Kiro specs are great at describing what to build. They are silent about whether the build kept its promises. Spec Auditor closes that loop: it reads requirements.md, finds every REQ-XXX, and checks that each one is referenced by source code AND by tests.
- COVERED / PARTIAL / MISSING / ORPHAN, one line per requirement.
- Exit 0 when everything is covered, 1 when there are gaps, 2 on usage errors.
- Stdlib-only Python, no network, language-agnostic (plain text search).
- Self-auditing: the tool's own spec is audited by the tool (10/10 covered). Honest limit, stated in the README: a tag match proves a requirement is referenced, not that the feature works. Test execution stays in CI.
How it was built with Kiro: Requirements, design, and tasks live under .kiro/specs/spec-auditor/; steering rules under .kiro/steering/. The implementation follows the tasks file in order, and the demo is real terminal output (demo.sh, demo-output.txt), not hand-written.
Demo: Run ./demo.sh: first the self-audit (all green), then an example calculator project with one deliberately missing requirement (REQ-103), reported MISSING with exit 1.
Log in or sign up for Devpost to join the conversation.