Repository navigation
Start here: what the next patch release covers, and how to help #6094
Description
Current intake (2026-10-14 UTC): thirteen contributor PRs have merged as their own PRs, with original authors and source branches retained. Latest #6977 by @Lstarsky0 landed at 88b4c58dfb, preserving the 300 ms cancellation phase and testing killed-host recovery separately under the default command deadline. Exact own-head macOS 19,038 passed / 0 failed / 33 skipped, Ubuntu 19,035 / 0 / 34, Windows 18,304 passed (one recorded browser leak) / 0 failed / 27 skipped; the changed fixture and owner recovery passed on all three, with CI-flags Clippy/budgets passing. #6924 also landed as its own PR with all three native suites passing. Those main changes await the next verified candidate fold. #6973's repaired original branch has audited Ubuntu 19,081 / 0 / 34 plus a separate 35-test portable pass, Windows 18,350 passed (one recorded browser leak) / 0 / 27, owner recovery and actual CI-flags Clippy 9m23s/budgets; macOS is still running. #6393's current-Engine/opaque-route/protected-reader repair passed Runtime 20 / 0 and Core 30 / 0 locally. Its initial TUI result was 631 passed / 5 failed / 636 run; the coherent cap/context/trust/path-warning repair then ran 57 passed / 1 failed / 58 run. Both failed gates are retained. The remaining existing fixture has now been migrated to actual eager-tool behavior, with 2 / 0 passing; feature registry is 2 / 0, and CI-flags workspace Clippy is running before npm/web/fmt/diff. The seven synchronous Echo canonicalization sites are explicitly recorded after all current consumers were traced to existing blocking workers; future callers must retain those fences. Original contributor inventory is corrected to 79 declarations, with a conditional Unix fixture's early-return limitation recorded. The goal-CRUD cleanup passed locally, but Windows leaked PID/cause/elimination remains unproven. Earlier main's Ubuntu job supplied only a CNB-routing notice, with no native result. A complete latest community scan found no new actionable report or Windows GIF. #6976's Windows Cargo publication gap remains labeled, answered and open. Remaining drafts preserve already-carried intent without dependency downgrades or obsolete UI behavior. Final candidate/main CI, package QA, human acceptance and explicit release approval remain required. 0.10.2 has not shipped.
This 0.10.2 candidate makes catalog facts, model admission and exact tiered pricing specific to the selected endpoint. It also fixes Windows shell approval, modal input, wheel capture and junction handling, Claude login input/state handling, bounded quota-worded rate limits and oversized-session diagnostics. 0.10.2 has not shipped.
Committed candidate branch head is ecd3db21dc, tree143c3aa85df3110d9f9b9b19d87d61513a1d5bc0, with main3929 and the four contributor merges below integrated. Its CI38489498503 has completed Ubuntu 19,032 passed / 0 failed / 34 skipped, Windows 18,301 passed (one leaky) / 0 failed / 27 skipped, and Lint. All native result ordinals and named Windows owner_crash_reentry PASS were independently checked. The leak is the existing external-browser launcher. macOS completed 19,035 passed / 0 failed / 33 skipped; all 19,035 contiguous result ordinals, owner recovery and 28 required cases were independently checked. Full exact-head CI is successful. Existing macOS LLVM/debug-strip warnings remain recorded. Squash is on hold for the remaining intake, final source freeze and fresh review admission.
CodeQL38489498516 completed four languages with zero returned analysis/SARIF results on synthetic111c1bb2, whose tree matchesECD. Rust extraction:1,211 files without errors/67with errors,3diagnostics and1,050 diff ranges114files. Error harmlessness and exhaustive coverage are not established. Audit and deny succeeded. These receipts qualify the committed candidate source components; they do not qualify the separate uncommitted lease integration or final distributions.
The frozen integration passed Summary [1.439s] 23 tests run: 23 passed, 15288 skipped (default-feature macOS, all23 unique PASS rows), npm540Node+772Vitest=1,312passed/0failed/7skipped, web/version/changelog/fmt/diff. Source hashes stayed unchanged and match the committed merge. It retains the stronger matching steering receipts/identity/HTTP projection checks, all15 loopback guards and three watchdog corrections, and one later atomic DeepSeek marker write from @gaord. Original contributor commits/source branches and specific release-note credits remain. The existing macOS linker warning is unsuppressed. These local results do not qualify final Windows distributions or provider grants.
Pre-integration4cf evidence remains historical: Ubuntu19,032passed/0failed/34skipped, Lint CI-flags Clippy7m09s/budgets, allfour CodeQL analyses zeroresults. Rust1,211 extracted without errors/67with errors,3 diagnostics/1,047 diff ranges115files; diagnostic harmlessness and exhaustive coverage are not established. The actual4cf CI cancelled automatically after this real verified integration push. Its partial Windows and macOS logs remain preserved, not relabeled as full passes. No old-source result qualifiesECD.
The preceding98 Windows job failed: Summary [1392.514s] 18301 tests run: 18300 passed (1 slow, 1 leaky), 1 failed, 27 skipped. Root independently checked all18,301 contiguous unique result ordinals:18,299 PASS,1 LEAK,1 FAIL. steer_and_interrupt_endpoints_work_on_active_turn expected count1 from a successful HTTP steer response but received0. The API explicitly permits a queued count0 after its unchanged two-second observation deadline. 4cfb453e78 corrects two existing tests to wait for matching durable delivery/completed-item receipts, then assert exact count1; all turn identity, modeled text, interruption, final status and event assertions remain. Five focused existing checks passed locally (5 tests run: 5 passed, 15306 skipped, default features), with fmt/diff. No production code, new test function or code comment. The exact delayed Windows settlement stage remains unknown; original failure/rawSHA17839497 are preserved with no waiver. Historical98 named Windows owner recovery, Ctrl-T and six review-fix checks passed, but the whole Windows job failed and cannot qualify this new head. Old98 macOS then cancelled during compilation, with no suite summary or test results; no macOS pass is inferred.
Historical98 Ubuntu completed19,032 passed/0failed/34skipped, with all19,032 unique PASS rows independently checked. Its Lint passed actual CI-flags workspace/all-target/all-feature Clippy7m15s,55 runtime metrics, persistence ceilings and locales. Historical98 CodeQL succeeded in all four languages with zero returned results, actual Rust extraction1,211 without errors/67 with errors,3 diagnostics and1,044 diff ranges across115files. Those errors are not proven harmless and zero results are not exhaustive assurance. The111 raw security hashes and all old platform/failure logs remain preserved. These are historical prior-source results; current ECD full native CI and security are recorded above.
The previous d83 Ubuntu run failed 19,031 passed / 1 failed / 34 skipped. Its existing Ctrl-T fixture published Models.dev metadata with endpoint-owned provenance and a fabricated fingerprint. 98d6003880 corrects only that fixture source tag; both exact ladders and all two-lap/effective-tier assertions remain unchanged. All five existing Ctrl-T checks passed locally (5 tests run: 5 passed, 15306 skipped, default features), followed by fmt/diff. No production code, new test or code comment. The full failure/raw19,032-ordinal audit is retained; no waiver. D83 CI cancelled remaining jobs after the real verified repair push; its Lint success and any partial platform observations remain historical. Current ECD full CI has since passed; the original failure and intermediate cancellations remain historical and are not substituted for final-source proof.
The founder authorized a squash after complete final-head CI/review, followed by return to canonical main. Proposed squash title: 0.10.2: endpoint-scoped pricing and Windows approval, input and scrolling fixes. Retain all contributor and release branches. Tag, Codewhale package/release publication and deployment require separate approval.
Latest review fixes and local evidence
d83e920e4c addresses the five ready-triggered Copilot/Codex findings: refresh frozen rebound facts; keep the xAI reasoning ladder on its endpoint; preserve per-request child/review usage; admit exact live/configured and selected-account model facts before legacy registry fallback; reject unsupported/nested configured pricing at load. Nested currency/unit fields and malformed/oversized usage remain fail-closed. Replies link the carrying source and actual local evidence; a fresh independent review of all13 committed paths/direct consumers found no blocking regression; complete hosted checks and final review-thread admission are still required.
The 13-path d83 review-fix commit independently matches its frozen verification inputs. 98 corrected only the existing Ctrl-T fixture provenance;4cf corrects two existing steering tests. All production inputs remain identical. The CLI binary was built at prior HEAD3c11 plus those13 frozen paths; its hash binds that source but its Git stamp is not a clean98 distribution or RC artifact.
| Check | Actual result and source boundary |
|---|---|
| New4cf existing steering contract checks | 5 tests run: 5 passed, 15306 skipped; default features. Delivery, caller cancellation, dropped and terminal-steer guards passed. Exact durable count1 and all original functional assertions retained; production two-second wait unchanged. fmt/diff pass; Current ECD hosted Windows passed these existing steering checks; final combined source still needs its own CI. |
| New98 existing Ctrl-T fixture | 5 tests run: 5 passed, 15306 skipped; default features. ModelsDevLive provenance corrected; both original ladders/all two-lap assertions retained. fmt/diff pass. Historical98 Ubuntu passed;98 Windows failed the separate steer contract assumption; The later ECD all-platform CI passed; original failures remain recorded. |
| Focused existing TUI checks | 98 tests run: 98 passed, 15213 skipped; after the helper's production initialization correction, 4 tests run: 4 passed, 15307 skipped. The four are repeated affected cases, not additional unique tests. Unchanged cost/reasoning proof retains its earlier frozen boundary. |
| Configured-model integration | 9 tests run: 9 passed, 0 skipped; unsupported/nested schedules and units rejected, supported context forms accepted. Source unchanged by the helper correction. |
| Existing CLI model-resolution integration | 15 tests run: 15 passed, 0 skipped, using the actual fresh executable. |
| Manual CLI roster matrix | 24 passed / 0 failed / 24 executed: 7 admissions, 15 refusals and 2 current-route diagnostic refusals. All 48 raw stdout/stderr hashes independently audited; isolated fake catalogs, no fabricated account grant. Network activity was not measured. |
| Production CLI build | PASS, 12m14s; SHA256 b6e712c0f2cd68bc75e612adb6f0af8601a497669c0e7b1e8a0cb7c3353f137b, at the frozen dirty-build boundary described above. |
| Workspace/all-target/all-feature Clippy | PASS with actual CI flags, 9m18s, no diagnostics. No lint allowances were added. |
| npm / web / fmt / diff | npm 1,312 passed / 0 failed / 7 skipped; check:web, fmt and diff passed. Unchanged JS proof was reused rather than repeated for Rust-only corrections. |
| Publication dependency order | Locked metadata confirms the 29-crate order with the registry dependency moved to its shared resolver owner. New archive/VCS provenance remains required at the clean final source. |
No new test functions or code comments were added. Two real compile failures remain recorded: the fixture's missing Tool qualification, and the production-only private Config receipt exposed by a struct update. Both were corrected after the compiler was idle; affected checks and production build then passed. The existing unsuppressed macOS __eh_frame linker warning remains in test/build logs. Current4cf hosted Windows/complete CI, final packages/VM and live provider/issuer acceptance are not inferred from these results.
Contributor work and carrying changes
- @SparkofSpike’s #6929 merged as its own PR at 5ebb2c14f8. Exact-head hosted Mac 19035passed/0failed/33skipped and Windows 18299/0/27; owner-crash reentry passed both. It accepts PowerShell call operators, quoted ampersands and combined redirection while preserving destructive-command detection. Release notes credit their fix and @jayanthvee’s detailed Windows runtime floor blocks read-only piped commands under Full Access, on a native (non-npm) install #6963 diagnosis.
- @Lstarsky0’s #6971 merged as its own PR at 39f073346f, now integrated conflict-free. Literal127.0.0.1 bypass repairs two loopback MCP fixtures. Exact-head Mac 19033/0/33, Windows 18297/0/27, Ubuntu 19030/0/34; both changed tests passed each platform. Original commits/source branches are preserved.
- @SparkofSpike’s #6949 merged as their own PR at 182bafe6bd, now integrated conflict-free. Exact PR-head Windows 18,301 passed / 0 failed / 27 skipped, macOS 19,035 / 0 / 33 and Ubuntu 19,032 / 0 / 34; Windows junction cases and owner recovery and Linux containment guards passed. Existing-target junction scope only; missing-target first use and packaged/native acceptance remain unqualified. Source branch retained; notes credit the contribution.
- 5975bd7a25: founder-approved normal approval for unclassified Windows input (Windows runtime floor blocks read-only piped commands under Full Access, on a native (non-npm) install #6963). Actual unbounded node.exe image kills retain their reviewing-mode hold; Full Access bypasses the floor.
- f9210a4428: modal paste cannot change hidden history-search input (未适配中文输入法 #2323). Native Microsoft Pinyin config-filter and ordinary write-approval commits leave the composer clean; OS preedit anchoring and all modal types remain open.
- cf7199b52e: releases Win32 mouse capture before external-editor reset (After running for a long time, TUI scrolling becomes laggy, like jelly. #6652). v0.10.1 wheel-loss reproduced after F4/Notepad; candidate preserves wheel, PageUp, tab focus, resize and inline/fullscreen transitions on Win11 ARM64/Terminal 1.24.12741. The separate hours-long slowdown remains open.
- bee861137e: Claude uses independent state and the sole terminal input pump handoff (Claude subscription sign-in through the shared OAuth and native Messages route #6932). Existing registration/experimental status remain; this is not provider permission or a valid grant receipt.
- 2b3decc006: accepted body retry hints <=120 seconds distinguish temporary quota-worded 429s; explicit durable codes and header-only evidence stay terminal (Bug or feature: in case of Gemini 429 error, wait and auto-retry last task #6923, thanks @Statter).
- 4dfe852c6e: compaction failures retain their cause, and an oversized source no longer falsely blames its active conversation (A saved session document larger than 8 MiB can no longer be resumed or forked, and the refusal names no remedy — the journal is what pushes it over #6955, thanks @7jrxt42BxFZo4iAnN4CX).
- 174a32b770: fail-closed per-machine Windows import checks for CLI aliases/optional host, NSIS setup/payloads and actual embedded plugins before upload (Release: check the Windows import floor before shipping (Windows 10 1607 regression guard) #6962/Windows 10 1607: v0.10.1 fails to start — unresolved load-time import DeriveCapabilitySidsFromName (0xC0000139) #6953). Unknown pairs/ordinals, delay imports, empty/malformed tables and machine mismatches fail with retained receipts.
- b8085bdcc0: finite 210-minute native budgets and matching Mac dev/test debug metadata. Test assertions, complete suite, deadlines and retries are unchanged. The previous Mac run was incomplete, not a pass:29m10s CLI+125m18s test compile, then 165-minute cap with 15,204 completed PASS / 0 FAIL / 0 TIMEOUT and no final summary.
- 9bfb3809d7: English/Chinese supported-platform contracts, first-turn/tool schemas, corrected notes and generated docs. The original Catalog and pricing: finish collapsing model facts and price tables into one authority #6396 source and cloud evidence below remain applicable only to their recorded revision.
Latest contributor PR integration
- @Lstarsky0’s #6967 durable steering acceptance fix merged as its own PR in 313cc4b4. Own-head macOS 19,033 / 0 failed / 33 skipped, Ubuntu 19,030 / 0 / 34, Windows 18,297 passed (one leaky) / 0 / 27; owner recovery passed. ECD integration retains the stronger matching receipt and identity assertions, verified again in the integrated23-case local batch.
- @gaord’s #6974 provider-table key marker fix merged as its own PR in e3436e48. Own-head macOS 19,035 / 0 / 33, Ubuntu 19,032 / 0 / 34, Windows 18,301 passed (one leaky) / 0 / 27. ECD reconciliation keeps one contributor marker write inside the atomic save and preserves read guards/rollback.
- @Lstarsky0’s #6975 explicit loopback bypass in15 existing fixtures merged as its own PR in 8aeeff6d. Own-head macOS 19,035 / 0 / 33 after one recorded retry, Ubuntu 19,032 / 0 / 34, Windows 18,301 passed (one leaky) / 0 / 27; all15 cases and owner recovery passed. Original partial macOS deadline remains recorded, not relabeled.
- @Lstarsky0’s #6966 separate no-receipt execution watchdog and approval-event readiness fix merged as its own PR in 3929bcee. Own-head macOS 19,033 / 0 / 33, Ubuntu 19,030 / 0 / 34, Windows 18,297 passed (one leaky) / 0 / 27; all three changed cases and owner recovery passed. The verified old review finding was resolved.
All four retain their original authored commits and source branches. These exact own-head gates do not qualify the new main merge SHA or assembled release packages. No release has been published.
@Lstarsky0’s #6956 is now merged as its own PR at bbae5043. Exact own-head macOS19,035/0failed/33skipped, Ubuntu19,032/0/34 and Windows18,301passed(oneleaky)/0/27; owner recovery passed all three. Both cleanup cases passed macOS/Ubuntu; the hook-output case is Unix-only, and the portable cleanup case passed Windows. The final strict worktree/branch assertions remain. Original authored commit and source branch are retained. Main has advanced to that merge; ECD has not yet folded it in.
@Lstarsky0’s #6958 is merged at 73e993a295. Own-head macOS19,035/0/33, Ubuntu19,032/0/34, Windows18,301passed(oneleaky)/0/27 and Lint passed; listing and owner cases passed. #6943 now has repaired headf6c0d4ff and new CI38502210725. Its old Windows clipboard acceptance passed, but the unrelated existing RLM deadline fixture failed. The repair waits for the pending request phase and retains all functional assertions; new whole-head Windows proof is pending.
@gaord’s #6959 now has head97aac3e: the maintainer registry-summary and generated-doc fixes are pushed, and Web38496306483 passed772Vitest/facts/docs/build. NativeCI38496306482 passed macOS 19,038/0/33, Ubuntu 19,035/0/34, Windows 18,304 passed (one leaky)/0/27, and Lint. The three contributor trust cases, registry and owner checks passed on the completed platforms. The contributor’s original commit and three tests remain; this PR merged at 947f670fee, preserving author commits/tests/branch; it is not yet carried byECD.
@gaord’s #6924 now carries 36d05258, preserving both human commits and merging mainbbae on the original branch. Lease lifetime, stable kernel-lock inode authority and protected-directory/Windows-handle guards are repaired. Scoped local proof:42existingTUI/19app-server/1registry checks passed; unchanged config12/npm1312/0/7/web gates reused; four actual Unix process journeys passed with the newly built isolated CLI. The own-fork files are reconciled against main and mapped to those local hunks; the whole own-head CI38505158242, Windows and final packages remain pending. Earlier source/fixture failures and the mistaken old-binary manual run are preserved and excluded. ECD does not yet carry it.
Older feature/draft PRs remain individually reviewed work, with explicit source and release-scope gaps. No release has shipped.
Historical platform and package evidence
D83 CI38463005110 has a confirmed Ubuntu failure:19,031PASS+1FAIL across19,032 unique ordinals,34 skipped, named owner recovery and the five production-review repairs passed. The only failure is the malformed Ctrl-T fixture corrected in98; its assertions were preserved. Hosted Lint succeeded with actual CI-flags workspace Clippy9m25s/all55 runtime-contract metrics/persistence ceilings. The new real repair push cancelled the remaining old jobs; they are incomplete, not passing98 evidence.
D83 CodeQL38463005240 completed successfully in four languages with zero results, exact synthetic merge d5b0393/candidate tree35ea. Rust extraction1,211 without errors/67 with errors,3 diagnostics,1,043 ranges/115files; root matched102 raw receipt hashes. This is historical scoped coverage, not exhaustive assurance or98/final-main qualification. Audit/deny succeeded, four old alerts remain open/undismissed, Dependabot reads were200empty.
Pre-review-fix3c11 CI38433637142 passed: macOS19,035/0/33, Ubuntu19,032/0/34, Windows18,301 passed(one slow,one leaky)/0failed/27skipped. owner_crash_reentry and the changed cancellation observer passed on all three. Windows had18,300 PASS+1LEAK across18,301 unique ordinals; the existing external-browser launcher leak is retained. The whole cancellation test took6.008s, which does not measure the observer's delayed stage.
Earlier6e Windows failed18,300passed(two leaky)/1failed/27skipped; owner recovery passed. The two-second cleanup conjunction was not observed in time; its delayed stage/iteration remains unproven. Commit3c11 reuses the existing ten-second settlement watchdog while preserving every functional assertion and production budget. Old cancelled macOS runs are not full passes.
Strict3c11 no-upload29-crate preflight and independent archive review passed:29 package verifications, clean exact3c11 VCS,1,319 matching tracked payloads, no unexpected/mismatched/node_modules files, every archive below10MiB; TUI9,431,091bytes/8.994189MiB. Earlier7bf/6e archive sets and source-qualified receipts remain preserved. This does not qualify d83 or the eventual squash/main VCS identity. Final-main preflight needs its own clean source receipt.
Pre-repair CodeQL38433637276 passed all four languages with zero returned results, on mergecb5fe84e/tree9f6b87b3. Rust extraction:1,211 without errors/67 with errors;1,015 diff ranges across111files. Three diagnostic messages/error files are not proven harmless; zero results are not exhaustive assurance. Four existing alerts remain open and undismissed; prior Dependabot reads returned200 with no open alerts. New-head and final-main security/source alignment remain required.
Earlier frozen-slice evidence (prior source)
These receipts cover their original frozen cf7199/6e/7bf inputs, not a fresh d83 execution. Subsequent contributor integrations were bound to their verified PR blobs; unchanged proof is reused with its boundary recorded.
| Check | Actual result |
|---|---|
| Focused existing TUI tests, all features | 170 tests run: 170 passed, 15157 skipped outside the selected filter; 0 failed |
| npm test | 1312 passed, 0 failed, 7 skipped (540 Node+772 Vitest) |
| check:web, fmt, dated versions/changelog sync, source budgets | PASS; generated agent-doc projection refreshed |
| Clippy,TUI all targets/features, actual CI flags | PASS;9m 35s |
| Actual Claude CLI/slash/picker malformed-input PTY | 3 cases PASS; cooked prompt, exact rejection, both TUI consumers restore raw input/clean composer/responsive new input. Wider repaint fixes fixture footer truncation; original failed observations retained. No valid grant/token request. POSIX only. |
| Actual normal CLI quota loopback | 7 cases PASS: accepted structured delay retries 2 byte-identical requests 258 ms apart and succeeds; 6 terminal cases each one request, exit 1, success false, empty output. Original scorer mistook echoed prompt marker for output; corrected JSON-field audit preserves original receipts, no execution repeated. Isolated path covered by existing tests, not this manual scenario. |
| Actual bounded Runtime history | 9 MiB synthetic source, archive-blocked 409 accurate refusal/all 4 entries retained; restored archive yields 1512 byte source + 3 archive entries losslessly and 1 message resume; 0 model POSTs. First long-path fixture stopped at macOS socket guard, corrected isolated SSD path passes. |
| Strict 29-crate preflight (historical6e) | All 29 built/verified without allow-dirty; 29/0 archive provenance audit, clean exact6e72400bbb VCS, 1,319 tracked payloads match, TUI 8.994222 MiB with no node_modules; no uploads. |
| Existing workflow contract/actionlint | 155 paths, 11 event cases x 12 steps, 13 hermetic checks, 7 targets/34 assets;PASS with actual CI lint flags |
| Static Windows guard | x64 v0.10.0: 411 importsPASS; v0.10.1: 454 examined, expected DeriveCapabilitySidsFromName rejection. ARM 453/453 exports on actual Win11 build 26300. Historical NSIS setup/System/nsDialogs163/19/54static importsPASS, x86 ordinal17 bound to SDK/export identity. Hosted final-artifact guard not yet run. |
Earlier exact cf hosted Windows: 18299 passed, 0 failed,27skipped including owner_crash_reentry; native Win11 ARM64 debug classifier 89/89 and owner 1/1. These receipts do not cover the new source head, final packages or Windows10. Native Ask /undo also restored two files byte-for-byte, removed a new third file and pruned the matching conversation after one three-write request (#6970); exact reporter session/packaged release still separate.
Remaining release gates
- Fold the remaining individually verified PRs, freeze and complete the resulting exact-head hosted CI/review, with literal Windows owner-crash PASS and no confirmed unresolved finding; only then perform the authorized squash. Preserve branches and restore canonical to actual current main without reset/history rewrite.
- Freeze actual main SHA/tree; inventory and complete full manual CI plus nonpublishing RC for7 targets/34 baseline assets, including the selected RC's own Parity, digests/checksums and import sidecars. Avoid duplicate queued/active runs or hiding automatic failures. Optional compiled-host delivery is currently disabled; recheck settings before dispatch.
- Final clean-main strict29 no-upload package/VCS provenance. Ratatui0.1.2 alone is already published with founder approval from frozen cf7199 vendor; downloaded registry bytes match the verified dry-run. Do not republish it.
- Actual final Windows native/archive/npm/real NSIS setup, embedded plugins, generated uninstaller, VC dependency closure and packaged/native-app owner recovery. Installer CI's two-byte MZ fixture and Ubuntu-only final npm smoke are not this acceptance. Win11ARM VM ordinary-user unlock remains unavailable; no authentication reset or privilege bypass. Actual x64Windows1607 startup/install or explicit founder acceptance remains required.
- Claude independent-state/input handoff has synthetic malformed-input proof, not a successful issuer grant. Registration/client identity, live sign-in and release visibility remain founder/external decisions or missing receipts. 0.10.2: repair ChatGPT sign-in rejected by unsupported originator parameter #6925 live authorization/login and remaining 0.10.2: every Bash/shell operation Codewhale starts or tracks is inspectable and stoppable #6931 terminal journeys remain open.
- Real-session codewhale resume refuses any session whose saved history still holds a raw sub-agent handoff — the canonical acknowledgement compares the Engine-restored messages with the saved document #6954 raw-handoff resume/fork; final native controls/recovery; 未适配中文输入法 #2323 all modal/OS preedit limits; After running for a long time, TUI scrolling becomes laggy, like jelly. #6652 separate hours-long slowdown. Reporter cannot reproduce wheel loss on an unidentified unreleased build, with no new GIF; keep the issue open for final retest.
- Final-source security receipts. Correct Linear workspace sync is unavailable; no retired-workspace writes. Issues close on release or explicit direction. Runtime API terminal contract: input attribution, restart-safe identity, Windows ConPTY #6160 ConPTY and MuseCode native Engine integration remain separate research/work, not release promises.
Original cloud verification (historical 2f0cac7; 9e8fa55 docs only)
| Gate | Result |
|---|---|
| TUI lib | 15280 passed, 26 skipped |
| TUI integration/cucumber/feature_registry | 243 passed |
| CLI | 541 passed |
| Rest workspace, all features | 2903 passed |
| Configlib | 801 passed, 0 failed |
| Clippy/fmt/version/changelog/facts | PASS |
| npm/web | 1312 passed, 0 failed, 6 skipped; web PASS except unreachable latest-release API (v0.10.1 confirmed independently) |
| Local release build | codewhale 0.10.2 (dev); source build only |
Cloud proxy/root/PID1 constraints were isolated for those checks; they are not substituted for current native or hosted receipts. Original temporary dependency patch size measurement was 8.99 MiB for TUI; fresh strict 29-crate verification/provenance audit passed at 7bfde0d; final-main publication preflight remains a separate freeze gate.
Refs #6396
Current security coverage detail:70 warning rows name67 distinct Rust paths (60 patch files,3 pet/design files and4 product-source files); none equals a file in the13-path review repair. The equal count suggests a correspondence to the67-error metric, but the saved output does not explicitly establish membership. The three higher-level diagnostic bodies/locations remain unavailable. This mapping is not harmlessness or complete-coverage proof.
Metadata
Metadata
Assignees
Labels
Type
Projects
- StatusShow more project fieldsTodo
Activity
Hmbown commentedon Sep 14, 2026
Tracker organization, 2026-09-13/14. Every open issue now sits on exactly one milestone, and milestones hold issues only (PRs carry none).
v0.9.13ata0b81f619). Its remaining open issues are the fixed defects that close with the Release page as evidence: the sub-agent fan-out set Sub-agent runtime: nine defects observed in a six-worker fan-out (0.9.13 blocker) #6121–action=status compact projection returns ~50k tokens of nested per-agent payload #6130, Extensions page: re-auth on an MCP server freezes the whole TUI, and the Diagnose button does nothing #5974–Session metrics: the tok/s number doesn't look right — audit the denominator and the token source #5977, Google Gemini problem on "from scratch" installation #6018, Google: thought-signature replay on a non-official route fails with Google's raw 400 instead of an actionable error #6048.codewhale execopt into hooks and typed permission rules #6099); session store integrity (save_session silently deletes the oldest transcript once the store reaches MAX_SESSIONS #6136–Design: what should resume do when the target session — or its provider — is gone? #6138, Session recovery is invisible to the model: after a force-quit, the agent cannot find what it was doing #5715); pet acceptance (Pet: qualify the /pet habitat in a real terminal and accept the shared owner across TUI + desktop #6155); security (security sweep 2026-09-11 #6058); declutter and UX (feat(tui): launch-screen MCP block — decide copy/paste (and click) affordances; it is read-only by design today #6085, feat(tui): declutter round 2 — /settings (71 rows), /setup providers (flat 30+), /fleet naming #6087, feat(tui): Session Picker UX — hide empty auto-sessions, highlight current session, page-scroll the list, widen the list pane #6014, /models command returns only partial model list — missing pagination support #6009, Fleet models: save-in-role from the workers view and an always-visible operator model list #5915, Agents view: message an agent from its row, show per-agent cost, document the roster #5479).Emptied milestones v0.9.10 and v0.9.11 are closed. If you pick something up, say so on the issue; if you think an issue is misfiled, comment there rather than moving it silently.
Hmbown commentedon Sep 14, 2026
Run summary for the planning issue: the v0.9.14 slice run is complete. Shipped and closed: #6153, #6156, #6159, #6118, #6119, #6136, #6137, #6138 — all in the single milestone PR #6161 (base 3b00c1a; main has since moved to e9b761c, deliberately not rebased). Every other open milestone issue has a fresh status/decision comment from this run (verified mechanism, next slice, or the gate that blocks it) — see the tracker. Gate receipts are in the PR body: nextest workspace 15645/15645, clippy/fmt clean, all Lint scripts pass, npm test + check:web green. Hosted CI on the branch has never run; opening the PR is its first run.
Hmbown commentedon Sep 14, 2026
Continuation run (2026-09-14): four more slices landed on
v0914-chunk1, all in PR #6161 — #6101 (opt-in approval timeout +Timeoutreceipt), #6147 (the engine's last two unbounded channels bounded), #6040 (prompt=consenton every interactive OAuth login + honest logout messaging), #5528 (failed workflow runs raise a sticky error toast).Full gates re-run on the branch:
cargo fmt --checkclean; clippy exit 0;nextest15654/15654 passed (21 skipped);npm test470 passed +check:webclean; all Lint scripts + selftests pass (locale parity 2251/2251 x 15); release checks OK.PR #6161 is currently CONFLICTING with main in three changelog surfaces (
CHANGELOG.md,crates/tui/CHANGELOG.md,web/lib/changelog.generated.ts) — reported in the PR body; not rebased per the brief, resolution is a maintainer action. #6009 closed with the documented-contract finding (OpenAI's spec documents no model-list pagination).Hmbown commentedon Sep 14, 2026
Maintainer action on PR #6161, 2026-09-14 (Claude Fable 5.1 at Hunter's direction, 0.9.14 takeover). The v0.9.14 slice run's PR was CONFLICTING only on the three changelog surfaces. Resolved by merging
mainintov0914-chunk1as a union of both Unreleased sections and regeneratingcrates/tui/CHANGELOG.md+web/lib/changelog.generated.ts(fc08c80cb0). Hosted CI then flagged one new CodeQL alert (rust/cleartext-logging) in the #6040 regression test, whose assertion message interpolated the OAuth authorization URL; fixed in16ebe8f390(message names only the fact under test). All 30 required checks passed; merged ascf37ea81ad, closing #6153, #6156, #6159, #6118, #6119, #6136, #6137, #6138, #6101, #6147, #6040, #5528.Now in flight on top of that main: #6163 (workspace files + session artifact routes for the GPUI desktop) and #6162 (canceled automation runs get a receipt), each on its own branch awaiting independent verification before landing.
Hmbown commentedon Sep 15, 2026
v0.9.14 progress note — slices landing on branch
v0914-chunk2(PR #6175):Op::SendMessagepayload →TurnSpec(wire format unchanged; residualsend().awaitsites audited)base64/shlexunified; remaining duplicates are upstream-pinned (documented on the issue)expectconversion landed, budget 369 → 254; blanket-allow and commented-allow phases remainFleetDenialGuardpolicyEarlier branch slices: #6099 #6037 #6149 #6141 #6033 #6140 #5715 #6014 #6085, #6132 qualified.
To avoid duplicating work: check PR #6175 and issue assignees before picking up a milestone item; in-progress slices are noted on their issues.
docs(contributing): stop pointing contributors at a branch that does …
[-]v0.9.14 — start here: what's planned, how to help, how to report[/-][+]v0.10.0 — start here: redesign, release checks, and how to help[/+]Hmbown commentedon Sep 20, 2026
The planned release is now 0.10.0, reflecting the broader workbench and settings redesign plus session/runtime improvements. This supersedes the unpublished 0.9.14 candidate; the milestone and its history stay intact.
The issue body above now describes the direction, remaining acceptance checks, and ways to give design feedback. We are also reviewing older issues against current code and product direction: completed or superseded work gets an explicit explanation, while valid gaps stay open. This is a planning update, not a release announcement. Thanks to everyone contributing fixes, reports, and testing.
Hmbown commentedon Sep 20, 2026
0.10.0 readiness pass, 2026-09-20 — branch
claude/trusting-shannon-sgusdh, draft PR #6370, head 2ca42e9, CI green on Lint, Ubuntu, macOS, Windows and the OpenHarmony check.Landed on the branch:
main's red legs: README translation lint, and the underwater widget tests that still counted ASCII fish after the native braille poses.ConfigFile.base, phased store-binding test, explicit 2 MiB guards (no stack ceiling raised).main's versions).SIGPIPE=SIG_DFL, so a stdio MCP server that exits beforeinitializeis written killed the whole session silently (6 of 12 launches on the launch-card fixture).run_tuinow ignores SIGPIPE; 0 of 12 after the fix, and the PTY harness prints the killing signal.Not verified in this pass and still open for a human with a native terminal: #6155 (
/petin Kitty), #5856 (Computer Use on macOS), and any real-provider turn. #6184 has status on its thread. Dependabot PRs were left open.Generated by Claude Code
[-]v0.10.0 — start here: redesign, release checks, and how to help[/-][+]v0.10.1 — start here: finishing what 0.10.0 started, and how to help[/+]chore(release): record combined 0.10.1 candidate validation
fix(mcp): handshake hygiene — empty client capabilities, accept 2025-…
[-]v0.10.1 — start here: finishing what 0.10.0 started, and how to help[/-][+]Start here: what the next patch release covers, and how to help[/+]test(runtime): reuse the durable user-input settlement watchdog