What problem does it solve?
Building Claude Code skills without enforced testing and verification leads to hollow, drifting skills that fail in real user scenarios, waste context tokens, and require costly post-ship fixes. This Skill eliminates that risk by enforcing a strict test-driven development workflow with executable gates, so every skill ships with proven, verified behavior.
Core Features & Use Cases
- Red-Green-Refactor TDD Workflow: Enforces writing failing eval cases before implementation, turning handoff specs into functional skills only when all required tests pass.
- Executable E Gate Validation: Runs a re-runnable harness, checks P0 action completion, adversarial checklist coverage, and genuine red test artifacts to prevent fake "green" builds.
- Empirical Trigger Eval: Measures real skill trigger rates against labeled prompts using majority voting and holdout test sets to avoid description overfitting.
- Security & Compliance Linting: Sweeps built skills for hardcoded secrets, shell injection risks, undeclared network fetches, and interactive scripts before shipping.
- Use Case: For example, if you have a handoff spec from skill-guidance for a new invoice processing skill, use this Skill to implement the logic, write failing eval cases for edge cases like malformed invoices, run the harness to verify all tests pass, and generate a build report for skill-zipper to compress for distribution.
Quick Start
Use the skill-engineer skill to build and test a new Claude Code skill from the handoff spec at .skill-guidance/handoff-spec.json, then pass the completed verified build to skill-zipper for final compression.
