The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and verify output from various CLI tools, which could potentially be manipulated by malicious code within the project being analyzed.
Ingestion points: Command output from flutter analyze, flutter test, and flutter build (SKILL.md).
Boundary markers: None; the agent relies on raw terminal output to validate its own work.
Capability inventory: Command execution via flutter, dart, node, and npx (SKILL.md).
Sanitization: No explicit instructions for sanitizing or escaping tool output before processing.
[DYNAMIC_EXECUTION]: The skill executes a local script (capture.cjs) using a computed path to a plugin directory.
Evidence: node <design-polish-path>/scripts/capture.cjs --wcag / (SKILL.md).
[COMMAND_EXECUTION]: The skill manages a local web server by starting a background process and later terminating it using a process ID.
Evidence: npx serve build/web -l 3000 & and kill $SERVER_PID (SKILL.md).

flutter-verification