01 · PROBLEM
Make destructive agent git ops structurally impossible on canonical history.
02 · HYPOTHESIS
Agents only ever get a TTL write token to their own fork; canonical has a single publisher; publication = platform re-validates head and fast-forwards; every token mint/revoke audited.
03 · EVIDENCE SO FAR
LATESTNot tested. The agents gathered reports of the pain behind this idea and scored it on paper only; no experiment has been run on it. Cited pain: force-push/reset/stash races (E-C135, E-C136, E-C138); wrong-checkout writes (E-X006, E-X009); unrelated deletions (E-X010); prompt rules ignored (E-C140).website/projects.json
04 · NEXT FALSIFICATION TEST
Not yet scheduled. Smallest demo on record: rogue agent runs `git push --force` and `reset --hard`; canonical untouched; footprint visible.
StatusParked; Task Passports refinement waits on first-hand demand
We would drop or park it ifIf any documented path lets a raw git client modify canonical state, or if equivalent GitHub setup takes <10 min, product value is low; keep as primitive.