SpecTrace

> Checking that a spec is fully implemented and tested is still manual work. Developers read each requirement, search the code, and hope a test exists. It's slow, requirements get missed, and gaps surface late as rework. > > SpecTrace turns this into a guided IBM Bob workflow. A custom **Spec Auditor** mode, driven by a reusable skill, reads the spec and extracts atomic requirements. Parallel subagents trace each one to the exact file and line that implements it. Python tools map tagged tests, run pytest, and produce a schema-validated traceability report and gap table. The auditor can only write to `reports/`. The developer chooses which gaps to fix. A second custom mode, **Spec Fixer**, fixes them test-first, is not allowed to weaken existing tests, and can only write to the target project. An incremental re-audit then reuses Bob's tracing, re-anchors code references to their new lines, and re-scores in under a second. > > On the included sample project (a 12-requirement expense splitter), Bob's audit took 136 seconds and found 6 gaps: a real currency-rounding bug, 2 untested features, and 3 missing ones. It scored 50%. After 4 approved fixes, the re-audit took 0.48 seconds and scored 83.3%. Every run is published to a static dashboard that shows the spec coloured by status, the before/after comparison, and links to the exact lines on GitHub at the audited commit. > > SpecTrace is repo-agnostic: the target repo and spec are inputs. Next steps are `.docx`/PDF specs, re-auditing in CI on every pull request, and a watsonx.ai Granite extractor.