The Agent Skills Directory

[PROMPT_INJECTION]: The skill's instructions in SKILL.md are purely architectural and procedural. There are no attempts to override agent behavior, bypass safety filters, or extract system prompts. Phrases like 'Non-negotiable rules' are used within the context of software design patterns.
[DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive file paths, hardcoded credentials, or unauthorized network operations were identified. The architectural patterns described (isolating SDKs behind interfaces) are actually security-positive by reducing the surface area of sensitive code.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The provided scripts (scripts/check-flavor-graph.sh and scripts/check-service-boundaries.sh) are local analysis tools. They utilize standard commands such as grep, find, and flutter pub deps. No remote scripts are downloaded or executed.
[DYNAMIC_EXECUTION]: All code examples and scripts are static. There is no use of eval(), exec(), or runtime code generation. The skill emphasizes static analysis and build-time configuration (flavors) over runtime detection.
[SAFE]: The skill serves as a best-practices guide for Flutter developers to maintain clean architecture and testability. All included scripts and code examples are benign and operate within the scope of the project's local environment.

service-boundary-and-native