docs: reconcile v0.8.65 release ledger to live state

[Hmbown]

Corrected v0.8.65 ledger update after maintainer direction.

Headline: the non-ledger PR churn is complete, but the v0.8.65 milestone is not empty. #3461, #3205, #2300, and #1519 are back in v0.8.65 and are recorded as the remaining release slice.

Records:

• feat(tui): harvest zh-Hans locale bundle #3559 landed the zh-Hans harvest and bare-v detail shortcut cleanup; feat(i18n): extract Chinese translations into dedicated JSON file #3549 was auto-associated/merged through that replacement.
• refactor(config): finish harness posture split #3560 landed the final harness split; [codex] refactor(config): extract harness posture module #3506 was closed as superseded and v0.8.65: Split config modules around provider/model/catalog boundaries #3311 was closed with evidence.
• Extract shared bridge core helpers #3561 landed bridge-core plus review fixes; Extract shared bridge core helpers #3432 was auto-associated/merged through that replacement.
• docs/V0865_REMAINING_AGENT_PROMPT.md is now the canonical remaining-work prompt for v0.8.65: MCP duplicate server instance lifecycle and doctor coverage #3461, v0.8.65: Fleet model classes, loadout auto, and semantic route roles #3205, v0.8.65: Multi-model compatibility, provider docs, and automatic Fleet loadout selection #2300, and v0.8.65: Custom provider endpoints, models, and auth within provider-scoped routing #1519. It tells the next implementation agent to start from a clean branch/worktree off origin/main, not this ledger branch.
• I searched tracked and hidden repo paths for older v0.8.65 handoff/objective docs; only this prompt and docs/V0865_RELEASE_LEDGER.md exist, so there was no stale handoff file to delete.
• v0.8.65: OpenAI Codex/ChatGPT OAuth route verification and usage display #2984 remains moved to v0.8.66 with unfabricated follow-up scope.
• The ledger keeps the security contact, release docs, provider/route/Fleet/usage/catalog closeout, and verification evidence truthful without treating this PR as a public release artifact.

Validation:

• cargo fmt --all --check
• git diff --check

Docs-only. This PR is a truthful handoff ledger, not a version bump, tag, publish, GitHub Release, or deploy.

[claude]

Claude finished @Hmbown's task in 1m 5s —— View job

---

Review: docs/V0865_RELEASE_LEDGER.md

• Read diff and full file
• Verify internal consistency (counts, cross-references, SHA formatting)
• Check factual claims within the document's scope
• Post review

---

Docs-only change. No code correctness, security, or provider-routing issues to flag. The reconciliation is thorough and the verdict ("0 closeable") is substantiated by the table. A few notes:

Correctness checks

• Issue math checks out. 18 partial + 7 not-started = 25 total. All rows in the table tally against those counts. ✓
• Cross-references are consistent. v0.8.65: Resolve every provider/model switch through a ReadyRouteCandidate #3384 appears in both the reconciliation table and decision item Simplify UI and improve footer display (v0.3.14) #2 with the same instruction ("build on feat(config): EPIC #2608 route foundation — canonical types + ReadyRouteCandidate/RouteResolver (#3084, #3384) #3458, not a parallel module"). feat(tui): resolve provider switches through an atomic RouteCandidate (#3384) #3492's closure rationale in the Closed-with-evidence section matches that decision. ✓
• fix: prevent ordinary prompt text from being interpreted as a mode switch #3455 appears twice correctly — once in Phase 2 Closed-with-evidence (closed in favor of fix: prevent ordinary prompt text from being interpreted as a mode switch (#3387) #3491) and is implicitly the "community PR" credited in the fix: prevent ordinary prompt text from being interpreted as a mode switch (#3387) #3491 row. No contradiction. ✓

One gap worth noting

PR #3484 (model-picker search) is mentioned in the verdict intro (line 19) as a recently landed foundation PR, but it doesn't appear in either the Phase 1 or Phase 2 merged tables. A reader tracing the merge history can't tell which session landed it or find the associated evidence. Suggest adding it to whichever phase table is correct (Phase 2 if it landed this session, Phase 1 if prior).

Minor

• Inconsistent SHA lengths: 5b8a5ac0b and 662a459ee are 9 chars; the other evidence SHAs are 10 chars. 9-char SHAs can theoretically collide on large repos. Not blocking, but worth normalising.
• The CodeWhale-trackd-routes worktree entry (line 114) correctly names this PR's branch — good self-consistency.
• The SECURITY.md .net→.com regression finding for feat(tui): enhance approval modal button prominence with visual grouping #3437 and docs: add official-site and mirror provenance copy to codewhale.net #3440 is clearly documented and correctly gates both PRs. Good call calling that out explicitly.

Overall

This is a high-quality state snapshot. The release-readiness verdict is evidence-backed and the decisions-for-Hunter list is specific and actionable. The #3384 gap (build on #3458 resolver, not the parallel WIP one) is the most architecturally load-bearing item and it's documented well. Ready to merge once Hunter reviews the decisions section — the docs themselves are sound.

---

[claude]

Claude finished @Hmbown's task in 1m 52s —— View job

---

Review: docs/V0865_RELEASE_LEDGER.md (updated)

• Read diff and full file
• Check if previous review gaps were addressed (feat(tui): add cross-provider model picker search #3484 table entry, SHA length)
• Verify internal consistency (counts, cross-references, formatting)
• Post review

---

Docs-only change. No code correctness, security, or provider-routing issues. This is the finalization update: the ledger now records the milestone-complete state (all PRs landed or explicitly moved) rather than the mid-triage "0 closeable" state from the first push.

Previous review gaps: both resolved

• PR feat(tui): add cross-provider model picker search #3484 now appears in the "Landed Release Work" table (Provider/model UX polish row, line 44) and in the "Milestone Issues" table under v0.8.65: Cross-provider /model search using explicit route rows #3075 (line 69). ✓
• SHA lengths are now full 40-character hashes for all three evidence merges (a7285ea5a2743d28d0c4bb4154526d0e727ac2fe, ec29998cce511047f0a237f2be4d95e9c5108a05, ead5165d433a3422625f55e0934443b50faad165). ✓

Correctness checks

• Verdict is internally consistent. "Zero open issues" (line 13) matches the Milestone Issues section which lists "None" for open issues and 5 explicitly moved to later milestones. The "not publicly shipped" qualifier (line 16–18) correctly separates milestone-completeness from release-owner actions. ✓
• Cross-references check out. v0.8.65: Split config modules around provider/model/catalog boundaries #3311→refactor(config): finish harness posture split #3560 appears in both "Landed Release Work" (Config module split row) and "Milestone Issues" (v0.8.65: Split config modules around provider/model/catalog boundaries #3311 row). v0.8.65: Resolve every provider/model switch through a ReadyRouteCandidate #3384→feat(config): EPIC #2608 route foundation — canonical types + ReadyRouteCandidate/RouteResolver (#3084, #3384) #3458+[codex] feat(route): gate runtime switches on RouteResolver #3521+release: v0.8.65 (provider/route + Fleet epics, hardening, version bump) #3544 is consistent across "Landed Release Work" (Provider route row) and the Milestone Issues table. feat(tui): harvest zh-Hans locale bundle #3559 is consistent between "PR Queue" and "Landed Release Work" (zh-Hans row). ✓
• SECURITY.md regression (feat(tui): enhance approval modal button prominence with visual grouping #3437/docs: add official-site and mirror provenance copy to codewhale.net #3440) is implicitly resolved. Those PRs are correctly absent from the final ledger; docs: update public security contact #3558 is recorded as the proper fix via hmbown@gmail.com. No contradictions. ✓
• Self-reference in docs: reconcile v0.8.65 release ledger to live state #3493 (PR Queue line 27, Required Closeout step 1) is correct — the ledger documents its own merge as the final closeout step. ✓

Minor note

The "PR Queue" section heading is now a slight misnomer: it contains the final landing batch (#3559/#3560/#3561) plus historical resolution notes for older PRs (#3549/#3506/#3432), not a queue of pending work. Only #3493 itself is actually open. The content is accurate; the heading just won't age perfectly. Not blocking.

Overall

High-quality closeout record. The verdict flip (not-ready → milestone-complete) is well-evidenced by the "Landed Release Work" and "Milestone Issues" tables. The release-owner caveat is clearly stated and matches repo policy. Ready to merge once checks are green.

---

[claude]

Review: PR #3493 (sync)

• Read diff and both changed files
• Check new file docs/V0865_REMAINING_AGENT_PROMPT.md
• Verify consistency with prior ledger state
• Post review

View job run

[claude]

Claude finished @Hmbown's task in 1m 35s —— View job

---

Review: PR #3493 (sync — ledger reconciliation + remaining-agent prompt)

• Read diff and both changed files
• Check new file docs/V0865_REMAINING_AGENT_PROMPT.md
• Verify consistency with prior ledger state
• Post review

---

Docs-only. No code correctness, security, or provider-routing code issues. Both files hang together well and the previous review gaps (SHA lengths, #3484 table entry) remain resolved in this push.

docs/V0865_RELEASE_LEDGER.md

The revision is sound. The verdict flip from "PR queue complete but milestone not empty" reads cleanly, the reinstatement of #3461/#3205/#2300/#1519 is clearly stated as a maintainer direction, and the release-owner caveat is repeated correctly.

Correctness checks:

• Issue tables are internally consistent. All four open issues in the "Open in v0.8.65" table match the "Current Verdict" paragraph and the agent prompt goal verbatim. ✓
• v0.8.65: OpenAI Codex/ChatGPT OAuth route verification and usage display #2984 is in "Moved out" with a stated reason and a target milestone — unfabricated and accurate. ✓
• Closed issues cross-reference correctly: v0.8.65: Split config modules around provider/model/catalog boundaries #3311→refactor(config): extract provider default seeds into provider_defaults (#3311) #3503/refactor(config): extract ProviderKind into a provider_kind module (#3311) #3505/refactor(config): extract harness posture types into a harness module (#3311) #3507/refactor(config): finish harness posture split #3560 and v0.8.65: Cross-provider /model search using explicit route rows #3075→feat(tui): add cross-provider model picker search #3484/[codex] feat(route): gate runtime switches on RouteResolver #3521/release: v0.8.65 (provider/route + Fleet epics, hardening, version bump) #3544 match the "Landed Release Work" table. ✓
• Self-reference for docs: reconcile v0.8.65 release ledger to live state #3493 in the PR Queue is accurate — it's the only open non-ledger PR after the closeout. ✓

Minor note: "PR Queue" heading is still a slight misnomer (flagged in the prior review) — the section is now a resolved-PRs record plus the one open ledger PR. Content is accurate; the heading just won't age well. Not blocking.

docs/V0865_REMAINING_AGENT_PROMPT.md

Well-structured handoff prompt. The hard constraints are the right ones, especially:

• "Do not implement from codex/v0.8.65-ledger-truth" (the current branch) — prevents an agent from building on an unmerged docs-only branch. ✓
• Provider prefix safety explicitly named in both the global hard constraints and the v0.8.65: Custom provider endpoints, models, and auth within provider-scoped routing #1519 direction. ✓
• "No hidden prompt sniffing, no LLM/router model" in the v0.8.65: Fleet model classes, loadout auto, and semantic route roles #3205/v0.8.65: Multi-model compatibility, provider docs, and automatic Fleet loadout selection #2300 direction — correctly applies the v0.8.65 EPIC: Separate provider facts, model facts, offerings, and route resolution #2608 invariant. ✓

One nit worth fixing:

The verification section says:

cargo test -p codewhale-tui --bin codewhale-tui --locked <filter>

--bin <name> scopes cargo test to binary targets only, which would silently skip lib-target tests in that crate. The correct form for running all tests in a crate with a filter is:

cargo test -p codewhale-tui --locked <filter>

This is the pattern used in every other crate invocation in the same block (codewhale-config, codewhale-protocol). The --bin flag here is likely a copy-paste artefact. Fix this →

Minor gap: The "Already done" list omits two ledger areas — "Website/docs provenance" (#3514/#3540/#3543) and "Community/contributor credit handling" (#3517/#3533/#3535). Those areas don't touch the 4 remaining issues, so the omission appears intentional. Not blocking, but if the next agent has to touch credit/docs provenance it won't know those are settled.

Overall

Solid pair of handoff documents. The ledger is an accurate snapshot and the agent prompt gives the next implementer concrete seams, invariants, and acceptance criteria without over-prescribing. The --bin nit is worth a one-line fix before merge, but is not blocking. Ready to merge once checks are green.

---