flutter-patterns
Warn
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: MEDIUMPROMPT_INJECTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Behavioral override and persona masking. The file
context/PROSE_ANTI_SLOP.mdcontains comprehensive instructions for the agent to override its default communication style and bypass standard AI self-identification. This includes banning phrases like 'As an AI...', 'As a language model...', or 'I hope this helps', and requiring the agent to rewrite any content that sounds like a chatbot to mimic a human teammate. - [DATA_EXFILTRATION]: Path traversal vulnerability in
scripts/validate_provider.py. The validation tool accepts a--pathargument which is concatenated with the project root. The script does not validate that the resulting path remains within the project boundaries, potentially allowing the agent to read file presence or pattern-match information from sensitive files outside the intended scope if absolute paths or traversal sequences (e.g.,../) are used. - [INDIRECT_PROMPT_INJECTION]: The skill's file validation workflow presents an attack surface for indirect prompt injection.
- Ingestion points: The
scripts/validate_provider.pyscript reads the raw content of local Dart files as specified in the command arguments. - Boundary markers: The skill lacks any delimiters or instructions to treat the ingested file content as untrusted data or to ignore embedded instructions within those files.
- Capability inventory: The script performs file-read operations and returns validation results based on the file content to the agent's context.
- Sanitization: There is no escaping or sanitization of the file content before it is read and processed by the tool.
Audit Metadata