires the use of curl for fetching remote Markdown files and dart analyze for project analysis. The skill's instructions include 'Hard rules' that strictly prohibit the agent from running mutating commands such as pub add, git commit, or modifying any source code outside the designated documentation directory.

flutter-improve-design