The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes extensive 'Prose Anti-Slop' instructions in context/PROSE_ANTI_SLOP.md. These rules strictly govern the agent's output style, banning specific words, punctuation (like em dashes), and common AI-generated phrasing. These are stylistic constraints designed to maintain a specific brand voice and do not attempt to bypass safety filters or ignore prior system instructions.
[INDIRECT_PROMPT_INJECTION]: The validate_provider tool in scripts/validate_provider.py reads user-provided Dart files to perform regex-based checks for architectural best practices.
Ingestion points: Files provided via the --path argument to validate_provider.py.
Boundary markers: None explicitly defined in the script's read operation, though output is returned to the agent as a JSON-structured warning list.
Capability inventory: The script performs read-only file operations and regex searches; no subprocess execution or network operations are performed.
Sanitization: The script performs light static checks and does not interpret or execute the code it reads, minimizing the risk of exploitation via file content.
[COMMAND_EXECUTION]: The Python scripts (scripts/validate_provider.py and scripts/_skill_cli.py) perform standard file system resolution and reading to support the linting tool. These operations are limited to the scope of the project root and do not involve arbitrary shell command execution or unsafe path traversal.

flutter-patterns