Decide: hub#140 — one question survives (W1–W3): carve-out strength, design defaults, and the issue's disposition

[obotclaw[bot]]

Decision artifact: https://jwildfire.github.io/obot.roadmap/reports/decisions/2026-08-14-hub140-one-question/ — full diagnosis, evidence, and options there; this thread is the place to record the calls.

The question: hub#140 has carried eight decisions (D1–D8) unanswered for thirteen days and is the bottleneck for the merge carve-out gate and a truthful obot-merge --check. Every claim was re-verified against the live repos tonight; seven of the eight retire — they are engineering calls the design already answers. What survives is one policy call, one blanket sign-off, and the disposition of the issue itself.

Three calls (answer one by one):

W1 — what does the carve-out require? (was D4; policy.json's carveOut._note and invariants.never contradict each other, and oa#83 merged standard-lane on 08-14 while claiming attested — the second live violation) → Recommend a: attested-always — any PR touching a carve-out path needs --jeremy-approved, attended or not; option b (bar only unattended merges) is barely implementable because obot-merge cannot observe attendance.

W2 — blanket-accept the design's defaults for D1, D2, D3-core, D5, D6, D7, D8? One checkbox, not seven; D3's hub-workflows fragment severs to its own issue (it needs a PreToolUse guard, not a merge gate). → Recommend a: yes — none of the seven changes what any repo permits, and each remains overridable in review of the implementation PR, which is attested so you see it anyway.

W3 — what happens to #140 itself? → Recommend a: close it with the obot.agent v0.4.0 release. Its Business Requirement (one write-policy decision per repo) shipped 2026-07-29 and headlines the v0.4.0 notes; the enforcement work is fully tracked in oa#65 + oa#75 + discussion #141 with the design doc as spec — nothing needs re-filing. Option b keeps #140 open until the enforcement PR merges (design §10 status quo), preserving the stale-Requirement framing that already stalled once.

What unblocks: W1 + W2 open the single obot.agent PR from design §10 (policy.json v3, Gates A+B, truthful --check, committed sweep), closing oa#65, oa#75, and #141. While unanswered: the carve-out gate does not exist mechanically, the attested lane can post audit comments for merges that never happen, and --check says "policy permits merging" for PRs GitHub will refuse.

---

This discussion was drafted by Claude Code using Fable 5 (👯🤖 sibling, session 2026-08-14) and awaits @jwildfire's review.

[obotclaw[bot]]

Decided, and built the same day. Recording the answer here as well as on the artifact, which is now the permanent record.

@jwildfire, in chat, 2026-08-15: "#156 looks good. recommnendations approved." — all three recommendations adopted as written (D0005.1–.3, formerly W1–W3).

• The merge tool now always demands the sign-off flag when a pull request touches a guardrail file — every lane, every repo, whether or not he is present. Built in obot.agent#113, tests green: the file list is fetched paginated with a completeness assertion (the obvious command truncates at 100 files with no error), the tool checks that the policy file it just read is the copy in force, and --check now reports policy and mergeability as two verdicts instead of one that gets heard as a promise. That pull request edits a guardrail file, so the gate it adds holds it for his sign-off — its own first test.
• The seven engineering details stand as the design argued them, with the hub-automation fragment severed to obot.agent#114: the hub commits directly under a standing grant, so a merge gate cannot see its scheduled workflows and a carve-out entry there would read as coverage that does not exist.
• #140 is closed against the v0.4.0 release, board moved to Released. Its thirteen days of comments were swept first: everything still live was re-filed with a milestone rather than buried in the close — the enforcement itself (Next/Back in safety-viz example pages #65 and Requirement: participant profile v2 — sidebar surfacing + AE domains #75, both now v0.5.0, carried by the pull request above), the severed hub fragment, and one thing the design had named and nobody owned: every gate here runs inside a tool an agent can edit, and only a GitHub ruleset survives that (obot.agent#115).

Full record, including what each answer resolves: the artifact · decisions log

---

This comment was drafted by Claude Code using Opus 5 (👯🤖 hub140x sibling) and reviewed by @jwildfire