Repository navigation
Scoped review invitations and module stewardship #4069
Description
Activity
DSH review routing update: @wujc12 is the first-review contact for the DSH plugin, installation and host-integration regressions. This replaces the earlier proposed DSH assignment; other contributors remain welcome to contribute and review. Shared lifecycle/replan changes still need the relevant maintainer review. The governance change is in #4071, and scope/availability feedback can stay here.
@huangruiteng Accepted, with the boundary below.
Scope I take: first implementation review for shared-goal-authority qualification: the stage ladder and its evidence rows, transaction-bound capture and drain, writer/cursor recovery, bounded qualification and rollback, and the provider conformance evidence that feeds them (file, PostgreSQL, NoKV). For each review I will record the exact head, the invariant checked, the validation actually run, and anything left unresolved.
Outside the scope: canonical-authority promotion, provider selection or activation, normative RFC changes, and anything that changes the authority source; those stay with the lead maintainer. On NoKV-provider PRs I review the LoopX-side qualification evidence only; NoKV-side decisions keep their own review in NoKV-Lab/NoKV, and the same LoopX acceptance gate applies to NoKV changes as to any other provider.
Availability: a few hours per week, usually a first substantive review within two working days of a request; no SLA implied, and I will say so on the PR when I cannot take one.
I will start with a cross-author PR in this scope, #4053, since it builds directly on the recovery boundaries of #3985.
@huangruiteng Accepted. I can serve as the first-review contact for Chat/runtime session lifecycle, covering resume, submit, close, request idempotency, focused regression coverage, and follow-up maintenance, with shared goal/quota/lease policy staying out of scope. My availability is best-effort without a fixed SLA; I will say on the PR when I need to pause or cannot take a review.
@huangruiteng Accepted. I can take a bounded first implementation review scope in reliability diagnostics: observer-envelope and ledger integrity, receipt/projection consistency, readback and time-evaluation correctness, and focused regressions that prove the diagnostics path remains read-only and does not influence scheduler inputs.
Privacy/export policy, retention/deletion authorization, provider activation, and shared Goal/quota/lease or lifecycle semantics remain separate maintainer reviews. DSH plugin, installation and host-integration reviews should continue through @wujc12; I will focus on the diagnostics contract at that boundary.
Availability is best-effort, with no fixed response-time SLA. I will state on the PR when I cannot take a review or need to pause. For cross-author reviews I will record the exact head, invariants checked, validation actually run, and unresolved findings. My own PRs, including #3975, still require another reviewer; author self-checks are not independent approval.
- added a commit that references this issue
on Sep 14, 2026 - added a commit that references this issue
on Sep 26, 2026 Maintainer-authorized access and review-scope update (2026-10-10).
Organization Member invitations with repository Write through the new
loopx-developersteam have been sent to @hhyykk, @LIHUA919, @NIU-123370, @xyx2002OvO, @jackie-cqz, @Jim-jimu and @BigDataDZ. These grants take effect when each invitation is accepted. @maxliux5 has also received an organization Member invitation; existing Write access and frontend code-owner scope continue.@mikamikasuki has received an external repository Write invitation for an initial four-week collaboration period. We will review the arrangement around 2026-11-07, considering review responses, regression follow-up and sustainable availability. GitHub Write does not expire automatically; this is a review date. Organization membership can follow an explicit scope/availability confirmation and that review.
Please reply here with the scope you want and realistic availability; declining, narrowing or pausing is welcome:
- @cocolord: would you take first review for a cohesive quota-admission and action-projection scope? Shared goal authority, provider activation and release policy remain separately reviewed. Your substantive cross-author reviews are useful evidence; we should map qualifying reviews to the scope you accept before proposing any code-owner route.
- @now-ing: please choose post-writeback hooks or periodic reporting as one initial scope. Please record future conclusions as GitHub reviews with the exact head, invariant, validation performed and unresolved findings.
- @yuefengw and @xiaods: would you like to continue contributing with organization Member + repository Write? For @yuefengw, the proposed task-lease/scheduler review scope remains open to adjustment. We will send the invitations after you confirm continued interest.
- @karenchuu, @kokokoXUY, @YZJF and @zyaiire: would you like to join the organization as community Members? Membership would not automatically grant Write; a sustained implementation/review scope can support a later Write proposal.
- @AronSwan and @JunZ-Leo: would you like repository Triage to help with reproduction, duplicate detection and issue labels? We will grant it after you accept that responsibility.
- @liubf21 and @wujc12: please confirm or adjust the proposed Usage/host-usage ingestion and DSH integration review scopes respectively.
- @liuyizhe, @JackyCSer, @ZaynJarvis, @Hoey041 and @Wanli-Lee: please let us know which responsibilities you want to continue, narrow or pause. Existing Write is retained; recent activity alone is not a reason to revoke access.
@Duang777 and @songoow already have active Write; their accepted Chat/runtime and reliability scopes continue, as does @wchwawa's accepted authority-qualification scope. @steven-kid's Lark appointment continues.
No new code-owner, Maintain, Admin or organization Owner appointment is being made in this rollout. Code-owner proposals still require accepted cohesive scope and at least three substantive in-scope cross-author reviews in the preceding eight weeks. Authored PR totals, author self-checks and unsupported automatic approvals do not establish that qualification. Existing substantive comments can supply technical evidence; only eligible GitHub approvals satisfy merge rules.
Write enables contribution through protected pull requests. Release, security, repository-settings and bypass decisions require lead-maintainer authorization. The lead maintainer has retained the existing automation permissions of @loopx-agent and authorized it to operate the protected formal release tags alongside @huangruiteng. This rollout adds no human approval gate to publication workflows. Contributor invitations do not grant that automation delegation. Silence is not acceptance.
@loopx-agent @huangruiteng Accepted — I take the repository Triage responsibility as scoped: reproduction, duplicate detection, and issue labels.
How I will work: my recent filings (#5462, #5592, #5885, #5886) already follow a repro-first pattern, and I will keep the same evidence discipline in Triage — exact head, reproduction actually run, validation performed, and unresolved findings stated. When confirming reports I reproduce against canonical main before labeling; duplicate detection will reference the canonical issue explicitly.
Availability: best-effort, no fixed SLA; I will say so when I need to pause. No review-scope or Write ambitions implied by this acceptance — Triage only, as offered.
Please grant the permission when convenient.
@loopx-agent @huangruiteng — following up on our acceptance with four items:
-
Recusal set (evidence-only, no labeling from us): threads we authored or filed (fix(tests): keep the single-owner scheme-list sentinel green on Windows #5356, Windows lock-holder sidecar left by killed processes crashes peers.returns() and inbox.pending() JSON scans #5397, Developer ergonomics: three undocumented silent-failure contracts in ChatSessionStore / drain (repro + production incidents) #5462, Windows: peers.py inbox read crashes on os.O_NONBLOCK (bare use; defensive getattr form already used elsewhere) — and the read receipt is persisted before the crash #5511, source_context_authority rejects source_session_v1 profile after #5522 — 46 exact-path tests red on main #5592, turn run-once: quota_spend CAS failure leaves turn permanently in_progress (completed work unauditable) #5885, Long-running goals: zero mid-run progress events (milestone RFC not landed); operator blind for the entire run #5886; deepseek-harness discussion 9084); threads deriving from our PR fix(chat): surface silent contract failures in session store and delivery #5554; and — during the Write trial review period ending around 2026-11-07 — threads authored by mikamikasuki, whose recent implementations (fix(collaboration): allow exact source-session context grants #5595, feat(turns): watch persisted journal progress #5998) built on our diagnoses. The same evidence-only stance applies to surfaces adjacent to fixes other contributors are actively driving.
-
How we read the lane: reproduction and duplicate detection, with evidence posted in comments; lifecycle labels (
duplicate/question/invalid/wontfix) stay with maintainers perdocs/operations/pr-issue-labels.md— we will suggest, not apply. -
Environment boundary: our repro environment is Windows + Python 3.14, editable install of canonical main with test extras (verified 2026-10-11 at
8d5a6fc17:tests/test_file_lock.py20 passed, 3 skipped). TypeScript control-plane surfaces are outside our repro capability; on those we will do code-level review only, and a bare "cannot reproduce" from us should always be read as "outside our repro coverage" — we will say which. -
Prior-knowledge disclosure: on 2026-10-04 we diagnosed the lock-holder sidecar ×
*.jsonglob interaction — the family filed as [Bug]: ephemeral lock holder sidecars race directory walks and match*.jsonglobs, causing Windows-only test flakes #6128 — and deferred a small fix PR until after fix(chat): surface silent contract failures in session store and delivery #5554 merged, then failed to follow up. Our miss. We have posted post-fix verification on [Bug]: ephemeral lock holder sidecars race directory walks and match*.jsonglobs, causing Windows-only test flakes #6128.
On rhythm: we will keep our triage work visible on a weekly cadence — actions taken, or an explicit note when there was nothing to do — and any pause will be announced in this issue.
-
Thanks for the invitation! I'd be happy to join as a community member. I'm a full-time student, so realistically ~3–5 hrs/week.
@loopx-agent @huangruiteng Accepted — confirming the repository Write invitation via
loopx-developers, with a bounded, best-effort first-review scope:Scope I take: Windows-runtime correctness in the Python CLI and its test harness — path/encoding issues (non-ASCII paths, reserved names), filesystem side effects, and repro-first validation on native Windows (exact head, commands actually run, evidence attached). This is the area of my previous contributions here (#4397, and the reviews of #5978/#5966).
Outside the scope: scheduler/lifecycle semantics, shared goal/quota/lease policy, and canonical-authority or provider-selection decisions.
Recusal (evidence-only, no review verdicts): threads where I am the author or a party — my PR #4397, my reviews of #5978/#5966, and #5895 if I end up picking up the GH-C88 implementation.
Availability: best-effort, no fixed SLA; I will say so on the PR when I need to pause or hand a review back. Happy to narrow or adjust this scope anytime.
Scoped review and module stewardship
Governance, review routing and merge-gate changes were merged in
#4071. Scope confirmations
belong in this issue; code and workflow review belongs on that PR.
Thank you for the implementation, fixes and cross-author reviews that keep LoopX moving. We are making the first-review routes clearer so contributors can participate in technical decisions as well as implementation.
How this works
The lead maintainer remains the fallback for availability and cross-module decisions. Please keep scope confirmations here so they are easy to find; implementation discussion stays on the relevant PR.