Repository navigation
Description
Part of the epic "Epic: slice 1, first approval. A Release on next reviews a pull request to this repo, and the machine user's approval stands" (moved from slice 0 on 2026-10-21, oversight steering point 3). This is org-wide in effect; the Factory's design is what drives it.
Why. donpetry-bot, the machine user and the Factory's approver (decision 0005, config/defaults.json), is a member of the org-leads team. That gives it two things it should not hold together:
- The maintain role, on this repo and on at least seven others (google-app-scripts, broodly, markets, TalkTerm, ContentTwin, .github, bmad-bgreat-suite). Approving needs only write; maintain adds repository-settings powers the machine user never uses.
- Code-owner status on every path, because every CODEOWNERS file in the org reads
* @combsmith/org-leads. Six repos require code-owner review. Design §5.1 makes the Maintainer the only code owner of the Factory's own rules, so a machine user whose approval satisfies code-owner review onlifecycle/,config/or.github/breaks that.
The Maintainer ruled on 2026-10-15 that the machine user's approval should satisfy code-owner review everywhere, and that its permissions should be redesigned to get there safely. Until then nothing changes, and the bot stays in org-leads.
Facts gathered on 2026-10-15
- Approving requires write. A code owner's approval also requires write, plus membership of the team that CODEOWNERS names.
- The bot's maintain role comes only from
org-leads; it is not a direct collaborator. The org's default role for members is write. - In the six repos that require code-owner review, the last four merges in each carry no reviews at all. They passed through the Maintainer's org-admin bypass, or through App 3167543's bypass. The bot's code-owner status is unused today.
- This repo's
mainruleset requires 0 approvals and no code-owner review. The tag ruleset forrefs/tags/factory/*(The Factory repo's rulesets hold the Release tags and each required check's App #62) lets only the admin role bypass, so the bot cannot move release tags.
Done when
- A decision record states the machine user's least privilege: its role, which teams it belongs to, and which paths it may own as a code owner. It reconciles "satisfies code-owner review everywhere" with §5.1. One way is a dedicated team holding the bot, named in CODEOWNERS for ordinary paths, while the Factory's rule paths (
lifecycle/,config/authority.json,config/rubric-*,config/defaults.json,.github/, the hooks, the gate and the brakes) name the Maintainer alone. - The bot holds write, not maintain, wherever that is enough, and loses nothing it needs to approve.
- CODEOWNERS in this repo, and wherever else the decision applies, reflects the decision. The change in the org's
repo-templateis noted, since decision 0014 copies the org file word for word. -
config/apps.jsonand decision 0005 agree with the result, and a fixture or check records the bot's real role, so drift is visible.
Needs. The Maintainer, for the team and role changes at org level.
Activity
don-petry commentedon Oct 19, 2026
Sequence (plan review, 2026-10-19)
Not on slice 0's exit: the
mainruleset requires no review until the Factory's reviewer serves the repo (design §9.12), and the machine user is a code owner through org-leads today, which is what slice 1's first approval needs. This is the last story before slice 1's ruleset switch (the new story "The ruleset switch: main requires the machine user's approval of the latest push" under the slice-1 epic), and it runs after P-24 has a fixture (#27), since P-24 is the evidence for the dedicated-path option.When its ruling is drafted, its last Done-when must name the check: a fixture under
trials/of the machine user's role on each repo and its team memberships, read through the Maintainer's login, compared by a check astools/check_apps.pycompares the Apps' grants, so that a change of role or team is visible.don-petry commentedon Nov 1, 2026
The Maintainer's answers on #89 (2026-10-31, relayed by the orchestrator).
org-leadsgets the org roleall_repo_write, andall_repo_maintaincomes off. The orchestrator reads this as removing the team's 7 repo grants too, so the org role alone governs. That is the orchestrator's reading, not his words.Why ruling 2 changed. His answer of 2026-10-21 chose to lower the team on "eight repos", with one
PUTa repo. A read of 2026-10-31 found the role is an org role on all 15 active repos,all_repo_maintain, and this repo has no grant of its own, so those commands would have left the bot at maintain here. Ruling 2, the table of acts in ruling 8 and the scope point are corrected in place, each marked.Nothing has changed at org level yet. The commands are with the orchestrator for him.
🤖 Written by Claude Code
The machine user's role: write on every repo through an org role, hel…