- Decision
- ADR-265
- Date
- 2026-11-18
- Status
- AcceptedAccepted (owner, 2026-11-18: the order of the four, then \"I knew these 3 things, it is not very bad\" for the three findings of platform, then \"yes, push all repos and make them public\" for services and manifest).
- Mentions
- ADR-243, ADR-244, ADR-245
- Mentioned by
- ADR-267
1Decision
Flip the four repositories to public, in the order of R-7 (docs, platform, services, manifest), each with only its default branch pushed: oblivios-docs 448bed9, oblivios-platform c1ff1d7, oblivios-services 1110f95, oblivios-manifest 0cfaafa. The branches fix/passwords-locked-device-biometrics and masklist-easylist of platform and masklist-easylist of services stay local.
Before services and manifest, the full history of the three code repositories was read for secrets and personal strings.
It showed, in the published history of platform: the second address of the owner (the one that ADR-243 removed) in the patch headers of 3bb6452 and 1d82d2c, which ADR-243 said no blob carries any more; the host of the private OTA server in 2be9c39 and 21a1b43 (ADR-245); and the commits from before the force-push of 2026-11-16 (for example c06f046), which GitHub still serves by id and whose ids the activity list of the public repository names (before and after of each push; ADR-244 (4) accepted them as reachable by id only).
Whether those old commits carry the second address was not measured. In services only a private-range test address in a test file was found; in manifest only the old home path of ADR-243.
2Why and evidence
Platform and docs were pushed and made public in one command that ran to its end before the owner stopped it to ask for a check of the other repositories, so platform was public before that check; the owner knew the three findings and accepted them. A history rewrite is no longer cheap: nobody can be told to drop a clone, and removing the old ids needs the repository to be deleted and created again (which drops its star and any fork or watcher). Rejected: setting platform private again and rewriting its history now (offered to the owner, who said the findings are known and not a problem, and then said to publish the rest); pushing all refs (R-2).
3Consequences
Done and read on 2026-11-18: gh api shows visibility=public for all four; an anonymous git clone (no credentials) of each lists only the default branch and a head equal to the local head; tools/docs/preflip-scan.py exits 0 on these heads and gitleaks finds no leak in any of the four; a clean clone of oblivios-docs regenerated with build-public-docs.py shows no changed file against 448bed9; V-2: from anonymous clones of the four at these commits every line of files.sha256 is OK and build-sbom.py --check exits 0 (without --schemas, so no schema validation was run); the four repository addresses and their issue pages answer 200 (404 before the flip, C-6); platform has 0 forks, 1 star and 2 clones of which 1 unique in the 14 days read before services and manifest were flipped.
Not done: the owner's dated read of the final set (O-3); a look at the READMEs as GitHub shows them (C-3); secret scanning, push protection and branch protection (R-3, for after the flip); V-3 to V-8 of the quiet week; the announcement (the third yes of O-4); the claim review of the READMEs and NOTICE files (C-1, C-2); the export position of the three prebuilt files and the terms of Anthropic for the logo; this ADR is not in the public snapshot until oblivios-docs is regenerated and pushed (a push needs a yes each time).
The local refs/original refs of platform still hold the old history and are not pushed.
4Review when
A reader reports an address or host in the history; the owner asks to remove the old ids (delete and create the repository again, then push the default branch only); a fork or mirror appears; the owner reads the final set and finds a change to make.
Read from docs/governance/decision-log.md, entry ADR-265.