4.0.0 Milestone — Roadmap

Planner synthesis, first written 2026-09-14 against the old 4.0.0 milestone (GitHub milestone 8, since renamed 4.x Work). Re-synthesized 2026-10-06 against the GA gate, the 4.0.0 release scope milestone (read with gh api repos/spec-kitty/spec-kitty/milestones/11 and its issue list), the 2026-10-01 direction update below, and the CI debrief of 2026-10-05. The durable "why" is the declaration of intent, 4.0.0.md. This roadmap is the authority for the active 4.0.0 cycle; the prior 3.2.x milestone roadmap points here (see its Addendum 2026-09-14). A summary for decision owners is in the 4.x Executive Overview.

Intent of 4.0.0

4.0.0 was declared on 2026-09-04 as the hosted-collaboration + structural-finish cycle: G1 hosted collaboration (Team Kitty), G2 next-generation product capabilities, and G3 the open-core structural strangler finish. The 2026-10-01 direction update made GA a local CLI integrity release. The 2026-10-12 direction update (below) expands the GA gate beyond local integrity to also gate on two pieces of the open-core structural finish: charter modular extraction (#645) and the mission-status API backend (the open-core producer surface). GA now ships "done when done" — it is not date-gated (see "Release posture"). The four operator-stated goal themes keep their numbering:

  1. Stability. The CLI never reports success while it loses, overwrites or strands local work. Failures are named, not swallowed.
  2. Maturity. One canonical local event store, honestly reported; large modules split into owned seams.
  3. Extensibility. A governed front door and a stable, versioned application API, so CLI, UI, MCP and SDK consumers build against one surface. Two of its pieces — charter modular extraction and the mission-status API backend — are now on the GA gate (Theme 3, below).
  4. Team Kitty enablers. Re-framed 2026-10-12 as "Multiplayer Mode" — the consumer side of the open-core producer API. The hosted backend lane stays frozen, but consumer-side UI/multiplayer on the mission-status API and the produce/drain adapter seam are no longer blocked (see "Direction update (2026-10-12)").

Same discipline as the prior cycles: no new shadow paths. Route or extract onto an existing authority; never build a parallel one.

Direction update (2026-10-12): GA is done-when-done, and the gate expands

The operator ratified six adjustments on 2026-10-12. They extend the 2026-10-01 update below (which still holds where this one is silent); where the two differ, this one wins.

  1. 4.0.0 GA is no longer date-gated. GA ships "done when done." Release candidates keep shipping to early adopters off main in the meantime, and red-CI-no-release (charter Standing Order #9) still holds per release candidate. The milestone-11 identity of the gate is unchanged; only the fixed due date is dropped. (The GitHub milestone object's own due date is handled outside this roadmap.)
  2. The GA gate expands beyond local integrity. It is now local CLI integrity plus (a) charter modular extraction and (b) the mission-status API backend. The "re-synthesized to local-integrity-only" framing from 2026-10-06 is superseded: those two structural pieces move from "4.x after GA" onto the gate.
  3. Charter modular extraction → its own repository, and it is GA-gating. The charter (src/charter/offering/, or all of src/charter/) becomes an independently-released unit consumed through its public surface, like spec-kitty-events / spec-kitty-tracker. Its precondition, #645 (move the ~129 specify_cli → charter imports onto the stable API), is now GA-gating. Grounded in #3101, ADR 2026-08-02-1 and ADR 2026-04-25-1. The repository extraction itself is recorded in ADR 2026-10-12-2; the extraction mechanics ride with a dedicated epic (to be filed). Superseded 2026-10-19 (operator ruling): the Python charter stays in core until it is retired and nothing Python moves to spec-kitty-charter; offering reads cut over to the spec-kitty-charter service through charter.offering.api, and the GA leg is a frozen seam plus a green conformance corpus, not an extraction. ADR 2026-10-12-2 is superseded by spec-kitty-charter ADR-0008 and ADR-0010; see ADR 2026-10-16-1, Amendment 2026-10-19.
  4. The mission-status read API is the open-core producer surface (internally "Build Kitty"), not a Team Kitty surface. Its Java implementation (services/mission-status-api/, epic #5528) is to start in-repo as a modular component and be extracted as needed as an open-core surface — never into a Team Kitty repository (the service directory is in-repo and the read service is substantially built). contracts/mission-status/ and the conformance fixtures already ship and stay canonical beside the Python reducer (ADR 2026-10-01-2, Accepted). Its backend is the GA-gating piece from rule 2.
  5. Team Kitty = "Multiplayer Mode," the consumer side. UI/multiplayer surfaces (mission-ui, the pack composer) consume the open-core producer API across its contract boundary. The "do not build toward Team Kitty" rule is scoped-lifted for (i) consumer-side UI/multiplayer built on that API and (ii) the event-dispatch / produce-drain adapter seam (status/adapters.py). Everything else hosted (zeitgeist auth, the hosted backend, the SaaS surface) stays frozen unless the operator asks (ADR 2026-10-06-1, Amendment 2026-10-12).
  6. The packs source home moves to spec-kitty-open-packs. packs/internal/ and packs/built-in/ here are fetched, pinned copies (pins in packs/<pack>.source.yaml, placed by make sync-packs); the source is authored in spec-kitty-open-packs. This is a source-location move only: the no-bundle invariant is unchanged and ADR 2026-08-16-3 is not reversed.

Direction update (2026-10-01): hosted collaboration is off the critical path

The CEO, CPO and CTO froze Team Kitty / SaaS work and deprioritised it while the commercial vision is reviewed. This section overrides anything below that treats hosted collaboration, auth or Zeitgeist as release-gating. Four rules follow for the rest of the 4.x line:

  1. 4.0.0 GA is the 4.0.0 release scope milestone and nothing else. No Team Kitty, Zeitgeist or hosted-auth item gates it. The hosted-facing epics (#3892 zeitgeist-client auth, and the auth P1 residuals) stay on 4.x Work and wait for the freeze to lift. The retired sync transport stays retired.
  2. All external interaction flows through the adapter module. The CLI core writes its events and does not know where they go. The status fan-out in src/specify_cli/status/adapters.py is that seam, and it is the starting point for the envisioned publish/subscribe system. Zeitgeist is one possible consumer; a local UI is another. No other module talks to an external system directly.
  3. Producing and draining are two separate concerns, each behind its own toggle. Producing an event (writing it to the bus) never depends on whether a drain is enabled, and a producer never knows which system a drain feeds. Today the code couples them: the fan-out functions check drain_posture() before any handler runs, and the hosted-posture contract (kitty-specs/hosted-opt-in-drain-ledger-01M3FFEV/contracts/hosted-posture.md) silences producers when drain is off. Decoupling them amends ADR 2026-09-26-3 and needs a 4.x ADR of its own.
  4. The charter ships on its own, in its own repository. Charter content that tells users to run the CLI is acceptable. The strangler blocker is code coupling: specify_cli imports charter internals directly (129 files at 8a581be1), and those imports must move onto a stable API (#645) before the charter can be packaged separately. As of the 2026-10-12 update, #645 is GA-gating and the charter becomes an independently-released external contract package in its own repository (not just "shippable alone"). Superseded 2026-10-19: the charter does not ship from its own repository as a Python package; see the note under direction item 3 above.

Theme 4's hosted backend lane is frozen; its consumer side was re-scoped on 2026-10-12 (Team Kitty = "Multiplayer Mode," above). The Stability and Maturity themes apply to the local CLI.

Release posture

Read 2026-10-06. Verify live state with gh issue view <n> --repo spec-kitty/spec-kitty before acting; milestones and counts move between release candidates.

Five release candidates have shipped; rc6 is open on main. PyPI upload dates: rc1 2026-09-13, rc2 09-14, rc3 09-17, rc4 09-21, rc5 10-02. main carries the unreleased 4.0.0rc6 ([Unreleased] in CHANGELOG.md, version in pyproject.toml).

rc5 shipped under a waiver. The release workflow requires a green nightly on the release commit (#5034). No nightly has been green since 2026-09-15, so on 2026-10-02 a skip_nightly input was added to manual release runs and rc5 was published with the gate skipped and the waiver logged. rc5 also carried defects in consolidate; their fixes are on main and reach users only with rc6.

The GA gate, milestone 11: 14 open, 210 closed (201 issues, 119 of them P0). GA is done-when-done — no fixed due date (see the 2026-10-12 direction update); the milestone object's own date field is managed outside this roadmap. Of the 16 issues open on 2026-10-01, every one except #5419 has closed. 58 issues closed on the milestone from 2026-10-01 to 2026-10-06, 29 of them P0. This milestone is local integrity; the two other GA-gating pieces (charter modular extraction #645, and the mission-status API backend) are tracked on their own epics — see the dependency spine.

The 14 open issues, by kind:

Kind Issues Gates GA because
Nightly #5419 (P0, performance suite) The 2026-10-06 nightly passed 56 of 57 jobs; this is the one red. rc6 needs it green or another waiver.
Correctness residuals #5638 (rollback leaves coordination status files dirty), #5704 (protected single_branch create leaves a scaffold without meta.json), #5726 (timing-dependent error code when specify re-runs) P2 bugs in the landing and create paths.
Agent integration #702 (P1, Codex .agents/skills drift) The one open P1.
Docs #5428 (P2) Docs cleanup after the 4.x re-anchor; this re-synthesis is part of it.
Structural hygiene #2560, #2561, #2562 (runtime_bridge strangler slices), #3732 (doctrine packs → charter packs rename), #4836 (doctrine→charter drift in operator surfaces), #5538 (residual kind-vocabulary mirror), #5629 (tasks_move_task.py split), #5715 (last CLI import in the lanes seam) P2/P3 debt. Each is a candidate to move to CLI 4.x stable if it does not land before GA.

Not on milestone 11, but raised as GA questions (2026-10-06):

  • #5780 and #5758 (P0, no milestone, no owner). consolidate and agent action review read only local branches, so a teammate's pushed rejection of a work package is ignored in a multi-clone team.
  • #5722 and #5603 (P1, no milestone). orchestrator-api consolidate-mission has no reconciliation gate and no rollback, and it force-deletes the mission or coordination branch. Automation that lands work through the API does not get the protections consolidate has.

Milestone routing. Milestone 11 is the local-integrity leg; the charter-extraction (#645) and mission-status API legs are tracked on their epics.

  • 4.0.0 release scope (11): the GA gate.
  • CLI 4.x stable (12): stable 4.x work after GA, including deferrals out of the gate (229 open).
  • 4.x Work (8): the former 4.0.0 milestone; hosted and product work planned beyond the stabilization release, including the frozen hosted lane (81 open).
  • Infra & Enablers 4.x (15): CI, tooling and quality-debt work, including #1928 and the CI follow-ups #5652 and #5624 (76 open).
  • Live Work NOW (post-launch) (13): five open, including the auth residual #2941.

Drift since the declaration

The 4.0.0 declaration was written on 2026-09-04, two days before the Convergence (#3881) retired the local sync transport and made this repository a client of spec-kitty/zeitgeist and spec-kitty/saas (ADR 2026-09-06-1), and four weeks before hosted work was frozen. Every epic it names has since moved:

Declared advance (2026-09-04) State (2026-10-06)
#1800 SaaS sync & event-envelope hardening Closed.
#1091 Team Kitty launch gate Closed 2026-09-13 (milestone 3.2.7).
#3322 CLI auth & token lifecycle Closed; the client-side successor #3892 is open on 4.x Work and frozen.
#3549 event-log integrity Closed; the local successor #3893 is open on 4.x Work.
#3278 P0 sync false success Closed as not planned (the sync transport is gone).
#3178 P0 wrong-authority egress Closed as completed.
#2519 charter authoring & lifecycle Open on CLI 4.x stable.
#2173 infrastructure-to-logic ports Closed 2026-09-27 on milestone 11.
#1619 one execution context Open on CLI 4.x stable (31 of 44 sub-issues closed on 2026-10-01).
#1797 degod / unshim delivery Closed as not planned.

Never design against or "re-enable" the old sync path; see docs/context/team-kitty.md.

The dependency spine

Re-anchored 2026-10-12. GA is done-when-done; the gate now has three legs.

   THE GA GATE (done when done — red CI == no release, per release candidate)
     ┌ LOCAL INTEGRITY (milestone 11)
     │    green nightly (#5419, 1 of 57 jobs red) · landing+create residuals (#5638 #5704 #5726)
     │    agent integration (#702) · docs (#5428) · structural hygiene (#2560-2 #3732 #4836 #5538 ...)
     │    UNDECIDED: pushed-rejection P0s (#5780 #5758), orchestrator-api gate+rollback (#5722 #5603)
     ├ CHARTER SEAM (offering reads — ADR 2026-10-16-1, Amendment 2026-10-19)
     │    #6188 imports onto charter.offering.api (DONE) ──▶ #6099 frozen seam + green conformance corpus
     │    ──▶ offering stays in core, replaced in place by the spec-kitty-charter service (no extraction)
     └ MISSION-STATUS API BACKEND + READ SURFACE (open-core producer surface, "Build Kitty")
          #5528 Mission Status Read API (contract `1.0.0-SNAPSHOT` in-repo, unreleased; services/mission-status-api/ in-repo, read service substantially built)
          ──▶ #957 mission detail (GA) ; #5532 readers re-point / #5533 routes rehome (OFF-GATE, CLI 4.x stable)
                                                                     ──▶  4.0.0 GA

   OFF-GATE 4.x STRUCTURAL (continues after GA)
       produce / drain decoupling (ADR tbd) ──▶ adapter seam as pub/sub (consumer side unblocked 2026-10-12)
       charter.activation drain (236 reaches) + activation rework (CLI 4.x stable — ADR 2026-10-16-1)
       #2519 later Java charter writes (CLI 4.x stable) · #1619 one execution context (CLI 4.x stable)
       Team Kitty = "Multiplayer Mode": consumer UI/multiplayer on the producer API

   FROZEN (hosted backend — waits for the commercial review)
       #3892 zeitgeist-client auth · auth residual #2941 · hosted backend · SaaS surface

   QUALITY DEBT (Infra & Enablers 4.x, OUTSIDE the blocking graph)
       #1928 ──▶ #4304 #4305 #2970 (security and BLOCKER findings still open)

Reading order:

  1. The nightly closes the release. Every consolidation-safety and coordination-status P0 that was open on 2026-10-01 has closed. #5419 is the only P0 left on the gate, and a green nightly is also what rc6 needs without a waiver.
  2. Two decisions sit outside the gate. The pushed-rejection P0s (#5780, #5758) and the unguarded API landing path (#5722, #5603) are in the same "says success, did the wrong thing" class as the defects the cycle fixed. Either pull them onto milestone 11 or record why GA ships without them.
  3. Two structural pieces are now on the GA gate (2026-10-12). Charter modular extraction — scoped to charter.offering (ADR 2026-10-16-1) — is GA-gating via epic #6099 (four offering-extraction phases). Its import-move precondition #6188 is done and #645 is closed not_planned; the charter.activation drain is post-GA (CLI 4.x stable). The mission-status API backend — the open-core producer surface (ADR 2026-10-01-2, Accepted) — is also GA-gating: the bundled dashboard is already removed (#5530) and the contract is in-repo as 1.0.0-SNAPSHOT, unreleased (contracts/mission-status/). The Java read service services/mission-status-api/ is in-repo and substantially built (read-only, loopback). The value-bearing read surface (epic #5528 + #957 mission detail) is GA-gating; the ledger already has live readers, so the reader-consolidation refactors #5532/#5533 stay off-gate (CLI 4.x stable), per ADR 2026-10-16-1. What stays off the GA gate and continues on the 4.x line: produce/drain decoupling (the precondition for any external consumer), the consumer-side UI/multiplayer ("Multiplayer Mode"), the charter read/write service strangler (ADR 2026-10-03-1, Proposed; starts only after callers move onto #645), and #1619.
  4. The hosted backend is frozen, not cancelled. Nothing on that lane is deleted; it waits. Its consumer side (UI/multiplayer on the producer API) and the produce/drain adapter seam were scoped back in on 2026-10-12 — see the direction update.
  5. Quality debt is a standing campsite epic, outside the blocking graph. It burns down per touched file and never gates a release candidate.

Per-theme progress and scope

State read 2026-10-06. An issue can serve more than one theme; it is listed under its primary theme.

Theme 1 — Stability (local integrity)

The bulk of the cycle, and nearly done for GA. 119 of the 201 closed milestone-11 issues were P0. The recurring defect was a command that exited 0 while it deleted, overwrote or dropped something.

  • Landing work (consolidate). Since rc5: canceled work no longer lands through a dependent lane (#5569); a resumed run no longer drops a lane (#5571); commits made after approval are refused until the work package is approved again (#5668, ADR 2026-10-04-5); a rollback keeps a teammate's commit (#5666); --abort reports what it could not restore (#5686); approved code missing from the target fails and rolls back (#5788). Before rc5: ignored local files are no longer overwritten or deleted (#5392, #5400) and every exit after the first mutation rolls back through one authority (#5385).
  • Workflow and coordination status. Status and decision events no longer split between two checkouts (#5519, #5513, #5440); a work package prompt no longer contradicts move-task (#3931); a copied project no longer moves the original's cursor (#5390); single_branch missions have no lanes (#5100).
  • Open on the gate: #5638, #5704, #5726 (P2).
  • Open off the gate: #5780, #5758 (P0), #5722, #5603 (P1). See "Release posture".
  • Record: the sync-era reliability items closed before 2026-09-14 (#3723, #3700, #2736, #2665, #2264 and others) belong to the retired transport.

Progress: heavy. The new behaviour adds refusals: a mission approved before rc5 stops at consolidate with APPROVAL_STAMP_MISSING until each work package is reviewed again or attested. The rc6 upgrade notes must say so.

Theme 2 — Maturity (one honest local store, owned seams)

  • Done: six of the largest modules were split after rc5 (mission_finalize.py 5,145 → 1,454 lines, consolidation/executor.py 4,145 → 833, orchestrator_api/commands.py 4,097 → 511, tasks_move_task.py 3,786 → 1,887, cli/commands/implement.py 2,304 → 443, core/mission_creation.py 2,173 → 742).
  • Open: #3893 (local event-log integrity, 4.x Work), #3887 (doctor.py residue, 4.x Work), and on the gate #5629 (finish the tasks_move_task.py split) and #2560–#2562 (runtime_bridge).
  • Closed as not planned: #2955 (the Team Kitty envelope producers it would have folded are frozen).

Theme 3 — Extensibility (governed front door + stable API)

Two of its pieces are now GA-gating (2026-10-12).

  • GA-gating (2026-10-12): #645 (stable application API) as the precondition for charter modular extraction into its own externally-released repository; and the mission-status API backend — the open-core producer surface (ADR 2026-10-01-2, Accepted), epic #5528, services/mission-status-api/ slated to start in-repo (direction; the directory does not exist yet — contracts/mission-status/ does), extracted when its contract freezes (never into a Team Kitty repo).
  • Open (off-gate, 4.x line): #901 (governed front door, 4.x Work), #5532 and #5533 (reader re-point / route rehome, CLI 4.x stable), and the consumer-side "Multiplayer Mode" UI/multiplayer that consumes the producer API.
  • Done: the bundled dashboard is removed (#5530); the Mission Status contract is in-repo as 1.0.0-SNAPSHOT (unreleased), with lint, breaking-change checks and a test against every mission in the repository; teams can share agent commands as pack skills (ADR 2026-09-27-1); org packs carry their own built-in override sanction (ADR 2026-10-05-3).
  • On milestone 11: #3732 and #4836 (the doctrine → charter vocabulary), #5538.

Progress: the producer API has a contract in-repo (unreleased) and a planned in-repo service build; its backend and the charter extraction are now gate items, and the governed front door and the consumer UI remain 4.x-line work.

Theme 4 — Team Kitty = "Multiplayer Mode" (consumer side; backend frozen)

Re-scoped 2026-10-12. Team Kitty is the consumer side of the open-core producer API (mission-ui, the pack composer). The backend lane stays frozen and nothing on it gates GA; the consumer side (UI/multiplayer on the mission-status API) and the produce/drain adapter seam (status/adapters.py) are no longer blocked (ADR 2026-10-06-1, Amendment 2026-10-12). Everything hosted below stays frozen unless the operator asks.

  • Open (frozen backend): #3892 (zeitgeist-client auth, 4.x Work), #2941 (widen bypasses renewable CLI auth, Live Work NOW), #4195 (review of the #4121 charter forward-port, 4.x Work).
  • Closed: #1091, #3322, #3233, #3980, #3277, #1621.
  • Closed as not planned: #3196, #3197, #3198 (sync consent and identity, retired with the transport), #2520 (charter events to SaaS; a hosted charter signal would need a new issue built on Zeitgeist).

Cross-cutting — Sonar / quality debt

Parent #1928 is on Infra & Enablers 4.x. The finding-class children #4299–#4303 were closed as not planned (the charter's Sonar standing orders handle them per touched file), and #2969 was fixed.

Issue Finding class State
#4304 S6350 ×17 and other src/ security findings Open, P1
#4305 S3516 constant-return BLOCKER in core/file_lock.py Open, P1
#2970 4 BLOCKER S2083 path-injection findings (≥1 likely false positive) Open, P2

Progress: deliberately post-GA. Adjudicate the security children before remediating, and record false-positive rulings.

Exit criteria for 4.0.0

Re-anchored 2026-10-12. GA is done-when-done; the gate has three legs (criteria 1, 5, 6).

  1. Local integrity: milestone 11 is closed. Every issue on 4.0.0 release scope is closed or re-milestoned with a rationale.
  2. The nightly is green on the release commit. Red CI means no release (charter Standing Order 9), per release candidate. A skip_nightly waiver is an operator decision, recorded each time; rc5 used one.
  3. The CLI never reports success while losing or stranding local work. This is what the consolidation, status and workflow P0s test. #5780 and #5758 are in this class and need an explicit in-or-out decision.
  4. No new shadow path. Every landing routes onto an existing authority. No re-introduction of the retired sync transport, and no module other than the adapter seam talks to an external system.
  5. Charter seam — offering leg (GA-gating; ADR 2026-10-16-1, Amendment 2026-10-19). The import move (#6188) is done and #645 is closed not_planned. The GA leg is a frozen charter.offering.api seam and a green conformance corpus (epic #6099, re-scoped), not an extraction: the Python offering stays in core and is replaced in place by the spec-kitty-charter service (spec-kitty-charter ADR-0008, ADR-0010). The charter.activation drain/rework is post-GA (CLI 4.x stable).
  6. Mission-status API backend + read surface (GA-gating; ADR 2026-10-01-2, 2026-10-16-1). The open-core producer surface ships its backend (contract canonical beside the Python reducer, services/mission-status-api/ in-repo) and the value-bearing read surface that realizes event value (epic #5528 + #957 mission detail). The ledger already has live readers; the reader-consolidation refactors #5532/#5533 stay off-gate (CLI 4.x stable).
  7. (Post-GA) Off-gate structural dispositioned. The produce/drain decoupling, #901, #1619 and the consumer-side "Multiplayer Mode" UI land or are re-milestoned on the 4.x line with a rationale.
  8. (Post-GA) Security debt adjudicated. #4304, #4305 and #2970 close or carry recorded false-positive rulings.

Risks / watch items

  • rc5 is the newest release and has known landing defects. Users get the fixes only with rc6. Four rc5 fixes were found not to hold one day after release; the window between a fix on main and a release is user exposure.
  • The nightly has not been green since 2026-09-15. The 2026-10-06 run had one red job of 57. Most reds came from suites only the nightly runs, which fell behind changes that merged green; the commit-recipe check now runs per PR (#5708), but the performance, Python 3.13 and integration suites have not moved. Tests that no CI job runs (#5652) hide pinned reds.
  • The release gate depends on a token. It fails closed without RELEASE_NIGHTLY_DISPATCH_TOKEN; the new dedicated token's scope is not yet written down (see RELEASE_CHECKLIST.md).
  • The gate refills. Milestone 11 went from 11 open (2026-09-30) to 16 (2026-10-01) to 14 (2026-10-06) while 58 issues closed. 159 issues were opened in the four days after rc5 against 90 closed, 33 of them new open P1s. With GA now done-when-done (no fixed due date), the cut-off is a judgment call at each release candidate, not a calendar date: a P2/P3 hygiene item that is not GA-gating moves to CLI 4.x stable with a one-line rationale rather than holding the release.
  • Review depends on one person. 60 of the 66 PRs merged after rc5 have one author.
  • The API landing path is unguarded. orchestrator-api consolidate-mission lacks the reconciliation gate and rollback (#5722, #5603).
  • Client-repo inversion. Auth and event work for the hosted path is authored upstream in spec-kitty/zeitgeist and spec-kitty/saas. A PR that edits a retired local-sync surface is designing against a dead subsystem; flag it.
  • doctor.py residue (#3887). #1623 closed without shrinking the module; keep the follow-up tracked.

Immediate next steps

  1. Green the nightly (#5419), then cut rc6 without a waiver.
  2. Decide #5780 and #5758: on milestone 11, or GA ships without them with the reason recorded.
  3. Decide #5722 and #5603: fix the API landing path for GA, or document that automation must use consolidate until it is.
  4. Judge the cut-off per release candidate, not by calendar date (GA is done-when-done). Move any non-GA-gating P2/P3 hygiene item to CLI 4.x stable with a one-line rationale.
  5. Write the produce/drain ADR (4.x), amending ADR 2026-09-26-3, then split the toggles in status/adapters.py and core/hosted_posture.py. The adapter seam is now unblocked for the consumer side (2026-10-12).
  6. Move the specify_cli → charter imports onto #645 so the charter can be extracted into its own externally-released repository — this is now GA-gating. Per ADR 2026-09-30-1, drain them rather than holding them in an allowlist. File the charter-repo-extraction epic/ADR.
  7. Ratified 2026-10-12: ADR 2026-10-01-2 is now Accepted (open-core producer surface; the service is to start in-repo and be extracted when frozen). ADR 2026-10-03-1 (charter read/write service strangler) remains Proposed and off-gate.
  8. Re-run this synthesis at each rc bump. Read gh api repos/spec-kitty/spec-kitty/milestones/11 and the milestone's open issues before acting.
Edit this page