tidylisted

# Keeping the tooling files honest Five sweeps — numbered 2 through 6, preserving the numbering the decision log already cites; Job 1, the catalog, lives in `--wss-catalog`. All are about the files that describe and drive the tooling: 2. **The claims inside those files** — `WSS.record.tooling.sources`. A skill or agent file that states something false about the project is worse than one that states nothing, and nothing else in the workflow owns them. 3. **The prose prune** — the same files, for text that is verbose and *true*. Job 2 deletes claims that have gone false and is silent on everything else; Job 3 is the deliberate sweep for prose whose removal changes nothing about what Claude does. 4. **The token-economy sweep** — the same files again, for structure that costs more context than the job needs: what could be a script, a cheaper subagent, a gated reference, a shared method, a cache hit. Job 3 asks whether the prose earns its bytes; Job 4 asks whether the *mechanism* earns its chain. 5. **The rot-resistance sweep** — the same files once more, for writing that is true today and shaped to go false: an uncompared second copy, a file two procedures write, a claim no reader can test, a drift nothing would report. Job 2 deletes what has already gone false; Job 5 finds what will. 6. **The routing sweep** — the same files for the one property none of the others measure: whether a skill is reachable. Jobs 3, 4 and 5 can all make a de