The Agent Skills Directory

[SAFE]: The skill defines a highly structured and disciplined software engineering workflow. It enforces separation between structural changes (refactoring) and behavioral changes (features/fixes), requires human-in-the-loop validation through 'Epistemic Gates', and performs regular architectural stewardship checks to prevent design drift.
[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting project-specific documentation to drive its development actions, which creates a potential surface for indirect prompt injection if those files contain adversarial instructions.
Ingestion points: Reads project artifacts including ./docs/domain-model.md, ./docs/system-design.md, ./docs/scenarios.md, and existing codebase files.
Boundary markers: The instructions establish a hierarchy of authority, designating domain invariants as 'constitutional authority' and 'highest-precedence', which helps anchor the agent to the user's intended project rules.
Capability inventory: The skill uses Write and Edit to modify code, and Bash and Task to execute the project's test suite and build processes.
Sanitization: There is no explicit content sanitization for documentation files; however, the skill mandates frequent user review ('User approves — then next scenario') and 'Epistemic Gates' at every logical boundary, ensuring that an actual developer oversees all generated code and test outcomes.

rdd-build