Repository navigation
[PR] release v0.1.214 - #6552
Conversation
…nstall dir The pure-SQLite storage default (owner decision 2026-11-13, 5a80e2b) stopped writing JSONL session envelopes by default. Five hermetic client e2e lanes (native-pi / native-dsh / native-opencode / pi-codex-ws / subagent-sessions) assert those envelopes and were not updated with the default flip: they now fail with 'proxy must persist the session (got 0 sessions)' (reproduced locally with the CI-pinned pi-stable 0.83.6 and on the PR head). Pin BILI_STORAGE_JSON_WRITER=1 in the spawned-client env: these lanes cover proxy/session semantics, not the storage default — the pure-SQLite path stays covered by unit tests and the real-model gate. Also: the advisory-rollback waitFor probes crashed with ENOENT when the updater's uninstall->reinstall half-deletes the install dir (npm drops package.json before the directory; reproduced locally). New pkgVersionOrAbsent() helper treats that window as 'not there yet'; non-ENOENT errors still throw. This was the ubuntu flake behind the 'hermetic registry e2e' red on 16b8d9b (run 43454198949).
The CI-only baseline preload (tests/preload-storage-baseline.mjs) pins BILI_STORAGE_UNIFIED=legacy for the legacy-default corpus. The sqlite-default notice test asserts the production default (absent spelling => pure-sqlite), so without an explicit unpin it passes locally (no preload) and reds every CI test shard carrying this file. Also unpin BILI_STORAGE_JSON_WRITER for the same predicate.
📦 Built Package ArtifactBranch: Latest build — install this exact versionnpm install -g billion-context@0.0.9896-pr.6552.7889c0cThis exact version is immutable and reproducible. The rolling Recent builds (commit → version)
Download artifact
tar xzf billion-context-pr6552.tgz
npm install -g packageInstall instructions are refreshed on each push. |
tests/e2e/e2e-native-dsh-persona.test.ts asserts JSONL session envelopes
('the raw conversation key persisted a session') but its spawned env missed
the same pin as the other client lanes — red in the e2e-dsh lane.
…EBUSY) The dual-write pins added with the pure-sqlite default flip make persist-zstd and persist-metadata-fold-gate open index.db; the suites never closed their stores, so the tmp-dir teardown raced a still-open SQLite handle — Windows cannot unlink an open file, and the retry budget in tmp-rm.ts lost deterministically (EBUSY unlink .../index.db on windows 22/1, 22/3, 24/1, 24/3; same suites green on v0.1.213). Close every store (closeQuietly: cancel pending writes + sidecar close) before rmrf. closeQuietly is idempotent-safe where flushAll already closed the sidecar.
v0.1.214 release review(master ← 2026-11-13_release-v0.1.214,616 files / +54981 / -12993)这是啥问题按"上一版本 v0.1.213 → 本版本"的大 diff 整体 review(不是逐 PR 看),重点查 dsh/性能/缓存、session msgid、存储四块的回归。结论:发现 3 个回归,全部已在本分支修复并验证;其余 dsh/性能/缓存、session msgid、存储主线均确认无回归。真模型回归门(bili gate,A 压缩 / B 切换 / U registry / restart / plugin 五段)对照 v0.1.213 基线跑,结论见文末。 根因与修复(3 个回归)1. e2e 客户端车道全挂:pure-SQLite 默认值把 JSONL 会话文件停了,但 e2e 断言还在读 JSONL
2. Windows 单测分片确定性挂:测试打开 index.db 后从不关句柄
3. ubuntu 单测挂:notice-registry 测试没防 CI 专属 storage preload
dsh / 性能 / 缓存(专项,无确认回归)
session msgid / block 生命周期(专项,无回归)
存储(专项,无 P0 回归;3 条建议关注)
真模型回归门(bili gate,本机 2 模型,qwen/glm 双端点,五段:压缩 / 切换 / registry / install+update / restart)对照组 v0.1.213 vs 候选 v0.1.214,两轮均完整跑完(约各 4.5h)。总体 verdict 两侧都是 RED —— 但逐段拆开看,RED 完全来自基线自身的既有失败(模型服务间歇空回复 / 超时),v0.1.214 没有任何新增回归,且多数段严格更优:
restart 段 27 vs 32 的逐项人工核对(这是唯一"变少"的段,不能只看数字):
唯一需要说明的 gate 侧发现(非 proxy 回归):plugin 车道 "plugin lane claimed" 断言在候选侧失败(gate 脚本扫 JSONL 会话文件找 结论:可以合并吗可以。 三个回归全部修复(本分支 af49339 / cc47215 / 7c152dc 三个 cherry-pick,另见配套 fix PR),全部 e2e 车道 + ubuntu/windows 单测 + build 在最终 head 全绿;真模型门相对 v0.1.213 基线无新增失败(见上)。合并由人工执行(禁非人类合并)。 URL
最终状态(发评时)
|
…ing the latest run at the sha (ranxianglei#6552) v0.1.214 one-click release: after the bot opened the fallback PR ranxianglei#6552, GitHub auto-created pull_request runs at the publish sha. Because the PR author is the github-actions app, those runs sit behind manual approval (ranxianglei#772) and complete 'action_required' without executing a single job — but with run ids NEWER than the workflow_dispatch compensation runs fired by release-manual.yml. latestRunAt() sorted by run id and returned runs[0], so the gate picked the approval noise as 'the latest run at the publish sha'. classify() has 'action_required' in RED_CONCLUSIONS, so --wait refused 26 seconds after the PR opened and the 60-minute supervision never ran; --dispatch would equally have skipped lanes whose only run was an action_required run. Treat terminal action_required runs as non-runs: the gate falls through to the run that actually executed, or to 'missing', which dispatches. per_page 5->10 so a stack of approval noise cannot mask a real run. Three regression tests against the mock API.
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) 我独立核验了这份 review 的关键结论,并发现一个 review 发布后分支已前移 的情况,需要说明。 已核验(review 结论成立)
关键新发现:review 发布后分支已前移
新增高危内容抽查(非全量重审,只点高危域)
结论(可合判定)
|
…mark fastpath (#5403/#5418 regression) The #4753 input-release (bodyText = undefined before streaming) made the fastpath gate in prepare-openai (bodyText !== undefined && ...) unreachable: the watermark fastpath silently never fired, and plugin wire sessions paid the full reparse+rehash on every turn. Release the text only on the slice where the fastpath cannot fire (non-openai wire or flag off); the strict gate (pluginMode + !titleGen) in prepare-openai stays authoritative. Also update the A/B pin: #5687 moved the dsh single-shot sidecar distinction from "tool-less" to the sidecar budget (hasSingleShotBudget, <=8192), so the test's 64K tool-less turn-0 is now (correctly) a main turn — model the real sidecar with a 4K budget to keep pinning "side lane never stamps".
The five 'HTTP CCR ... fails closed after cold disk restore' tests corrupt and re-read a .content-store.json envelope, which the pure-SQLite default writer (2026-11-13) never emits — ENOENT on sessions/<protocol> turned the e2e-dsh-plugin-lane red on this branch. Pin BILI_STORAGE_JSON_WRITER=1 (restore in finally) like the other e2e lanes (d252543/74a925db); the tests still delete index.db before the cold restore, so the fail-closed 409 contract is unchanged.
|
CI fix: e2e-dsh-plugin-lane was red (run 43548669714) because of the pure-SQLite storage default (2026-11-13): the five 'HTTP CCR … fails closed after cold disk restore' tests in tests/plugin-fork.test.ts corrupt and re-read a .content-store.json envelope, but the default writer no longer emits JSON envelopes, so Fix in b604ff1: pin This fix also lands on master automatically when this PR merges (branch = master + release commits). |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) 更新:当前 head CI 实测(直接查 GitHub check-runs,镜像缓存当时是旧的没刷新)
关键判断:这是 review 之后新引入的回归
我卡住的说明
结论(可合判定,改)
合并仍由人工执行(禁非人类合并);当前状态:拦下,等分片 3 修绿。 |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) 收到你的 CI 修复报告(b604ff1,plugin-fork CCR cold-restore 5 个测试)。我交叉核验了分支与 CI 当前状态: 分支状态(已 fetch 最新)
CI 当前状态(直接查 check-runs,head b604ff1)
判断
合并仍由人工执行;当前状态:CI 未跑完,暂不能判定。 |
'child persists its own companion and survives a proxy restart (#1341)' counts the .content-store.json companion envelope on disk, which the pure-SQLite default writer (2026-11-13) never emits (0 !== 1 — red in the e2e-dsh-plugin-lane, run 43550710438). Pin BILI_STORAGE_JSON_WRITER=1 (restore in finally) like the other e2e lanes (74a925d/d252543a and the plugin-fork pin in b604ff1). Full lane verified locally: 12 suites, 165 tests, 0 failures.
|
Second CI fix: b604ff1 turned 11/13 green; the two remaining reds:
|
(ai: vllm-qwen/qwen3.8-27b) Confirmed the "Second CI fix" — I verified both fixes locally on head
State as of this run:
八问:
收工判定:两处红的处置我已本地核验(①修复已上分支且本地全绿;②按 flake 观察合理)。无需我再推改动; |
|
为什么没有release报告 |
The e2e-fake-codex lane's #2195 test failed in CI (run 43557997361) with ERR_SQLITE_ERROR 'database is locked' from the very first statement of a fresh proxy: the machine-global $XDG_DATA_HOME/.../index.db is written by every bili instance on the box (#5082 supported topology), and a peer still holding the file (draining proxy, sibling job on the same runner) refused the startup outright: - the pre-migration quick_check ran before any busy_timeout existed on the connection -> raw SQLITE_BUSY instead of a bounded wait; - deeper: PRAGMA journal_mode needs exclusive access and BYPASSES the busy handler entirely - with the file in its legacy journal mode and a peer in a write transaction it throws at 0 ms even with busy_timeout=5000 installed (deterministic repro now in tests/storage-open-busy-lock.test.ts, which fails pre-fix). Fix, per the #5082 owner decision that a live foreign writer never blocks startup: install busy_timeout right after open; skip the journal switch when the file header already declares WAL (strict no-op, plain file read); defer it to the next open when the switch comes back BUSY. Pinned by tests/storage-open-busy-lock.test.ts (child-process lock holder; legacy-journal shape fails pre-fix, WAL shape guards the skip).
These lanes pinned XDG_CONFIG/CACHE/STATE per test but inherited XDG_DATA_HOME, so their spawned proxies wrote their session index to the machine-global sessions dir. The CI jobs are bare runs-on: ubuntu-latest (no container) and GitHub's shared-pool runners run several jobs of several repos side by side under the same user - every concurrently running bili proxy on that runner shares one index.db. #2195's t3 launcher burned its 60 s health budget on exactly this. Every other e2e lane already pins a per-test data dir (e2e-native-pi/dsh/opencode/pi- codex-ws); pin the same shape in e2e-codex-fake/-persona/e2e-codex.
|
第三个 CI 失败修掉了,只触发了失败的 lane。 失败: 根因(两层):
修复:
验证:本地 |
(ai: vllm-qwen/qwen3.8-27b) 第三个 CI 修复已本地核验(head 现
关于 @Dog 问的「为什么没有 release 报告」: v0.1.214 的 release PR 还在 CI 修复阶段、尚未合并,所以还没有 GitHub Release / 发布产物——GitHub Release(含 release notes)是 PR 合并后由 CI 自动创建的,现在还没到那一步。 另外, 如果你指的是别的「release 报告」(比如 PR 正文的 "Changes since vX" 清单、或发布后的 GitHub Release 文案),说清楚要哪种,我来补。 八问:
收工判定:第三个 CI 修复(库代码 + 测试 pin)已本地核验(新单测 2/2 全绿、diff 合理); |
c1355dc added tests/storage-open-busy-lock.test.ts, which spawns lock-holder child processes, but never registered it in the shared LIVE_FILES table — the run-ci-tests.mjs tripwire (#1248/#3182) dies before any test runs, so every unit shard on every OS fails at the guard step. One-line registration; suite itself passes locally (2/2, hermetic mkdtemp, no port binds).
|
大面积 CI 失败原因 + 修复(commit acc7944) 现象:merge master(be980cc2f)后,ubuntu/windows 全部 15 个 unit 分片在测试开始前就挂了。 根因:c1355dc18(sessions index open survives a concurrent writer, #2195)新增的 修复:acc794423 一行注册(附并发假设注释:锁持有者是子进程、hermetic mkdtemp、不绑端口)。本地验证:guard 检查 PASS + 该套件 2/2 通过。 新分片跑起来后如有真实失败再报。 |
v0.1.214 大 Diff 回归 Review —— 最终结论范围:v0.1.213 → v0.1.214 整段大 diff(616 文件 +54981/-12993,448 commits),以及后续合入 master 的 82 个新提交(36a04a7 线)。 结论:发现 5 个回归,全部已修复并有对应验证;其余各专项未发现回归。CI 当前全绿,可以合并。发现的回归与修复链
另有 1 个 CI 守卫问题(非产品):master 合入的新测试 四个专项(本 PR 范围)
Gate 对照(真模型,8299)
master 轮全部红灯 = 与基线同机制的环境类失败(模型空回复/超时、dsh 的 pi-ai 客户端 CONTEXT_WINDOW_EXCEEDED 自伤)或已知 #4;无新增产品回归。master 轮最难的持久性断言全过:pi 杀进程后恢复出站压缩 41.7KB vs 原始 127.9KB(24/25,仅中窗自校准一项环境类失败);dsh 跨重启 cache 恢复 100%(38080/38094)。 门禁 harness 修复记录(~/projects/bili-gate)
最终状态
|
…n wire log plugin-lanes.mjs — bring the lane script forward to the state that ran the ranxianglei#6552 full gate (the repo copy predated it): - plugin claim is read from the sessions index.db metadata FIRST and the JSONL envelope scan is kept only as a fallback. Pure-SQLite has been the storage default since 2026-11-13, so JSONL-only claim checks could go red on a perfectly healthy session (that is exactly what the ranxianglei#6552 gate review found — the data sat in index.db while the lane asserted on the JSONL that was never written). - CODEWORDS / buildCorpus are exported: restart-lanes.mjs reuses the same corpus generator (commit 2). - requireSync via createRequire (node:sqlite is loaded lazily, only when a candidate session actually has an index.db). - keep the recorder-spawn comment for the 2026-11-14 wedged-tee fix. upstream-recorder.mjs — record a capped 256KiB rolling tail of each response body plus per-request ms: usage / cached_tokens live in the FINAL stream chunk, and Lane R's cache-hit assertions read them from the recorded tail. Sensitive headers (authorization / x-api-key / cookie / ...) stay masked — bili forwards the model API key into this hop on every request.
Auto-fallback from the one-click release workflow: direct push to master was blocked by branch protection. Merging this PR publishes via the standard release.yml flow.
Changes since v0.1.213
test: isolate XDG_DATA_HOME in the three codex e2e lanes
fix: sessions index open survives a concurrent writer ([Bug][Windows Codex] WAL-only 提交时 SQLite 合并选错世代:顺序启动新增 thread 被移入 conflict #2195 CI gate)
test: pin dual-write storage in fork-ccr-adoption companion test
fix(update): never pre-create the install backup before the atomic swap ([Bug][update] In-place self-update is not atomic: if the process dies mid-copy, the install directory ends up with a mix | [Bug][update] 原地自更新不是原子的:拷到一半进程死掉,安装目录会新旧混杂(或连入口都没有),而备份从来不会被拿来恢复 #6309 regression)
test: pin dual-write storage in plugin-fork CCR cold-restore tests
fix: keep the raw body text available to the plugin full-resend watermark fastpath ([Bug][dsh] 20k-message / 81 MiB-inbound session: host aborted 6× in 2 h — the #3050 resident-AST OOM is not bounded by
sessions.hardMaxBytes(Windows, 0.1.210) #5403/[Bug][dsh] Plugin lane performs a full re-parse and resends the entire body every round (81 MiB → hundreds of MB AST), w | [Bug][dsh] plugin 车道每轮全量解析重发 body(81 MiB→数百 MB AST)是 #5403 OOM 的生产者:用水位线断言替换逐条复核,服务端只解析新增尾段 #5418 regression)test(pi lane: no-proxy direct send is unguarded — oversized raw history egresses and blows the provider window #5025): reset the native origin waiter so the suite's static pi-native import cannot leak INLINE_ORIGIN egress
test(persist): close store sidecars before tmp-dir teardown (Windows EBUSY)
e2e: pin dual-write storage in dsh-persona lane too
tests: unpin CI storage-baseline env in notice-registry suite
e2e: pin dual-write storage in client lanes + tolerate half-deleted install dir
test(notices): delete the CI baseline's BILI_STORAGE_UNIFIED override for this suite
release-ci-gate: ignore approval-gated action_required runs when picking the latest run at the sha ([PR] release v0.1.214 #6552)
Revert "sync: decide block liveness by coverage shape, not existence"
fix(storage): mysql archive restore keeps the NEWEST matching dump line
Revert "kernel: bridge span-safe gaps at the merge gate so mega-turn islands survive minCompressRange (production 2026-11-09 no-ranges incident)"
test([Bug] dsh 车道切模型 fork 出的
_sub:会话不继承父会话 fold 块:上游上下文从 ~83K 瞬间膨胀到 794K、命中归 0、节省转负(0.1.187/0.1.188) #2817): turn 1 carries the pinned one-shot budget for the narrowed residual ([PR] fix: unknown dsh one-shot sidecars steal the raw-key anchor and kill compression (context-explosion root cause) #5687 follow-up)fix(lock): restore root package version 0.1.213 (sed hit the wrong 0.0.196 line during the kernel bump)
kernel: version 0.0.197 (batch4 kernel fixes: [PR] fix(kernel): heal unpaired server_tool_use / *_tool_result blocks after folds #6428 server-tool pair heal, [PR] kernel: bridge span-safe gaps at the merge gate so mega-turn islands … #6345 span-safe gap bridging, [PR] sync: decide block liveness by coverage shape, not existence #5166 liveness-by-coverage)
chore: re-run kernel guard (kernel-change declaration added to PR body)
docs(governance): announcement policy — no broadcast-to-all, owner confirmation, blast-radius PRs, plugin notices trigger-only ([notice] The dsh-live-compact plugin is not installed, yet a "dsh-live-compact conflicts with bili compression" notice s | [notice] 没装 dsh-live-compact 插件,却出现「dsh-live-compact 与 bili 压缩冲突」公告 #6508)
retract: withdraw the dsh-live-compact announcement from the remote feed ([notice] The dsh-live-compact plugin is not installed, yet a "dsh-live-compact conflicts with bili compression" notice s | [notice] 没装 dsh-live-compact 插件,却出现「dsh-live-compact 与 bili 压缩冲突」公告 #6508)
docs: seed the requiresScan text in config-reference so docs-sync regenerates it
fix(notices): gate announcement records on the live conflict scan via requiresScan ([notice] The dsh-live-compact plugin is not installed, yet a "dsh-live-compact conflicts with bili compression" notice s | [notice] 没装 dsh-live-compact 插件,却出现「dsh-live-compact 与 bili 压缩冲突」公告 #6508)
chore(pi-subagents): bump version 0.2.3 -> 0.2.4 for the fleet-inspector CRLF fix
docs(architecture): add src/server/rejection-diagnostics.ts to the module map
test(notices): pin the sqlite-default notice gate (resolves unused-exports)
storage: pure-SQLite default — JSON envelopes stop being written unless dual-write is requested (owner decision 2026-11-13)
test(review): pin the documented explicit "auto" spelling of BILI_PLUGIN_BYPASS (x-bili-plugin-bypass: 1 Allow forced specification | x-bili-plugin-bypass: 1 允许强制指定 #6220)
docs(review): add ja docs rows for BILI_PLUGIN_BYPASS (ja work [PR] docs: add Japanese versions of repository documentation (#6072) #6076 landed on master during rebase)
fix(review): one-time warn when BILI_PLUGIN_BYPASS forces the bypass decision (x-bili-plugin-bypass: 1 Allow forced specification | x-bili-plugin-bypass: 1 允许强制指定 #6220)
test(review): pin + document the unrecognized-value fallback of BILI_PLUGIN_BYPASS (x-bili-plugin-bypass: 1 Allow forced specification | x-bili-plugin-bypass: 1 允许强制指定 #6220)
fix(handoff): stop coupling the handoff canSave to res.destroyed ([PR] fix(handoff): persist a completed reply when the client hangs up #4861)
fix(review): cite GitHub x-bili-plugin-bypass: 1 Allow forced specification | x-bili-plugin-bypass: 1 允许强制指定 #6220, not the mirror number [Bug][storage] Verification digests ignore own __proto__ fields in retained JSON content #6222, in BILI_PLUGIN_BYPASS artifacts (x-bili-plugin-bypass: 1 Allow forced specification | x-bili-plugin-bypass: 1 允许强制指定 #6220)
fix(loop): restore the streamedMarkers push the Plugin lanes: host declares exposed bili tools per request via x-bili-plugin-exposed-tools (structural fix for #3884/#3600 name-guessing) | 插件车道:宿主按请求声明暴露的 bili 工具(根治 #3884/#3600 认名字脆弱性) #3906 strip mangled
fix: drop unused export from PluginBypassOverride type (x-bili-plugin-bypass: 1 Allow forced specification | x-bili-plugin-bypass: 1 允许强制指定 #6220)
docs: register BILI_PLUGIN_BYPASS in the lifecycle ledger (x-bili-plugin-bypass: 1 Allow forced specification | x-bili-plugin-bypass: 1 允许强制指定 #6220)
feat: BILI_PLUGIN_BYPASS operator override for the x-bili-plugin-bypass passthrough (x-bili-plugin-bypass: 1 Allow forced specification | x-bili-plugin-bypass: 1 允许强制指定 #6220)
fix(test): match the [PR] fix: release input-only request data before streaming #4756 release hook to forward()'s master signature
fix(test): repair comment/test join dropped by the [PR] fix: forward toolless structured-output requests as side requests #5074 union in the lane-invariant pins
fix(handoff): complete fix: release request input buffers before long-running responses #4753 release-wiring dropped by a marker slip
test: pin the full Codex exec note constant in the Codex exec/namespace ACP hints miss direct-call path when native MCP binding is not ready #5592 regression tests
fix: teach direct ACP calls for Codex exec namespaces
kernel: heal unpaired server_tool_use / *_tool_result blocks before forwarding
fix(generation-boundary): arm the native-compaction rebase on a proven ref-ledger reset
kernel: bridge span-safe gaps at the merge gate so mega-turn islands survive minCompressRange (production 2026-11-09 no-ranges incident)
fix(test): remove duplicate buildSessionCacheReport import
test: pin [storage] The newly created index.db actually ends up with auto_vacuum=0: this pragma is placed after journal_mode=WAL, | [storage] 新建的 index.db 实际是 auto_vacuum=0:这条 pragma 排在 journal_mode=WAL 之后,
--vacuum incremental什么都回收不了 #6323 auto_vacuum-before-WAL pragma orderingtest(loop): restore OpenAI Routing: Requests where the upstream omits cache fields are misrecorded as "unmeasured" and excluded from attribu | OpenAI 路由:上游省略缓存字段的请求被误记为"未测量"并剔除出归因,命中率与归因面板差 12.6pp 无法解释 #6329 regression — OpenAI cache-field omission is a measured miss
perf(storage): skip rewrites when large session columns are unchanged
test: regenerate storage-render goldens for OpenAI cache attribution change
fix(update): prefer atomic directory swaps during in-place updates
fix(storage): serialize archive exports and restore the newest dump
fix(openai): treat omitted cache fields as zero cached tokens
fix(Replay a stream cut that lands after reasoning-only output #5951): commandcode truncation record uses single-newline JSONL terminator
fix(Replay a stream cut that lands after reasoning-only output #5951): let the OMP host own truncation recovery, keep diagnostics local
test(win-cmd): pin the comspec trailing-backslash round-trip in CI
fix(win-cmd): double trailing backslashes when quoting for cmd.exe
docs(omp): document exported PI_ANTHROPIC_MESSAGE_THREADS default ([Bug][omp] OMP ≥ 18.18 sends Claude turns as Anthropic message threads →
bili ompand the omp native lane never compress (fix: PI_ANTHROPIC_MESSAGE_THREADS=0) | OMP 18.18 起 Claude 走 message threads,每轮只发最新一条 → bili omp 不压缩 #6088)fix(omp): default PI_ANTHROPIC_MESSAGE_THREADS=0 so bili sees the full history
test: pin both ends of the dsh 0.2.2-rc.1: auto-review sidecar no longer matches isDshAutoReviewRequest — reviews ride the main pipeline and tombstone the fold ledger as DEAD #6104 head window
test: pin the dsh 0.2.2-rc.1: auto-review sidecar no longer matches isDshAutoReviewRequest — reviews ride the main pipeline and tombstone the fold ledger as DEAD #6104 review shape in the side-lane invariants suite (lane-guard dsh 回归防护:泳道判定真值表 + 侧请求不变量 + dsh e2e lane + 运行时金丝雀 + lane-guard(5 项) #2170)
fix: recognize dsh 0.2.2's auto-review sidecar shape (dsh 0.2.2-rc.1: auto-review sidecar no longer matches isDshAutoReviewRequest — reviews ride the main pipeline and tombstone the fold ledger as DEAD #6104)
fix(google): promote stripped proxy call thoughtSignature onto surviving sibling functionCall ([Bug] Antigravity Google lane: post-fold remaining functionCall missing thoughtSignature → sticky 400 (gemini-4-argon-medium) #6079, [Bug] Retained functionCall after Google lane folding loses thoughtSignature, guaranteed 400 on the next turn (gemini-4- | [Bug] Google 车道折叠后保留的 functionCall 丢 thoughtSignature,下一轮必 400(gemini-4-argon / Vertex) #5252)
fix: preserve position fingerprints when reaping folds
fix(opencode): ride REAL host msgIds into the incremental lane from both seams
test(rolling-marks): compare instances on substance, not wall-clock tag times
fix(incremental): continue repeated-content IDs across delta turns
fix: create dist/agent in write-build-info so the shim write is self-sufficient
docs: drop stray closing paren in delegate-notify header comment
test: align remaining pi-lane e2e + docs to the billion-context.js entry
fix: identify billion-context in pi extensions
test: update stale side-lane comments and pin hasSingleShotBudget boundary rows
fix: align the single-shot sidecar doc comments with the narrowed residual
fix([PR] fix: unknown dsh one-shot sidecars steal the raw-key anchor and kill compression (context-explosion root cause) #5687): a tool-less MAIN turn must not be mistaken for a dsh sidecar
chore: drop the full-request hash from WS cache diagnostics
chore: add redacted Responses WebSocket cache diagnostics
test(ci): issue5152 startReceiver waits for a valid port value, not file existence (CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599)
fix(diagnostics): strip C1 control chars (incl. NEL) from the bounded rejection cause
fix(diagnostics): distinguish size rejections and retain bounded causes
fix(loop): classify proxy tool failures by status line
fix: report nested transport timeout diagnostics without assuming idle expiry
sync: decide block liveness by coverage shape, not existence
fix(hermes): route a plain-http provider base_url through the proxy ([bug][plugin] hermes native lane never routes a plain-http provider base_url (only HTTPS_PROXY is set) #5116)
fix(hermes): align the Hermes lane: proxy adds ACP tools to toolless structured-output requests from child processes, which strict backends reject (400) #5073 side-call fix with [bug] bili dsh 长会话中 max_output_tokens 被压到 ≤200,主请求被误判为 side request 并绕过压缩 #546/标题生成等侧请求污染主会话状态:prepareAnthropic 缺 isTitleGen 守卫、快照无条件覆盖、survivedCount 每请求空转 #388/[bug] Claude Code 的 WebSearch 子请求被当成主对话:nudge 基准被搜索的 usage 重锚,导致提前折叠;每次搜索还被注入 bili 工具 #2500
chore: anchor the new side-request rule comments to Hermes lane: proxy adds ACP tools to toolless structured-output requests from child processes, which strict backends reject (400) #5073
fix([PR] fix: forward toolless structured-output requests as side requests #5074): drop the unused export from requestsStructuredOutput
fix: forward toolless structured-output requests as side requests
fix: arm the pi lane's direct-send window guard so a proxy outage cannot egress an oversized history (pi lane: no-proxy direct send is unguarded — oversized raw history egresses and blows the provider window #5025)
fix(plugin): announcement-tail detector must read answer prose, not proseAcc (bili forwards all x-bili-plugin* host headers (and its own chain markers) to upstream model endpoints — egress strip gap | bili 把全部 x-bili-plugin* 宿主头(及自身链标记)转发给上游模型端点——出站剥离缺口 #3905)
test(issue3905): cover plugin chat/responses lanes like the DSH native mode: assistant text ending with </summary> is treated as a completed step, so the turn silently stops (149 samples from session transcripts) #2303 sibling; pin the 400-char cap
fix(degenerate-turn): implement the action-cue + closed-fence gates the PR body describes
refactor(loop): strip Plugin lanes: host declares exposed bili tools per request via x-bili-plugin-exposed-tools (structural fix for #3884/#3600 name-guessing) | 插件车道:宿主按请求声明暴露的 bili 工具(根治 #3884/#3600 认名字脆弱性) #3906 streamMarkers hunks — owned by PR [PR] feat(compress): add streamMarkers setting to suppress streaming receipts (#3906) #4934
fix(loop): retry stalled bare-announcement terminal turns (bili forwards all x-bili-plugin* host headers (and its own chain markers) to upstream model endpoints — egress strip gap | bili 把全部 x-bili-plugin* 宿主头(及自身链标记)转发给上游模型端点——出站剥离缺口 #3905)
fix(reconcile): recover exact user text after line wrapping
test: address instruction-preservation review findings
fix: preserve changed user instructions during reconciliation
fix(handoff): drop leftover conflict markers (merge slip)
fix(handoff): persist a completed reply when the client hangs up
fix: release input-only request data before streaming (fix: release request input buffers before long-running responses #4753)
fix(window): size a child conversation from its parent's runtime-info ([Bug][wiring]
:side:child conversations resolve the registry/#1321 cap instead of the parent's runtime-info window #4682)fix(pi): normalize Fleet CRLF and paint truncated rows
fix(overflow): recognize the DashScope input-length-range 400 so the probe path can arm ([BUG] DashScope compatible-mode 400 "Range of input length should be [1, N]" is not recognized as an overflow → the #2313 probe path livelocks #4262)
test(5631): bill the fake upstream at token scale with a disjoint anthropic cache partition
docs: root redirect stubs for Japanese + ja links in docs indexes (All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072)
docs: add Japanese version of configuration.md + generator/guard/CI support (All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072)
docs: ja hub/module pages link to ja hub; add English link in ja hub
docs: add Japanese architecture hub + 8 module pages (All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072)
docs: add Japanese version of technical-notes.md (All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072)
docs: add Japanese version of clients.md (All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072)
fix: README.ja.md — use standard Japanese term ランディングページ for 'landing page'
fix: message-identity.ja.md — close full-width bracket in '(rollout gate)' summary line
docs: add Japanese version of the docs directory landing page (All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072)
docs: message-identity.ja.md — close full-width bracket in heading (Snapshot as session reconstruction: identity claim order takes priority, message hash demoted to reference signal (root- | 快照即会话重构:身份认领顺序优先,消息哈希降级为参考信号(根治 #3588/#3640 一类误判) #3641)
docs: session-identity.ja.md — keep TRUNCATION_LINEAGE_WINDOW code span on one line
docs: add Japanese translations for message-identity & session-identity (All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072)
fix: opencode-responses-websocket.ja.md — restore dropped 'responsesCodec' registration name
fix: msgid-infrastructure.ja.md — space after '5.' so item 5 renders as a list item
docs: un-cross README row labels in docs/index.zh-CN.md (English label pointed at the Chinese file)
docs: align identity-channels.ja diagram indent with EN (byte-identical fence)
docs: add Japanese translations for identity & protocol docs (All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072)
test: adapt zero-fold anonymous arbitration pins to master's measured-body gate ([Bug][dsh][0.1.208] Kill switch / egress fuse only measures message.text: a session with 26,903 messages and only 1.44 M | [Bug][dsh][0.1.208] 判死/egress fuse 只量 message.text:26,903 条消息、core text 仅 1.44 MiB 的会话绕过 fuse 发出 55.2 MiB 请求体,被上游 413 拒收(ModelArts.… #5000/[Bug] CJK payloads slip past the egress byte-floor fuse: a 4.1 MiB outbound was forwarded at 1,121,674 tok into a 1,000,000-token window (0.1.206, plugin lane) #4543, 2e02110)
docs(preflight): restore the image-reserve rationale thinned by the budget repair
fix(preflight): account for images removed by folding
fix(storage): hash the split media form on both verification legs
fix(storage): align the envelope tail leg to the row tail by suffix
fix: name the real cause in the proven-share log when codex netting skips the reservation (review nit for [PR] fix: do not reserve output headroom twice on a codex live-table window #6165)
refactor: resolve the headroom compress config once per request (review nit for [PR] fix: do not reserve output headroom twice on a codex live-table window #6165)
fix: do not reserve output headroom twice on a codex live-table window
docs: describe the watchdog fatal path as a whole-process SIGKILL, not process.exit(1) ([Bug] event-loop watchdog never kills the proxy: process.exit() in a worker thread ends only the worker, so the wedged proxy keeps the port and runs on unwatched #6157)
fix: event-loop watchdog kills the whole process, not just its worker
fix(preflight): preserve calibrated fit after zero folds
fix(codex): reconcile opaque native compaction checkpoints
fix(responses): preserve block parts through projection rebuilds
fix(responses): bind surviving parts to their original carrier
fix: preserve multipart Responses user runs in host encoding
docs: add Discord badge to zh-CN and ja README headers
docs: use Discord badge icon in header instead of plain text
docs: add Discord community link to README
fix(stream): arm the ledger align only after a compress created a block
fix(storage): set auto_vacuum before journal_mode so a new index.db really gets INCREMENTAL
fix(update): run the boot smoke when the dsh desktop copy updates itself in place (dsh中工具会突然消失 #2082)
fix(storage): no archive line for a failed sweep; restore the newest matching entry
fix(advisories): cap id/reason length in publish validation, mirroring the client caps ([Bug][advisory][update-notes] Remote
reason/summaryhas no length limit, preserves newlines, and is passed into ac | [Bug][advisory][update-notes] 远程的reason/summary不限长、保留换行,原样进到 acp_status(模型上下文) #6306)fix(advisory,update-notes): cap length and collapse line breaks of remote reason/summary
test(block-ops): pin the acp_rewrite net-shrink 'never block an improvement' exception ([Bug][blockOps] The summary stored by acp_rewrite is not subject to compress's length min/max bounds or net-shrinkage ch | [Bug][blockOps] acp_rewrite 存进去的摘要不受 compress 的长度上下限和净收缩检查约束;blockId 不接受数字 #6319)
fix(block-ops): hold acp_rewrite to the compress summary bounds and accept a numeric blockId
fix(dsh-ledger): ignore the pnpm peer suffix when comparing lock versions and keep it when rewriting
kernel: version 0.0.196
fix(prompts): teach the canonical string form in compress call examples
test: pin ingestRowsNow — rows durable before the envelope, stale guard stays observable (Compression fails in long sessions while new sessions work fine | 压缩失败.新会话反而没事 #3397)
fix: point the durability-before-render comments at Compression fails in long sessions while new sessions work fine | 压缩失败.新会话反而没事 #3397 (PR [PR] fix: persist ledger rows before the model call (durability-before-render) #6326), not the closed unrelated [PR] remove: the experimental compressAtFoldPoint feature — fold-point relocation (v1 #5471 + unreleased v2 #5949), #6132 #6150
fix: keep the multi-instance stale-write guard observable in the durability path
fix: persist ledger rows before the model call (durability-before-render)
fix(ci): preserve live fork PR runs during janitor cleanup
fix(storage): match KV key prefixes literally
perf(affinity): construct only sampled SimHash shingles
fix(render): avoid argument limits when locating large fold anchors
fix(storage): skip non-object JSONL entries when locating archives
fix(summary): keep exponential backoff for blank Retry-After headers
perf(storage): append split media without copying prior entries
fix: enforce readable byte limit before credential store replacement
fix(storage): treat malformed integrity caches as misses
fix: exclude tool data from keepalive TTL detection
fix(storage): preserve extension timestamps across archive restore
fix: recognize valid JSON layouts in raw media accounting
fix: redact JSON-escaped home paths in published diagnostics
fix(storage): include prototype-named JSON fields in verification digests
fix: recover Responses turns with string input
fix: expand Responses WebSocket history in linear time
fix: preserve deferred WebSocket listener options and close state
fix: honor leading Windows proxy bypass wildcards
test: accept platform-specific export destination errors
fix: honor output paths for JSONL exports
fix: bound uncovered message range queries
fix: accept completed summaries with empty tool call arrays
fix: enforce extension storage validation across backends
test: scope archive permission checks to POSIX
fix: replace archive logs atomically after restore
fix: restore archive content ownership per reference
docs: register shared TOML integer reader in architecture map
fix(config): apply strict TOML integers to sibling model readers
fix(kimi): preserve TOML integer model limits
docs: record BMP DIB field offsets in bmpDims ([Bug] BMP CORE headers are assigned the wrong dimensions and pixel-token estimate #6179)
fix: ground BMP CORE dimensions in encoder output
fix: correct CORE header offsets so real spec-valid BMPs parse ([Bug] BMP CORE headers are assigned the wrong dimensions and pixel-token estimate #6179)
fix: decode BMP dimensions by DIB header layout
fix: select image downscale work after archiving
fix: limit Google inline media shedding to declared images
docs: point the range-spill comment at bili goose: launcher discovers only config.toml + custom_providers/*.toml — Goose ships config.yaml + <id>.json, so the lane silently runs uncompressed #6191 (was [Bug] Automatic decompression spills can overwrite previously returned files #6175, an unrelated slow-ingest issue)
docs: restore feat: CCR v2 — covered-ref spans, range-level decompress, ID-returning search_context; CCR default-on (#1179) #1207-F4 provenance and 0600 rationale in the range-spill comment
fix: give every automatic decompression spill a unique file
fix: bracket IPv6 addresses in port-holder health probes
fix: stabilize identity for requests carrying multiple credentials
fix: index Gemini inline images referenced by function responses
test(google): pin the image/* note gate negative direction for nested tool media ([Bug] Gemini tool-response screenshots have zero image cost and disappear from summary notes #6146)
fix(google): account for images nested in tool responses
fix(server): validate catalog identity before replaying 304 responses
fix(storage): retain malformed records during session garbage collection
fix: repair overlapping tool groups with a linear sweep
fix: classify native compaction only from complete SSE events
test([Bug] Native compaction treats non-completed JSON output as a successful rebase #6135): pin that a completed claim carrying a real error still fails
fix: honor native compaction JSON terminal status
fix: propagate upstream error-body failures without replay
test: pin leaf feed order in the deep-JSON runaway-guard tests
fix: avoid recursive runaway guard traversal of provider JSON
fix: harden runaway guard NDJSON mode selection; document framing rules
test: pin the NDJSON opening-token gate with a discriminating case
fix: preserve runaway guard frame boundaries across chunks
fix: parse summary SSE data fields as complete events
fix: recognize legal JPEG marker fill during image billing
fix: strip credentials from loopback diagnostic URLs
fix: refresh upstream CA cache on configured path changes
test: register pack installer subprocess suite
fix: select the requested prompt pack from npm dependencies
fix: separate IPv6 socket hosts from CONNECT authorities
fix: repair incomplete image spills when originals are resent
fix: scope image encodes to their resize policy
docs: clarify witness hash index lifecycle
fix: collect expired witness rings during record-only traffic
fix: ignore stale keepalive response bodies after rearming
web: MdRenderApi is module-internal (unused-exports gate)
web: hover-widened tree topics, tick-proof rich nodes, restamp-proof snapshot hydration
web: persistent compression-tree tooltip + frozen-snapshot hydration for pruned leaf messages
fix: settle and reset unified snapshot write batches
fix(guards): lock-sync + lane-guard for the Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 blockOps gate
fix(guard): drop two INTERNAL_ONLY exports left by the Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 mcp-note split
feat: gate the Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 block-lifecycle tools behind compress.blockOps (default off)
web: handoff tab renders the Compressed block summaries appendix as its own section
thirdparty-scan: flag dsh-live-compact as a known conflict ([Bug][compress] After rebase, the ref reported by acp_status/nudge is inconsistent with the ledger used by compress for | [Bug][compress] rebase 后 acp_status/nudge 报的 ref 与 compress 校验用的台账不一致:提示的区间必然 unknown,连续三次即熔断 #5984)
web: eWork-grade handoff document rendering (markdown + syntax highlight + role cards)
test(e2e): key the fake upstream's directive queue on the source ref + tag-stripped body (e2e-native-pi lane flaky red since 2026-11-10: fake upstream's directive-queue key contains volatile ACP render-tag bytes, so a scripted fold replays the queue ("compress must be CALLED exactly once ... got 2") #6160)
docs(architecture): register src/acp-block-ops.ts in the file map (Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765)
fix(review): drop two orphaned imports left by the fold-point removal (Remove the compressAtFoldPoint feature — it reorders resent history and caused an authorship-misattribution incident in a live session #6132)
test: retune payload-calibrated e2e pins for the Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 body growth
test: regenerate storage-render goldens for the Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 tool surface
test: update pinned ACP tool lists + prompt text for the Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 tool surface
remove compressAtFoldPoint from host, web, docs and site (Remove the compressAtFoldPoint feature — it reorders resent history and caused an authorship-misattribution incident in a live session #6132)
kernel: remove the compressAtFoldPoint fold-point relocation feature (Remove the compressAtFoldPoint feature — it reorders resent history and caused an authorship-misattribution incident in a live session #6132)
fix: restore render-tag example stripped from stripTextSlot header comment
test(pf): cover the routed branch for the [Bug][dsh] watermark fastpath never hits: the client never sends x-bili-pf-count because pfRequestHeaders is handed the pre-merge headers object #5813 headersFor-conversation shape
fix(agent): pass merged headers to pfRequestHeaders so watermark count is emitted ([Bug][dsh] watermark fastpath never hits: the client never sends x-bili-pf-count because pfRequestHeaders is handed the pre-merge headers object #5813)
test(hermes-native): pin the never-routed direction of the unload env gate ([PR] fix(hermes-plugin): restore routed process env on unload and watchdog exit (#4644 follow-up) #5991)
fix: fail open when a marker strip would blank the latest user turn (marker strip blanks a user message whose entire text is a
[ACP]line — the model receives a single space #5812)test(hermes-native): regression — unload restores the host's own proxy env ([PR] fix(hermes-plugin): restore routed process env on unload and watchdog exit (#4644 follow-up) #5991)
fix(hermes-plugin): unroute only in _shutdown — the retire branch needs no env work
fix(hermes-plugin): restore routed process env on unload and watchdog exit
refactor: consume the kernel's insertBeforeTrailingSystemRun, drop the host interim copy (Anthropic 400: mid-conversation system must FOLLOW a user message — heal only checks the follower, and trailing nudge/notes land after a trailing system #5432)
test: pin Anthropic 400: mid-conversation system must FOLLOW a user message — heal only checks the follower, and trailing nudge/notes land after a trailing system #5432 adapter fixture to legal-follow shape after the kernel two-sided heal landed (Anthropic 400: mid-conversation system must FOLLOW a user message — heal only checks the follower, and trailing nudge/notes land after a trailing system #5432)
fix: keep post-hoc anthropic body mutations off trailing client systems (Anthropic 400: mid-conversation system must FOLLOW a user message — heal only checks the follower, and trailing nudge/notes land after a trailing system #5432)
fix: insert anthropic ephemeral tails before trailing client systems (Anthropic 400: mid-conversation system must FOLLOW a user message — heal only checks the follower, and trailing nudge/notes land after a trailing system #5432)
fix(ci): align hermetic registry e2e entry assertion with [Bug] Plugin fails to load under OpenCode 2.1.0: bare package name entry dist/index.js lacks a default export (Plugin mu | [Bug] OpenCode 2.1.0 下插件加载失败:裸包名入口 dist/index.js 没有默认导出(Plugin must export a default definition) #3976 2.x shim-dir semantics
fix: cite the upstream QLC field-report issue billion-context Physical Write Volume on QLC SSD: 31-Day Field Test, Four Governance Tiers, Residual Attribution | billion-context 在 QLC SSD 上的物理写入量:31 天实测、四道治理、剩余归因 #6058, not internal mirror number Time-context snapshots injected as role:"user" cause spurious "fake user messages" #6062
refactor: drop unused absPath parameter from coerceModule (# [Feature] Support user-registerable egress request body rewriting plugin pipeline | # [Feature] 支持用户可注册的出站(egress)请求体改写插件管线 #6003)
docs: pipeline module pages describe the egress-transforms request stage (# [Feature] Support user-registerable egress request body rewriting plugin pipeline | # [Feature] 支持用户可注册的出站(egress)请求体改写插件管线 #6003)
fix: satisfy unused-exports gate for egress-transforms (# [Feature] Support user-registerable egress request body rewriting plugin pipeline | # [Feature] 支持用户可注册的出站(egress)请求体改写插件管线 #6003)
docs: list src/egress-transforms.ts in the architecture file map (# [Feature] Support user-registerable egress request body rewriting plugin pipeline | # [Feature] 支持用户可注册的出站(egress)请求体改写插件管线 #6003)
feat: user-registerable egress request-body rewrite pipeline (# [Feature] Support user-registerable egress request body rewriting plugin pipeline | # [Feature] 支持用户可注册的出站(egress)请求体改写插件管线 #6003)
docs(config): correct persist.maxDelayMs writer attribution in configuration.md en+zh (Time-context snapshots injected as role:"user" cause spurious "fake user messages" #6062)
docs: regenerate config docs from seed for persist.maxDelayMs (docs-sync drift)
docs: point stale slow-session constant comment at its new home ([PR] fix(diagnostics): report large histories without attributing stalls #5481)
fix: cancel idle raw upstream reader in reasoning-guard verbatim pipe ([Bug] Cancel idle raw upstream reads on client disconnect #6041)
docs: document persist.maxDelayMs and a physical write-volume observation recipe (Time-context snapshots injected as role:"user" cause spurious "fake user messages" #6062)
fix: wire persist.debounceMs and new persist.maxDelayMs into the session-order and prefix-affinity snapshot writers (Time-context snapshots injected as role:"user" cause spurious "fake user messages" #6062)
fix: teach the billion-context: 400 Bad Request when codex sends
role: "developer"messages — request a request-body compat transform hook #552 role-rejection ladder to recognize serde 422 rejections (for [Bug][Responses] injectResponsesDeveloperMessage overwrites client input[0].role="system" with "developer" (ignores DSH compat.supportsDeveloperRole:false) -> 422 on DeepSeek | Responses 协议未继承客户端头部 role:"system",强制改写为 developer 导致上游 422 #6063)fix: traverse raw-body token estimates iteratively as well ([Bug] Account for deeply nested media without recursive stack growth #6038)
fix(responses): correlate proxy-owned ACP calls by stable identifiers so they never leak to the client dispatcher (Pi subagents: proxy-handled ACP calls leak to the client on Copilot Responses #6056)
kernel: version 0.0.191
kernel: reuse the client's leading message role when rebuilding the Responses input (for [Bug][Responses] injectResponsesDeveloperMessage overwrites client input[0].role="system" with "developer" (ignores DSH compat.supportsDeveloperRole:false) -> 422 on DeepSeek | Responses 协议未继承客户端头部 role:"system",强制改写为 developer 导致上游 422 #6063)
docs: list src/exit-deadline.ts in the reference map (Existing JSON State Files: Inventory and First-Batch Unified Storage Migration | 存量 JSON 状态文件:盘点与第一批统一存储迁移 #5753)
fix(Existing JSON State Files: Inventory and First-Batch Unified Storage Migration | 存量 JSON 状态文件:盘点与第一批统一存储迁移 #5753): per-scheme legacy import for resign-pending hydration
fix(tests): stop server.ts from pulling restart.ts into the static import phase (Existing JSON State Files: Inventory and First-Batch Unified Storage Migration | 存量 JSON 状态文件:盘点与第一批统一存储迁移 #5753)
storage: harden unified-kv callers against partial store stubs + review fixes (Existing JSON State Files: Inventory and First-Batch Unified Storage Migration | 存量 JSON 状态文件:盘点与第一批统一存储迁移 #5753)
storage: move six JSON sidecar states onto the unified kv_store (Existing JSON State Files: Inventory and First-Batch Unified Storage Migration | 存量 JSON 状态文件:盘点与第一批统一存储迁移 #5753)
refactor: name importSnapshot carrier lists after their content ([Bug] Reject corrupt positional affinity snapshot carriers #6037)
fix: reject corrupt tailFrags positional carriers in importSnapshot ([Bug] Reject corrupt positional affinity snapshot carriers #6037)
docs: clarify noteColdLoad contract for missing/invalid savedAt ([Bug] Reject invalid attach timestamps before scheduling alarms #6043)
docs: adopt the [PR] feat: declarable custom-wire model endpoints + commandcode-cli wire family (#1295) #1444 branch's wire-contracts doc wording (relative link + index/architecture lines)
Revert "[PR] fix: classify the dsh image-offload placeholder so it cannot steal the last-user anchor ([Bug][dsh] image-offload placeholder text hijacks the last-user anchor: model treats a two-day-old screenshot as the cur | [Bug][dsh] image-offload 占位文本抢走 last-user 锚点:模型把两天前的旧截屏当成当前提问(bili 0.1.213 / DSH 0.2.2-rc.1) #6021)"
fix: keep commandcode plugin-pipe filter state across deltas
docs(pf-channel): state the LRU eviction policy at the capacity constant
test: refresh storage goldens for corrected SimHash votes
storage: CI gate — jsonWriter=off may only write whitelisted .json in the sessions dir (Identity pipeline: key sessions by (agent, conversation id) in the unified store — decouple fold ledgers from protocol/host origin (#2938 / #2847 / #2671 defect-1) #2960)
storage: jsonWriter=off also stops the CCR content-store sidecar (Identity pipeline: key sessions by (agent, conversation id) in the unified store — decouple fold ledgers from protocol/host origin (#2938 / #2847 / #2671 defect-1) #2960)
docs: relocate WIRE-CONTRACTS ledger under docs/ ([PR] docs: batch-migrate 14 root-level documents into docs/ (consolidation + master index + redirect placeholders) | docs: 14 个根级文档一次性迁入 docs/(集中 + 总目录 + 重定向占位) #5920 layout)
docs(architecture): register the three [Feature] Declarable custom-wire model endpoints — route providers like commandcode's Go plan (/alpha/generate) into compression #1295 modules in the source map
test: reject holes in positional snapshot carriers
fix: tolerate JSON null content during storage identity backfill
fix: evict watermark sessions incrementally at cache capacity
fix: reject invalid attach timestamps before scheduling alarms
fix: honor backpressure and completion when dumping SSE streams
fix: cancel idle raw upstream reads on client disconnect
fix: retain failed dump deletions in GC byte accounting
fix: preserve busy liveness when health response bodies time out
fix: account for deeply nested media without recursive stack growth
fix: reject corrupt positional affinity snapshot carriers
fix: prevent SimHash feature vote overflow
fix: age idle open upstream circuits out of status reports (The DSH upstream header is permanently occupied by the cooled OPEN circuit. | DSH 上游横幅被已冷却的 OPEN circuit 永久占据 #6026)
fix: hydrate web block-detail leaves by stable msg_id, not restamped ref number
test([Bug][dsh] image-offload placeholder text hijacks the last-user anchor: model treats a two-day-old screenshot as the cur | [Bug][dsh] image-offload 占位文本抢走 last-user 锚点:模型把两天前的旧截屏当成当前提问(bili 0.1.213 / DSH 0.2.2-rc.1) #6021): add end-to-end anchor-geometry pin for the offload placeholder (folded in from parallel PR [PR] fix: classify the dsh image-offload placeholder so it cannot steal the last-user anchor (#6021) #6027)
test([Bug][dsh] image-offload placeholder text hijacks the last-user anchor: model treats a two-day-old screenshot as the cur | [Bug][dsh] image-offload 占位文本抢走 last-user 锚点:模型把两天前的旧截屏当成当前提问(bili 0.1.213 / DSH 0.2.2-rc.1) #6021): pin the words-before-marker splice shape for the offload placeholder
docs: changelog entry for hermes top-level base_url discovery (Hermes plugin: MITM-domain discovery misses top-level base_url, so model traffic silently blind-tunnels (no compression) #6022)
fix(hermes): quoted endpoint values in providers maps no longer die silently (Hermes plugin: MITM-domain discovery misses top-level base_url, so model traffic silently blind-tunnels (no compression) #6022)
fix(hermes): launcher route discovery reads the top-level base_url too (Hermes plugin: MITM-domain discovery misses top-level base_url, so model traffic silently blind-tunnels (no compression) #6022)
fix(hermes): discover top-level base_url so stock configs join the MITM whitelist (Hermes plugin: MITM-domain discovery misses top-level base_url, so model traffic silently blind-tunnels (no compression) #6022)
fix: stale-epoch config overlay loses the named-provider recipe table ([Bug][config] 0.1.212: stale-epoch overlay expands externalSummary with empty provider recipes ("known: none") and disables a valid chain #5912)
fix: name the unattributed-chain class in the zero-progress fits-forward warn (The default
modelContextLimitof 200000 underestimates the context window for custom provider routing, and over-window |modelContextLimit缺省 200000 会把自定义 provider 路由的窗口算小,且超窗的「无归因」请求会被照发而不是本地拒绝 #5907)refactor: unify relay recovery re-sends through one egress chokepoint (# [Feature] Support user-registerable egress request body rewriting plugin pipeline | # [Feature] 支持用户可注册的出站(egress)请求体改写插件管线 #6003)
fix: classify the dsh image-offload placeholder so it cannot steal the last-user anchor ([Bug][dsh] image-offload placeholder text hijacks the last-user anchor: model treats a two-day-old screenshot as the cur | [Bug][dsh] image-offload 占位文本抢走 last-user 锚点:模型把两天前的旧截屏当成当前提问(bili 0.1.213 / DSH 0.2.2-rc.1) #6021)
fix(dsh): classify the image-offload placeholder as a one-shot carrier ([Bug][dsh] image-offload placeholder text hijacks the last-user anchor: model treats a two-day-old screenshot as the cur | [Bug][dsh] image-offload 占位文本抢走 last-user 锚点:模型把两天前的旧截屏当成当前提问(bili 0.1.213 / DSH 0.2.2-rc.1) #6021)
docs(architecture): map the Responses separator module
fix(stripRawStage rescans all history text every turn — a fourth O(history) sync path beyond #4528/#5744 #6018): gate the per-slot MARKER_LINE regex behind a literal check and attribute raw-strip wall time
test: rerun ci — New test: after repeatedly killing (dsh, pi agent) and restarting, the compression state of long sessions must survive a | 新测试:反复杀(dsh、pi agent)并重启后,长会话的压缩状态必须存活、缓存命中必须能恢复——宿主 CI 与 bili-gate 两边补自动化断言 #5631 restart-survival dsh-lane startup flake (no code change)
fix(fold-reconcile): run the surviving-majority test on active coverage only ([Bug][DSH] Partial interior loss is classified as total drift despite surviving direct-ID majority, latching compression off #5715)
test(web): stop i18n ref scanner treating filenames like package.json as catalog keys
test(launcher): point the one-observation-loss comment at where the contract actually lives
fix(web): show built-in notice banners (missing .show class)
fix([Bug][preflight] Folding pass expansion asymmetry: view meets target but outbound reconstruction exceeds window, only ~5 | [Bug][preflight] 折叠 pass 展开不对称:视图达标但出站重建超窗,每轮只推进 ~5k token (#4543-①) #5974): judge the mid-walk futility deficit on the corrected scale too
fix: warn about iot-thinking only while it is enabled
feat: warn when dsh's iot-thinking plugin is installed
fix: abort the HTTP fallback egress when the Codex WS client disconnects
fix: keep UnsignedThinkingStrip module-internal (unused-exports lane, Violates Anthropic's constraint that "preserved thinking must be replayed verbatim." | 违反了 Anthropic「preserved thinking 必须原样重放」的约束 #5998)
docs: document the pre-fix Codex transport:sse fallback for the dsh lane ([Bug][DSH] Codex auto/WebSocket bypasses native fetch interception while compaction is disabled #5838)
docs: register src/server/thinking-posture.ts in the architecture file map (Violates Anthropic's constraint that "preserved thinking must be replayed verbatim." | 违反了 Anthropic「preserved thinking 必须原样重放」的约束 #5998)
docs: document third-party Codex transport boundary in docs/clients ([Bug][DSH] Codex auto/WebSocket bypasses native fetch interception while compaction is disabled #5838)
docs: correct the stale [Bug] DSH native fork adoption skips zstd-compressed Codex SSE request bodies at body-parse gate #4829 comment in the dsh takeover gate
test: pin that installDshIntercepts arms both egress legs ([Bug][DSH] Codex auto/WebSocket bypasses native fetch interception while compaction is disabled #5838)
fix: route DSH Codex Responses WebSocket through the native proxy ([Bug][DSH] Codex auto/WebSocket bypasses native fetch interception while compaction is disabled #5838)
test([Bug][preflight] Folding pass expansion asymmetry: view meets target but outbound reconstruction exceeds window, only ~5 | [Bug][preflight] 折叠 pass 展开不对称:视图达标但出站重建超窗,每轮只推进 ~5k token (#4543-①) #5974): pin case-3 failure kind, drop tautological assertion
fix(preflight): judge fit on the post-reconstruction measurement when view and rebuild diverge ([Bug][preflight] Folding pass expansion asymmetry: view meets target but outbound reconstruction exceeds window, only ~5 | [Bug][preflight] 折叠 pass 展开不对称:视图达标但出站重建超窗,每轮只推进 ~5k token (#4543-①) #5974)
fix: drop signature-less thinking blocks for upstreams that reject them (Violates Anthropic's constraint that "preserved thinking must be replayed verbatim." | 违反了 Anthropic「preserved thinking 必须原样重放」的约束 #5998)
test(launcher): cover the [Bug][dsh] 0.1.212: Zombie proxy holds port + fingerprint branch only skips without terminating → oversized sessions stu | [Bug][dsh] 0.1.212: 僵尸代理占端口 + fingerprint 分支只 skip 不 terminate -> 超大会话持续 422(BUSY 判定 × EADDRINUSE 死锁) #5955 wedged same-lane takeover ladder
fix: make the upstream circuit's half-open probe slot a lease
fix: arm the opencode V1 lane's direct-send window guard (omp/opencode lanes: oversized history egresses uncompressed when the proxy is down — direct-send window guard still unarmed (#5025 family residual) #5032)
docs: add Japanese README ([Docs] Japanese README: where should it live? | [文档] 日文 README 应该放在哪里? #6000)
fix: opencode 2.x native install writes the shim dir instead of the bare package name ([Bug] Plugin fails to load under OpenCode 2.1.0: bare package name entry dist/index.js lacks a default export (Plugin mu | [Bug] OpenCode 2.1.0 下插件加载失败:裸包名入口 dist/index.js 没有默认导出(Plugin must export a default definition) #3976)
docs: allow root README..md community translations ([Docs] Japanese README: where should it live? | [文档] 日文 README 应该放在哪里? #6000)
fix(native-intercept): name the floored local bar distinctly from the resolved window in the empty-slot refusal (bili-native (dsh): empty intercept slot sends a 6.7 MiB / 2.1M-token body direct — the 8 MiB byte cap does not catch it, provider answers 400 CONTEXT_WINDOW_EXCEEDED #5996 review)
fix: empty-slot direct send now refuses over-window bodies locally (bili-native (dsh): empty intercept slot sends a 6.7 MiB / 2.1M-token body direct — the 8 MiB byte cap does not catch it, provider answers 400 CONTEXT_WINDOW_EXCEEDED #5996)
docs(clients): link the zh-CN Command Code note to its section heading
docs(clients): add Command Code BYOK entry point to verified-entry-points table (Support Command Code (commandcode.ai CLI) — BYOK /bili/ works today; gateway wire + native seams for a fuller lane #5995)
fix(ci): register tests/issue5955-wedged-same-lane-takeover.test.ts in LIVE_FILES
test(ci): register tests/issue5955-wedged-same-lane-takeover.test.ts in LIVE_FILES
ci([PR] Fix #5807: Classify TTFB timeout separately on gateway hang and replay within the proxy | 修 #5807:网关装死时,首字节超时单独分类并在代理内重放 #5947): sync requireNudge seed wording with credit-window change; un-export internal gate types
fix(compress.requireNudge is only valid in the round where the nudge is injected: if the model comes back in a later round t | compress.requireNudge 只在 nudge 注入的那一轮有效:模型隔轮回来压缩就被拒,多轮清理式压缩全部白拒 #5943): requireNudge judges a nudge credit window, not this-round arm
fix([PR] fix(#5951): replay a stream cut that lands after reasoning-only output #5956): wire the responses pipe read-error emit to name its read error
test(ci): align issue411 waitFor with the CI: windows-latest node-22 shard 2 intermittently red on timing-sensitive e2e tests under load (rotating flakes: issue760, #4079, storage-full-flip #3818) #4161 timing-window idiom (PR [PR] loop: retry degenerate turns that end with whitespace-only text (#732 follow-up) #5986 red CI)
storage: jsonWriter=off also stops the CCR content-store sidecar (Identity pipeline: key sessions by (agent, conversation id) in the unified store — decouple fold ledgers from protocol/host origin (#2938 / #2847 / #2671 defect-1) #2960)
fix(stream): extend the return-to-task instruction to all parse-failure receipts ([PR] loop: retry degenerate turns that end with whitespace-only text (#732 follow-up) #5986 review)
test(loop): pin the double-degenerate warn both directions ([PR] loop: retry degenerate turns that end with whitespace-only text (#732 follow-up) #5986 review)
fix(loop): correct double-degenerate warn attribution to the request-scoped budget ([PR] loop: retry degenerate turns that end with whitespace-only text (#732 follow-up) #5986 review)
refactor(loop): drop redundant forwardedVisibleText from the degenerate-retry gate ([PR] loop: retry degenerate turns that end with whitespace-only text (#732 follow-up) #5986 review)
fix: context-analysis harvest denominator follows the injected tier (Context: "Stuck at N%, won't compress any further" — missing analysis tools: initialization size / injection frequency / | 上下文"卡在 N% 压不下去"缺少分析工具:初始化多大 / 注入多频 / 压缩接受率 / 诊断意见(2026-11-08 session-aedc188c 排查) #5963)
loop: retry degenerate turns that end with whitespace-only text ([BUG] In some cases, with Codex's auto model + DeepSeek, it will suddenly terminate automatically. #732 follow-up)
test: pin the bytes-present gate direction of the compress refuses with kind=missing-content keys=[] when the upstream stream was cut — receipt says do-not-reissue and the compression is abandoned #5981 lost-in-transit branch
fix: bound the lost-in-transit re-issue advice with the session breaker (compress refuses with kind=missing-content keys=[] when the upstream stream was cut — receipt says do-not-reissue and the compression is abandoned #5981)
fix: tell the model to re-issue a compress whose args were cut in transit (compress refuses with kind=missing-content keys=[] when the upstream stream was cut — receipt says do-not-reissue and the compression is abandoned #5981)
test(launcher): cover the [Bug][dsh] 0.1.212: Zombie proxy holds port + fingerprint branch only skips without terminating → oversized sessions stu | [Bug][dsh] 0.1.212: 僵尸代理占端口 + fingerprint 分支只 skip 不 terminate -> 超大会话持续 422(BUSY 判定 × EADDRINUSE 死锁) #5955 wedged same-lane takeover ladder
test(fork-adoption): add deep-copy isolation, straddled-count and ref-carry regressions
fix(launcher): field-tune wedged-takeover timings + quiet win32 ESRCH alarm ([Bug][dsh] 0.1.212: Zombie proxy holds port + fingerprint branch only skips without terminating → oversized sessions stu | [Bug][dsh] 0.1.212: 僵尸代理占端口 + fingerprint 分支只 skip 不 terminate -> 超大会话持续 422(BUSY 判定 × EADDRINUSE 死锁) #5955)
fix: exempt image-bearing bodies from the ingest cost band (image doctrine PR #489 后续: 默认图片计费 b64/4 在像素 tile 计费上游(官方 Anthropic/OpenAI)会硬 502 误杀多模态会话 #496/Image session permanent 503: re-sending 48.6 MiB outbound body every round (fold blocked by cadence, archived images sti | 图片会话永久 503:每轮重发 48.6 MiB 出站体(fold 被 cadence 挡住,已归档图片仍在 wire 上) #3909)
docs: correct mis-cited issue numbers in batch-PR comments and test file name
fix(launcher): stop exporting wedgedLedgerFilePath (unused-exports gate)
test: pin the ACP 状态标记被拼接进用户消息文本块;跨会话代际 stale refs 压缩整批失败 #1029-② one-liner against the real lecture signature, not a string no receipt carries
test(Anthropic stream:false: a model compress call ends the client turn (receipt returned with end_turn, no re-request) #5957): pin the MAX_LOOP_ROUNDS degradation gate for the non-stream Anthropic loop
ci: fix unused-export gate (MeasuredOutbound internal) + regenerate config docs ([docs][website] 配置文档单一事实源:从声明式 schema 生成 CONFIGURATION.md 与网站 /docs/(一步到位) #2107 drift guard)
fix(anthropic): never send an outbound body that ends on an assistant turn ([Bug] failed-compress recovery receipt (#5220/#5231) leaves the re-request ending on an assistant turn → upstream 400 (trailing whitespace / prefill not supported), subagent dies #5604)
test: drop duplicate overflowArmCompletionReserve assertion (review cleanup)
fix: 会话历史按「真实维度」拦死 — 条数闸(BILI_MAX_MESSAGES) + 32MiB 摄入成本带(BILI_INGEST_BODY_BYTES) + 判死信息点名超重字段 ([Bug][dsh][0.1.208] Kill switch / egress fuse only measures message.text: a session with 26,903 messages and only 1.44 M | [Bug][dsh][0.1.208] 判死/egress fuse 只量 message.text:26,903 条消息、core text 仅 1.44 MiB 的会话绕过 fuse 发出 55.2 MiB 请求体,被上游 413 拒收(ModelArts.… #5000 [perf][dsh] 51.2 MiB / 29.6k-msg inbound has no cost admission - DecodedRequestAdmission only sees compressed bodies | 51.2 MiB 入站没有成本准入:admission 只覆盖压缩体 #5794 [Bug][dsh] Rewind (fork+seed) leaves non-monotonic refs: preflight folds only ~5k tokens/round, so 31 rounds (649 s) cannot close a 283k deficit — and the "NOT forwarded" payload is sent anyway #5885)
fix(ci): sync root + kernel lockfiles to acp-kernel 0.0.189
fix(launcher): force-takeover the wedged same-lane proxy holding the canonical port ([Bug][dsh] 0.1.212: Zombie proxy holds port + fingerprint branch only skips without terminating → oversized sessions stu | [Bug][dsh] 0.1.212: 僵尸代理占端口 + fingerprint 分支只 skip 不 terminate -> 超大会话持续 422(BUSY 判定 × EADDRINUSE 死锁) #5955)
fix: context-analysis idle gates — drop dead reason-string patterns, surface other bucket (Context: "Stuck at N%, won't compress any further" — missing analysis tools: initialization size / injection frequency / | 上下文"卡在 N% 压不下去"缺少分析工具:初始化多大 / 注入多频 / 压缩接受率 / 诊断意见(2026-11-08 session-aedc188c 排查) #5963)
chore: pass unused-exports + architecture-docs gates (Context: "Stuck at N%, won't compress any further" — missing analysis tools: initialization size / injection frequency / | 上下文"卡在 N% 压不下去"缺少分析工具:初始化多大 / 注入多频 / 压缩接受率 / 诊断意见(2026-11-08 session-aedc188c 排查) #5963)
test: allowlist context-analysis array slices in the compactEntry summary clamp (slice(0,199)+'…') can strand a lone high surrogate → same deterministic UTF-8 gateway 500 family #816/[bug] summaryFingerprintLine 的 head/tail 裸码元切片产生孤立代理项 → 上游 JSON 400、会话永久卡死(#816/#828 同族第三站点;0.1.169 仍未修) #1615 family gate
feat: context analysis — init size, nudge cadence, harvest acceptance, verdict (Context: "Stuck at N%, won't compress any further" — missing analysis tools: initialization size / injection frequency / | 上下文"卡在 N% 压不下去"缺少分析工具:初始化多大 / 注入多频 / 压缩接受率 / 诊断意见(2026-11-08 session-aedc188c 排查) #5963)
fix(google): keep tool_result-first part order for Claude-through-Google targets (for [PR] fix: keep tool_result-first google part order for Claude targets (#5392) #5397, closes [Bug] Host-side tool-result pruner mutates folded messages → permanent fold residual (614 ids) re-reported every turn, v0.1.210 plugin lane #5392)
fix(codex): keep marketplace Git repos inside the codex overlay tmp containment (for [PR] fix(codex): keep marketplace Git repos inside CODEX_HOME (overlay owns .tmp) #5586, closes [Bug][Codex Launcher] Marketplace Git repositories escape CODEX_HOME through Bili .tmp symlinks #5584)
fix(codex): answer WS-lane failures with object error frames incl. oversize passthrough hint (for [PR] fix: answer WS-lane failures with object error frames and true 1009 sizes (#4601) #4624, closes Codex WS lane: proxy-failure error frames carry
erroras a string, so codex ignores them and waits for its idle timeout; 1009 diagnostic reports 0.0 MiB #4601)fix: name the no-guard fetch postures in bili.log; document the direct /bili/ fallback (The default
modelContextLimitof 200000 underestimates the context window for custom provider routing, and over-window |modelContextLimit缺省 200000 会把自定义 provider 路由的窗口算小,且超窗的「无归因」请求会被照发而不是本地拒绝 #5907)fix: slow-session warn fires once per session with size picture and cleanup hint (The default
modelContextLimitof 200000 underestimates the context window for custom provider routing, and over-window |modelContextLimit缺省 200000 会把自定义 provider 路由的窗口算小,且超窗的「无归因」请求会被照发而不是本地拒绝 #5907 minor item b)fix: name the expected duplicate-instance outcome in the terminal EADDRINUSE diag (The default
modelContextLimitof 200000 underestimates the context window for custom provider routing, and over-window |modelContextLimit缺省 200000 会把自定义 provider 路由的窗口算小,且超窗的「无归因」请求会被照发而不是本地拒绝 #5907 minor item a)fix: fall back to the 1M window floor in dsh gate oversize arms when no window is declared (The default
modelContextLimitof 200000 underestimates the context window for custom provider routing, and over-window |modelContextLimit缺省 200000 会把自定义 provider 路由的窗口算小,且超窗的「无归因」请求会被照发而不是本地拒绝 #5907)fix: recover retained compression blocks against fork replay
tests: pin the new receipt ladder + credit-overrides-staleness doctrine; gate the one-liner collapse on the tracked lane
fix: parse LiteLLM input and completion overflow counts
fix: match custom-gateway 'vendor/prefix:model' ids in the context-limit table (The default
modelContextLimitof 200000 underestimates the context window for custom provider routing, and over-window |modelContextLimit缺省 200000 会把自定义 provider 路由的窗口算小,且超窗的「无归因」请求会被照发而不是本地拒绝 #5907)fix: execute proxy-owned tools on the non-stream Anthropic path (Anthropic stream:false: a model compress call ends the client turn (receipt returned with end_turn, no re-request) #5957)
fix: shape the double-degenerate retry terminal by streamErrorShape ([dsh] Retry-exhausted degenerate turn is serialized as successful completion (0.1.213) #5958)
Stop the compress-burst loop: ack fold-point pointer echoes, collapse repeat failure receipts, credit overrides cycle-stale meter
fix(Replay a stream cut that lands after reasoning-only output #5951): replay a stream cut that lands after reasoning-only output
docs: correct the split-condition wording in fetchWithTimeout ([bug][fetch] Upstream TTFB idle inherits the 12-min default → hangs with a pointless 35 s wait; suggest splitting header | [bug][fetch] 上游首字节空闲沿用 12 分钟默认值 → 卡死白等 35 s;建议拆分 headers/body 预算并让 header 超时可重放 #5807)
chore: sync kernel/package-lock.json version fields to 0.0.189 (lock-sync)
chore: sync package-lock acp-kernel to 0.0.189 (lock-sync)
kernel: relocated fold summaries ride the fix: detect codex compaction requests before preflight; verbatim passthrough when not intercepted (#332) #335 live-carrier protection
kernel: relocate-compress v2 — summary text moves to the fold point, the pair stays in its slot with pointer args ([Feature][experimental] After compression, relocate the compress call pair to its own fold point (weak-attention models | [Feature][experimental] After compression, relocate the compress call pair to its own fold point (weak-attention models | [Feature… #5470 follow-up)
test: pin first-byte timeout classification and in-proxy replay ([bug][fetch] Upstream TTFB idle inherits the 12-min default → hangs with a pointless 35 s wait; suggest splitting header | [bug][fetch] 上游首字节空闲沿用 12 分钟默认值 → 卡死白等 35 s;建议拆分 headers/body 预算并让 header 超时可重放 #5807)
fix: give a tightened first-byte timeout its own replayable kind ([bug][fetch] Upstream TTFB idle inherits the 12-min default → hangs with a pointless 35 s wait; suggest splitting header | [bug][fetch] 上游首字节空闲沿用 12 分钟默认值 → 卡死白等 35 s;建议拆分 headers/body 预算并让 header 超时可重放 #5807)
docs: record integrity worker stderr capture fix in CHANGELOG [Unreleased] ([Bug][Pi native] SQLite ExperimentalWarning from background integrity worker leaks into Pi TUI #5940)
fix(ci): tripwire error points at scripts/live-files.mjs where LIVE_FILES actually lives ([Bug][Pi native] SQLite ExperimentalWarning from background integrity worker leaks into Pi TUI #5940)
fix(ci): register tests/integrity-worker-stderr.test.ts in LIVE_FILES ([Bug][Pi native] SQLite ExperimentalWarning from background integrity worker leaks into Pi TUI #5940)
fix: capture background integrity worker stderr so the SQLite experimental notice no longer leaks into the host TUI ([Bug][Pi native] SQLite ExperimentalWarning from background integrity worker leaks into Pi TUI #5940)
fix(log-mask): scrub JSON-quoted credentials and Groq/Hugging Face keys
fix: 计量器补排 responses 车道 image_url 键 + 判死只比文本量(CI 回归 [Bug] Jimp is active, but image accumulation still triggers Responses WS 1009; duplicate request size not reduced (v0.1. | [Bug] Jimp 已生效,图片累积仍触发 Responses WS 1009;重复请求未缩小(v0.1.200) #3701/[Question] 非白名单上游(如 DeepSeek)图片按 b64/4 估算,实测高估约 100×;且该估算被写回 lastInputTokens 并自愈成 confirmed 窗口,使 #496 逃生门永久失效 #857)
fix: 超窗请求不再裸发上游 — 三道转发闸按「真实出站体」计量 + 32MiB 成本带准入 ([Bug][dsh][0.1.208] Kill switch / egress fuse only measures message.text: a session with 26,903 messages and only 1.44 M | [Bug][dsh][0.1.208] 判死/egress fuse 只量 message.text:26,903 条消息、core text 仅 1.44 MiB 的会话绕过 fuse 发出 55.2 MiB 请求体,被上游 413 拒收(ModelArts.… #5000 [Bug] CJK payloads slip past the egress byte-floor fuse: a 4.1 MiB outbound was forwarded at 1,121,674 tok into a 1,000,000-token window (0.1.206, plugin lane) #4543 [Bug][dsh] Rewind (fork+seed) leaves non-monotonic refs: preflight folds only ~5k tokens/round, so 31 rounds (649 s) cannot close a 283k deficit — and the "NOT forwarded" payload is sent anyway #5885 [perf][dsh] 51.2 MiB / 29.6k-msg inbound has no cost admission - DecodedRequestAdmission only sees compressed bodies | 51.2 MiB 入站没有成本准入:admission 只覆盖压缩体 #5794)
docs: add docs/README navigation landing page, en+zh (owner request 2026-11-08)
docs: refresh two stale facts in slimmed AGENTS.md (mysql2/jimp lazy deps; src/server/ + src/storage/ in module map, architecture-docs CI lane)
fix([PR] Upstream hang on compressed summary no longer silent: added first-byte budget + watchdog/preflight logging (#5904) | 压缩摘要上游卡死不再静默:新增首字节预算+看门狗/预检日志(#5904) #5917 review): unset first-byte knob follows the RESOLVED upstreamTimeoutMs
docs: slim AGENTS.md — KDD full text to reference/design-decisions.md, triage table to reference/triage.md (owner request 2026-11-08)
fix: prepare DSH WS frames and guard cached identity ([Bug][DSH] Codex auto/WebSocket bypasses native fetch interception while compaction is disabled #5838)
docs: document third-party Codex transport boundary in CLIENTS ([Bug][DSH] Codex auto/WebSocket bypasses native fetch interception while compaction is disabled #5838)
docs: correct the stale [Bug] DSH native fork adoption skips zstd-compressed Codex SSE request bodies at body-parse gate #4829 comment in the dsh takeover gate
test: pin that installDshIntercepts arms both egress legs ([Bug][DSH] Codex auto/WebSocket bypasses native fetch interception while compaction is disabled #5838)
fix: route DSH Codex Responses WebSocket through the native proxy ([Bug][DSH] Codex auto/WebSocket bypasses native fetch interception while compaction is disabled #5838)
chore: sync root package-lock.json with kernel 0.0.188 bump (New error in v0.213, following up on the previous bug where old blocks can't be compressed and tokens double. | 0.213版本新报错,接着上次的旧块无法压缩token翻倍bug #5919)
kernel: version 0.0.188 (New error in v0.213, following up on the previous bug where old blocks can't be compressed and tokens double. | 0.213版本新报错,接着上次的旧块无法压缩token翻倍bug #5919 shredded-merge recovery)
kernel: close shredded-merge dead-ends — auto-extend to stranded partners or no-op with exact guidance (New error in v0.213, following up on the previous bug where old blocks can't be compressed and tokens double. | 0.213版本新报错,接着上次的旧块无法压缩token翻倍bug #5919)
docs: all 8 architecture module pages live (owner request 2026-11-08)
fix(docs): hub ledger links — module pages live in docs/architecture/, hub is in docs/ (owner report: storage.md 404)
fix(docs): hub ledger links 404 — module pages live in docs/architecture/, hub is in docs/ (owner report: docs/storage.md 不存在)
fix(docs): hub diagram links 404 — hub lives in docs/, assets in docs/architecture/assets/ (owner report: 整体架构不显示); guard now checks hub diagram links
fix(docs): drop duplicated 'Architecture docs are LIVE' bullet in AGENTS.md
docs: live architecture hub — overall diagram, module pages with diagrams, file-map CI guard (owner request 2026-11-08)
docs: codify documentation layout rules in AGENTS.md §3 (owner request 2026-11-08)
fix(acp-panel): keep legacy CONFIGURATION.md pointer prefix in status-strip grammar ([PR] docs: batch-migrate 14 root-level documents into docs/ (consolidation + master index + redirect placeholders) | docs: 14 个根级文档一次性迁入 docs/(集中 + 总目录 + 重定向占位) #5920)
docs: centralize 14 root docs under docs/ with hub index + redirect stubs (owner request 2026-11-08)
fix(gate): opencode lane one-time warmup round for fresh storage dirs (PR [PR] gate: add a one-time warmup round to the opencode lane to absorb the v1 CLI sqlite storage migration (guaranteed to trig | gate: opencode 车道加一次性 warmup 回合,吸收 v1 CLI 的 sqlite 存储迁移(容器每跑必中) #3497)
fix(opencode): route host console chatter to bili.log and teach the config key (PR [PR] fix(#5706): opencode plugin startup lines go to bili.log; "opencode" config key no longer warned as unknown #5710, closes opencode native plugin: un-silenceable startup console.log noise + "opencode" config key falsely warned as unknown #5706)
fix: native liveness log says re-attach when the probe can self-heal (PR [PR] fix: native liveness log says re-attach when respawn returns the origin we just let go (#5823/#5449) #5828)
fix(dsh): await ACP tool registration before releasing activation (PR [PR] fix(dsh): await ACP tool registration before activation releases — first round composes with the full tool set #4136, fixes In each new session, the first turn carries no ACP tools: the second turn triggers a full prefix recomputation, and the | 每个新会话第 1 拍没带 ACP 工具:第二拍整前缀重算,前缀缓存每次冷启动断一次 #4134)
docs(kernel-boundary): note mechanical-bug exemption in contributing checklist (PR [PR] docs(kernel-boundary): add mechanical-bug exemption to Kernel Boundary rule 1 #5822, complements merged [PR] Kernel Minor Bug Exemption Clause: Fix mechanical bugs directly + report in-thread; merges still require human review. | 内核小 bug 豁免条款:机械性 bug 直接修+线程内上报,合并仍人审 #5819)
docs(website): add missing bili sessions / storage gc command rows (PR [PR] docs(website): add the missing bili sessions / bili storage gc rows to the Commands section #4565, closes /docs/ command section is missing the bili sessions / bili storage gc shipped on Oct 20. | /docs/ 命令区漏了 10-20 上线的 bili sessions / bili storage gc #4562)
docs(clients): disclose pi-claude-bridge compression-state reset on reopen (Pi + claude-bridge: native sends no model requests to proxy; launcher compress returns 404 #3429, PR [PR] docs: document the pi-claude-bridge compression-state limitation (#3429) #3553)
docs: document opencodex (ocx) as upstream gateway (PR [PR] docs: opencodex (ocx) as upstream gateway — verified recipe #5829)
docs: document omp advisor session isolation ([Bug][omp] Identity cross-talk between advisor and main session, causing DEAD refs, fold drift, and repeated compaction | [Bug][omp] advisor 与主会话身份串扰,导致 DEAD refs、fold drift 和反复压缩(0.1.212) #5567, PR [PR] docs: document omp advisor session-isolation workaround (#5567) #5568)
docs(claude): disclose that base-URL routing disables account-bound features (Installing billion-context removes '/remote-control' command from claude-code #4693, PR [PR] docs(claude): disclose that bili routing disables account-bound features like Remote Control (#4693) #4709)
docs(The summarization request silently stalls for 5 minutes inside the agent before being dispatched to the model, bringing | 压缩摘要请求在代理内部静默卡 5 分钟才发给模型,把整轮对话拖死(gate Lane A 三连红根因) #5904): align first-byte-knob comments and docs with shipped values
docs(The summarization request silently stalls for 5 minutes inside the agent before being dispatched to the model, bringing | 压缩摘要请求在代理内部静默卡 5 分钟才发给模型,把整轮对话拖死(gate Lane A 三连红根因) #5904): regenerate config artifacts for upstreamFirstByteMs + ledger entry
fix(The summarization request silently stalls for 5 minutes inside the agent before being dispatched to the model, bringing | 压缩摘要请求在代理内部静默卡 5 分钟才发给模型,把整轮对话拖死(gate Lane A 三连红根因) #5904): phase-1 first-byte budget for upstream summary calls + watchdog/preflight logging
release-gate: prove real inference in preflight, refuse evidence overwrite, print evidence map on red
docs: replace How-it-works ASCII diagram with SVG, en/zh variants (owner request 2026-11-08)
docs: state last-resort fallback semantics for top-level modelContextLimit / ACP_MODEL_CONTEXT_LIMIT (Prompt-cache hit rate 0.88–0.92 with a pinned window in long multi-turn pi sessions (pure proxy): what triggers the mostly-uncached requests? #5908)
test: pin the one-shot saturation warn in the 429 governor hold is keyed by origin+credential with no model dimension — one model's quota exhaustion freezes the whole lane (healthy models get local 429/424) #5199 saturated-map case
fix(docs): sync config-reference seed with batch-total minSummaryLength semantics (Batch compression failed: insufficient output for individual paragraphs. | 批量压缩个别自然段输出过少失败 #5894)
fix: docs-sync/lock-sync — throttle429Scope notes belong in the process.yaml seed (CONFIGURATION.* are generated), kernel lockfile 0.0.187
fix: [PR] release v0.1.213 #5898 review batch — 429-hold saturation warn + scope-value guard, media-refs carry-forward filter & cloud prefix families, oversize refusal messages that name the real fix, dsh durable lines with readable numbers, website streamKeepAliveMs typo, 5755 ord-crossing guard kept with regression tests
feat: panel quick-config row for compress.minSummaryLength + batch-total docs (Batch compression failed: insufficient output for individual paragraphs. | 批量压缩个别自然段输出过少失败 #5894)
kernel: judge the summary floor at batch level for multi-range compressions (Batch compression failed: insufficient output for individual paragraphs. | 批量压缩个别自然段输出过少失败 #5894)
fix(ci): close unterminated quote + missing event-name injection in kernel-guard Rule-1 ([PR] fix: add workflow_dispatch trigger to boundary guard (release path guard blind spot, with branch-exempt dispatch hardeni | fix: 边界守卫补上 workflow_dispatch 触发(发版路径守卫盲区,含分支豁免 dispatch 加固) #4428 piece)
ci: run boundary guards on workflow_dispatch too (release-lane coverage gap)
fix(test): pin BILI_MESSAGE_TIMESTAMPS in the storage goldens' own test files (storage-render golden suite is not hermetic: locally regenerated digests cannot be reproduced on CI runners #5402)
docs(tests): correct stale storage-default comment in preload-storage-baseline ([PR] fix: make full-mode storage tests deterministic (#4004) #4011 review)
fix: recover x.ai Responses WS continuations rejected without a type
test(golden): pin additive requireUsageSampleForPressure in wire-contract kernel.json
test(golden): realign storage-render snapshots with post-[PR] fix(nudge): no billing sample this round → hold pressure-band injection one round (phantom-pressure guard) #5860 persisted-state shape
fix(plugin): don't drain held tags on inert mid-stream frames ([pi/OpenAI] text-less mid-stream frames (delta:{}) drain the tag-echo filter → opener-less render-tag residue (one mechanism of #5192) #5848)
fix: hide the host composer seat while the Context tab is open ([Feature][dsh] The "Compression Agent" tab should not render the host composer (input field / task list / goal box) | [Feature][dsh] 「压缩代理」标签页不应渲染宿主 composer(输入框 / 任务列表 / goal 框) #4406)
test: align the Unified Storage: MERGE_STATE_FIELDS and mergeState out of sync — full mode hydration drops sweptOrphanCallIds/pendingOrp | 统一存储 MERGE_STATE_FIELDS 与 mergeState 不同步:full 模式水合丢 sweptOrphanCallIds/pendingOrphanSweepMinRef(#2388 类) #4192 field-list mechanism pin with the probe-derived list ([PR] fix(resume): inherit timestamp ledger so resumed sessions re-render provider-cached tag bytes #5609)
fix: derive the kstate field list from a fully-populated probe ([PR] fix(resume): inherit timestamp ledger so resumed sessions re-render provider-cached tag bytes #5609 CI red)
test(side request(max_tokens<=200)绕过 preflight 且不检查载荷体积,携带全量历史直转上游导致 400 #554): pin the honest-NULL poison field in the side-request isolation suite
# Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608-B: honest-NULL arms only where identity is PROVEN (fix side request(max_tokens<=200)绕过 preflight 且不检查载荷体积,携带全量历史直转上游导致 400 #554 state poisoning)
kernel: keep # Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608-B/C new state fields key-free on legacy states
chore: regenerate package-lock.json version fields after master sync (lock-sync gate)
fix(ca): let hasKeyCertSign see critical-marked keyUsage ([Bug][codex][macOS]
bili codexfails on every connection: combined-ca.pem includes non-CA keychain certs, andcom.apple.kerberos.kdcmakes Codex reject the whole bundle #5802)# Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608-D: defer citation-space alignment from cold load to the next post-fold save
kernel: # Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608-B/C — time-attr high-water mark + honest-NULL minting arm
fix: keep only CA certificates in combined-ca.pem ([Bug][codex][macOS]
bili codexfails on every connection: combined-ca.pem includes non-CA keychain certs, andcom.apple.kerberos.kdcmakes Codex reject the whole bundle #5802)docs: re-apply local-forge accounting wording lost in the master-merge ([PR] fix(diagnostics): report large histories without attributing stalls #5481)
test: pin D-lane retired-block anchor suppression (Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 review)
fix: export ledger range fallback avoids arg-spread crash on huge blocks (Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 review)
fix: block-op receipts report only what actually changed (Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 review)
chore(web): color the new block-op tool chips like acp_cache
fix(acp-block-ops): normalize retire receipt block ids via parseBlockIdArg
chore(tests): name Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 regression tests after the GitHub issue number
chore(kernel): align CompressionBlock field indentation to 2-space convention
test(wire-contract): regenerate goldens for the Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 tool-surface additions
fix(test): classify Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765 block-op tools in the kernel partition guard
feat: acp_retire/acp_restore/acp_rewrite host surface + export ledger (Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765)
kernel: block retirement and external summary rewrite (Request two capabilities: compressed block "retirement" and summary "external rewrite". | 请求两个能力:压缩块「退役」与摘要「外部重写」 #5765)
fix(ci): opt plugin-agent tests into the full pi.commands surface ([PR] feat(pi): pi.commands — configurable slash-command registration surface #5659)
TESTS ONLY: await hydrate-on-listening before the phase-2 first request (# Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608 Windows race)
ci: nudge the pull_request event (head sync lag) for the # Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608 Windows triage commit
TESTS ONLY: make the miss-seed/gate e2e failures self-diagnose on Windows CI (# Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608)
fix(ci): drop unused PiCommandName export; regenerate config docs from seed ([PR] feat(pi): pi.commands — configurable slash-command registration surface #5659)
fix(fold-reconcile): keep the [Bug][DSH] Partial interior loss is classified as total drift despite surviving direct-ID majority, latching compression off #5715 majority-loss note off already-explained episodes
fix(fold-reconcile): a surviving direct-ID majority is partial loss, not total drift ([Bug][DSH] Partial interior loss is classified as total drift despite surviving direct-ID majority, latching compression off #5715)
TESTS ONLY: reshape the miss-seed e2e to the real continuation shape and pin the v3 adoption gate (# Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608)
resume-miss: require a verified continuation before weakly adopting (# Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608)
fix(test): number the drift diag suite by its GitHub issue + pin the async-lane latch gate
fix(storage): name divergent drift fields and force the full path while drift-latched
fix(invariants): carry proven separator ownership into the next round's continuity verdict
refactor(review): pi.commands — gate indent, import order, per-class warn flags
feat(pi): pi.commands — configurable slash-command registration surface
fix(Service crashed, sessions exploded instantly. | 服务崩溃,会话瞬间爆炸 #5219): bound the bounded-boot parse window by decoded-file count
fix: align stray indent in maybeJudgeContinuity
test(resume): defeat the suite's timestamp pin in the # Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608 wiring tests
test(golden): adopt ci-fixes converged storage-render goldens (In 0.1.210, sessions with frequent interruptions barely trigger compression reminders — tokens accumulate to 330k before | 中断频繁的会话在 0.1.210 里压缩提醒几乎不触发,积到 33 万 token 才压一次 #5415 nudge state, Mismatch between tool copy and actual behavior (false-advertising cleanup): on compress failure, falls back to the retir | 工具文案与实际行为不符(false-advertising 清理):compress 失败回执教回已退役形态 #5555 copy sweep)
fix(resume): weak adoption on resume-inheritance miss restores provider-cached prefix (# Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608)
fix(resume): carry timestamp ledger across public plugin forks too (# Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608)
fix(resume): carry timestamp ledger across resume/fork inheritance (# Cache hit collapses to 1–12% for a whole generation, then the very next request returns to 100% — repeated 15× in one day #5608)
test: cover round-two thinking hold edge shapes
docs: correct CHANGELOG local-forge accounting wording ([PR] fix(diagnostics): report large histories without attributing stalls #5481)
test: pin large-history threshold boundary and unknown-count case ([PR] fix(diagnostics): report large histories without attributing stalls #5481)
fix(logging): settle the retention drain gate on stream close, not only end()
fix: reload retained logs when view controls change
refactor: extract canonical Responses separator primitives
fix: retain unchanged separator-shaped client messages
test: clarify retention drain and destination ownership
fix(logging): drain replaced streams before retention
test(ci): register log-retention spawner in LIVE_FILES (tests/claude-native.test.ts linux live-E2E subtests fail/hang in shared multi-agent sandboxes #1248 tripwire)
fix(invariants): preserve separator lineage through final-wire rewrites
fix(diagnostics): measure the final forwarded request body
fix(logging): account direct appends and current active bytes
docs(config): register private log retention controls
docs(config): document private log retention knobs
docs: correct changelog wording — the import moved, 10 call sites untouched ([PR] fix(diagnostics): report large histories without attributing stalls #5481)
test: pin large-history warning wording against stall-attribution regression ([PR] fix(diagnostics): report large histories without attributing stalls #5481)
fix(diagnostics): retain bounded private log history
fix(invariants): bind separator exemptions to producer-owned wire items
fix(diagnostics): report large histories without attributing stalls
chore: bump acp-kernel 0.0.170 -> 0.0.171 after master sync (kernel-change-vs-base rule)
chore: bump acp-kernel 0.0.169 -> 0.0.170 after master sync (kernel-change-vs-base rule)
chore: sync acp-kernel 0.0.169 into kernel lockfile packages[""] entry (lock-sync)
chore: bump acp-kernel 0.0.168 -> 0.0.169 after master sync (kernel-change-vs-base rule)
test([Bug][omp lane] lane port moves under live omp hosts, and the host's cached proxy hop survives re-targeting — every Anthropic turn dies (connection refused / SELF_SIGNED_CERT_IN_CHAIN) #4566): pin the two missing gate directions — non-loopback keep + divert-site strip (Test Blind Spot Retrospective: Reinforcing Negative Assertion Discipline, Tightening Fake Upstream Strictness, Thin Inst | 测试盲区复盘:补反向断言纪律、假上游严格化、薄安装冒烟、形态矩阵与生命周期覆盖 #3617)
fix: drop stale host proxy hops from intercepted dispatches ([Bug][omp lane] lane port moves under live omp hosts, and the host's cached proxy hop survives re-targeting — every Anthropic turn dies (connection refused / SELF_SIGNED_CERT_IN_CHAIN) #4566 part 1)
chore: sync acp-kernel 0.0.164 into root and kernel lockfiles
kernel: heal system positions on the final Anthropic wire array ([Bug][anthropic] 0.1.206: system 400 still fires on nudge turns: the heal keeps a trailing system, then prepare-anthropic appends user notes after it #4570)
fix(loop): drop the export from ParseStreamOpts — the unused-export gate flags it INTERNAL_ONLY
fix(loop): hold a bili-owned round's thinking until its signature, so a cut inside round >=2 stays re-fetchable ([bug] anthropic: a cut inside bili's own re-request round (round ≥2) is unrecoverable — round-2 thinking streams unsigned, then #2501 correctly bars the only re-fetch #4679)
fix([PR] feat: declarable custom-wire model endpoints + commandcode-cli wire family (#1295) #1444 merge): seed modelEndpointPatterns in config-reference + unexport internal ModelEndpointWire
fix([PR] feat: declarable custom-wire model endpoints + commandcode-cli wire family (#1295) #1444 merge): restore master's pickAdapter/emitStreamError trailing args (anthropicCacheMarks, streamErrorShape) around the commandcode ternary
fix: once-gate the declaration/URL-pin conflict warn ([Feature] Declarable custom-wire model endpoints — route providers like commandcode's Go plan (/alpha/generate) into compression #1295 review F4)
fix: review hardening for declared custom-wire endpoints ([Feature] Declarable custom-wire model endpoints — route providers like commandcode's Go plan (/alpha/generate) into compression #1295)
docs: fix WC-4 unwrap/rewrap direction swap in wire-contracts ledger
feat: commandcode-cli wire family — JSONL envelope codec for declared custom endpoints ([Feature] Declarable custom-wire model endpoints — route providers like commandcode's Go plan (/alpha/generate) into compression #1295)
feat: declarable model endpoints — one config source for both claim gates ([Feature] Declarable custom-wire model endpoints — route providers like commandcode's Go plan (/alpha/generate) into compression #1295)