The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs strong instructional steering in SKILL.md and flutter-design-thinking.md, using directives like 'MANDATORY: Read Reference Files First' and '⛔ DO NOT code until...' to override default agent behaviors and enforce specific coding protocols. While intended for quality control, these patterns function as behavioral overrides.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill utilizes a Python script (scripts/flutter_ui_audit.py) that performs static analysis on local source code. This creates a potential vector where malicious instructions embedded in a project's Dart files could be reflected in the audit results to influence the agent.
Ingestion points: The read_dart_files function in scripts/flutter_ui_audit.py recursively reads all .dart files within the target project directory.
Boundary markers: Audit findings are output as plain text without delimiters or instructions to the agent to treat the reflected code segments as untrusted data.
Capability inventory: The agent's execution environment allows the use of Bash, Read, Glob, and Grep tools as specified in the skill frontmatter.
Sanitization: The audit script does not implement sanitization or escaping of the content it extracts from analyzed files before printing it to the console.
[COMMAND_EXECUTION]: The skill includes a local Python script (scripts/flutter_ui_audit.py) designed to be executed by the agent to audit Flutter projects for performance and architectural issues. The script performs filesystem traversal and file read operations to identify code anti-patterns using regular expressions.

flutter-ui