[ranxianglei]

Closes #4872

What

The #3354 load-time overlay (bili-compaction-gate) marks BOTH @deepseek-ai/dsh-compaction-basic AND @deepseek-ai/dsh-command-compact disabled: true in dsh web-profile presets. compaction-basic is the right target (its auto-compression checkpoints shadowed the fold substrate, #1729), but command-compact carries the EXPLICIT /compact command — the DSH-official bridge third-party plugins borrow via ctx.commands.find(agent, "compact") (e.g. dsh-switchman's /ctx-handover). Disabling it made those plugins degrade to undefined with nothing pointing at bili.

Owner decision (#4872 floor): ship an opt-in switch this release; full unbinding of the auto-compaction constraints waits for server-side compaction to fully take over next phase (see Follow-up below).

config surface

New keys (both default OFF = today's behavior):

key	kind	default	effect when on
dsh.keepCommandCompact	file (billion-context.json)	false	overlay disables only compaction-basic; command-compact rows pass through untouched
BILI_DSH_KEEP_COMMAND_COMPACT	env-only	false	same; env wins over file

Why the existing surface could not express this: the overlay scope was a hardcoded two-package constant list (src/agent/dsh-compaction-disable.ts, rule at :159-161) with no knob to narrow it. Alternatives considered: reusing dsh.authoredCompaction — rejected, that knob controls whether bili ANSWERS dsh compaction calls locally (#3733), not which rows get disabled; conflating them would let local-answering users still break /compact-dependent plugins. Per-package booleans — rejected as overkill; the real design axis is one flag: "auto vs explicit".

Owner sign-off pre-implementation (#4872 floor): 「ework 可以给出配置开关 允许本issue发生,本期这样」.

Behavior change disclosure

• Default (knob off/unset): zero drift. The overlay disables both packages exactly as before; every existing test pinning that behavior passes unchanged.
• Knob on: command-compact rows are no longer disabled → /compact is reachable again in web-profile presets and ctx.commands.find consumers work. A /compact-triggered compaction call STILL passes the nudge 死锁无检测:同一 session 交替收到两种结构互斥的请求流(巨型单消息 ↔ 多轮)时不告警 #1729 wire guard: with active fold blocks it is answered locally by bili-authored client compaction: let the host's own compaction flow run, but bili writes the checkpoint #3733 authored compaction (zero egress); otherwise the terminal, diagnosable refusal fires. Keeping the explicit command does NOT re-introduce nudge 死锁无检测:同一 session 交替收到两种结构互斥的请求流(巨型单消息 ↔ 多轮)时不告警 #1729's silent checkpoint shadowing — that mechanism lives in the auto path (compaction-basic), which stays disabled.
• Timing: the flag is read once at bili-native activation (host process) and takes effect at the next dsh/host restart — the plugin-mode convention for host-side settings.
• Telemetry count follows the same flag (countCompactionRows takes it), so covered-row counts can never drift from what the rewrite actually did.
• Verdict channel shape changed from string "active" to {active: true, keepCommandCompact?}; the reader tolerates the legacy string so mixed-version mid-update windows degrade to today's behavior.

Follow-up direction (recorded, NOT implemented here)

Per the owner decision, the next phase (等服务端压缩彻底转正) removes the auto-compaction constraints entirely, on the principle that the compression plugin depends only on incremental messages and can ignore stock history. That is separate work; this PR ships only the opt-in switch.

Tests

• New tests/issue4872-dsh-keep-command-compact.test.ts: pure-function scope (default pins both-disabled; keep=true passes compact rows through BY IDENTITY at top/group/deep nesting depths, input untouched), verdict channel (flag rides the declaration; legacy "active" string tolerated), transform e2e (telemetry count locked to the rewrite; compact-only tree returns the input object untouched), knob tiers (env > file > default via BILI_CONFIG_FILE).
• Local targeted runs green: issue4872 + issue3354 + issue3662 + issue4147 + config-surface-guard batch (30/30), dsh-native (68/68, extended with declaration-flag assertions incl. kill-switch still silent), dsh-settings-ui (25/25). npm run typecheck clean. Full suite = CI.

[github-actions]

📦 Built Package Artifact

Branch: 2026-10-27_dsh-keep-command-compact (d61d472)

Latest build — install this exact version

npm install -g billion-context@0.0.9298-pr.4883.a70a0c9

This exact version is immutable and reproducible. The rolling pr tag tracks the newest test build across all PRs.

Recent builds (commit → version)

commit	version	build
d61d472	billion-context@0.0.9298-pr.4883.a70a0c9	build
2470631	billion-context@0.0.7349-pr.4883.6203999	build
d15ab28	billion-context@0.0.7057-pr.4883.d166a00	build
76c45cd	billion-context@0.0.6842-pr.4883.cd68f90	build
6f0ba1a	billion-context@0.0.6346-pr.4883.bcd5e9c	build
995d1ed	billion-context@0.0.6342-pr.4883.c43888d	build
337a658	billion-context@0.0.6332-pr.4883.0bcd39e	build
25eaa12	billion-context@0.0.6330-pr.4883.e633b46	build
77c60b0	billion-context@0.0.6325-pr.4883.f6dd118	build

Download artifact

1. Download the artifact from the Actions run
2. Extract the tarball and install:

tar xzf billion-context-pr4883.tgz
npm install -g package

---

Install instructions are refreshed on each push.

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)
Merged origin/master into the branch (clean, no conflicts; 77c60b0 → 25eaa12), local typecheck passes — PR is now in sync with master, mergeable_state blocked only while CI runs on the new head.

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

查重: 不重复。 这是 #4872 的唯一实现 PR(Closes #4872;检索引用 4872 的 PR 只有本条),dsh-compaction-disable / keepCommandCompact 主题无并行工作项。本条是既有 agent 工作的 [PR] 镜像,按「识别自己的 PR」规则处理:核对分支与 CI、做完整 review,不重做。

标签: dsh、scope/wiring、p3 —— 功能交付(opt-in 开关,默认 OFF 零漂移),非缺陷不打 bug;底层插件失效问题在 #4872 侧已按 p2 管理。

状态(初核完成):

• 分支已拉取核对。注意:PR head 已从 77c60b09f 推进到 25eaa12e8——两个原始提交之后有一次「merge origin/master」提交(并入 [PR] test(e2e): scripted compress calls must use the single-string form (#4823 follow-up) #4873/[PR] feat: serve a bundled favicon.ico at /favicon.ico so browser icon probes never relay upstream #4835/[PR] External summary failures: name the real abort source and expose request size (#4833) #4839/[PR] fix(dsh): make dsh.contextTab work on DSH Desktop — always emit the __BILI__ boot-snapshot row #4848/[PR] docs: message-identity hierarchy is now msgid > order > hash (#4491/#4834) #4871 等 13 个 master 提交),分支现在与 master 对齐(距离 0)。该 merge 触碰了 PR 自己改过的 src/agent/dsh-native.ts,冲突解决方式在 review 范围内。
• CI:旧 head 77c60b09f 除 lifecycle-ledger-guard 外全绿。红因明确:新增 env var BILI_DSH_KEEP_COMMAND_COMPACT 被 src/ 读取但未登记进 LIFECYCLE.md ledger(guard 输出点名该变量)。master HEAD 全绿 → 非继承性红,是 PR 侧遗漏。新 head 的 CI 正在重跑中。
• 处置:按 CI 协调政策(master 绿 = 自己的问题自己修),我直接在 PR 分支补上缺失的 ledger 条目(独立标注 commit),然后继续完整 review(原 issue 方向/回归审计/代码质量/测试有效性)+ 本地定向测试,结论按八问回帖。

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Merge-guard sync done: merged origin/master into the branch cleanly (no conflicts; 77c60b0 → 25eaa12), local typecheck passes, pushed to 2026-10-27_dsh-keep-command-compact. Verified via API: head=25eaa12e8, mergeable_state=blocked — i.e. in sync with master, only CI checks pending on the new head. No design decisions required.

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

CI 确认:新头 337a65835(含 LIFECYCLE.md 台账登记修复)全部 check 绿(rollup SUCCESS,含 lifecycle-ledger-guard 与 ubuntu/windows × node22/24 全部分片)。无阻塞项,可以合并:#4883

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Resynced again (master moved twice in between): merged current master cleanly into the branch, no conflicts, local typecheck passes — pushed 337a658 → 995d1ed → 6f0ba1a. Verified via API: head=6f0ba1a32, 0 commits behind master, mergeable_state=blocked = in sync, only CI checks pending on the new head. No design decisions required.

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

七问 — PR #4883 conflict-resolution run (attempt 1/3)

1. What was the blocker? The PR branch 2026-10-27_dsh-keep-command-compact was 1766 commits behind master (upstream merge state dirty/behind) — the merge_conflict blocker.

2. What did I do? Merged origin/master (33104b1) into the PR branch — a merge, not a rebase, matching the established merge-guard pattern of the prior merge commits already on this branch.

3. What conflicts, and how resolved? 8 conflicts in 6 files. All were the [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 doc-reorg (root docs → docs/ + redirect stubs) plus one code file:

   • src/agent/dsh-native.ts (2): kept master's dsh Long-Session Drift Comprehensive Remediation: Sealing the Last Leak Window for Native Compression Local Deployment + | dsh 长会话漂移综合治理:堵死原生压缩本地落地的最后漏窗 + 会话身份分叉可见化与自愈 #5686 boot-note latch structure verbatim; re-injected the PR's keepCommandCompact text into the web-app branch and the non-web trace branch (using master's docs/configuration.md path, not the PR's stale CONFIGURATION.md).
   • CONFIGURATION.md + .zh-CN.md: root = master's redirect stub; ported the PR's 5 keepCommandCompact entries into docs/configuration(.zh-CN).md.
   • CLIENTS.md + .zh-CN.md: root = master's stub; ported the PR's keepCommandCompact dsh-section edits into docs/clients(.zh-CN).md.
   • LIFECYCLE.md: root = master's stub; ported the PR's BILI_DSH_KEEP_COMMAND_COMPACT ledger row into docs/lifecycle.md.
   • Auto-merged clean (no action): src/config.ts, src/knobs.ts, src/web/client.ts, src/web/i18n.ts, tests/dsh-native.test.ts, website/config-reference/{env-only,server}.yaml.

4. What did I verify? npm run typecheck clean (kernel + pi-subagents built, tsc --noEmit no errors); node --import tsx --test tests/dsh-native.test.ts = 76/76 pass. (Full-suite verdict is the remote CI on the pushed head, per §7.2 — not a local full run.)

5. Did I push? Force or clean? Clean fast-forward push — the remote head (2470631) was an ancestor of my merge commit, so no force-push was needed. Fetched + checked the remote head immediately before pushing, per push discipline.

6. Current state? Branch is now 0 behind / 12 ahead of master — the merge_conflict blocker is cleared. The mirror check cache is still stale (fetched_at 2026-10-31, sha 2470631 = the pre-merge head); the new CI run on d61d472 is pending and not yet visible in the cache, so I cannot assert its green/red live in this run.

7. What remains / risk? Only the remote CI on the new head remains (re-runs automatically). Risk is low: the conflict was doc-layout reorg + one code file where I kept master's structure verbatim and only re-injected the PR's existing keepCommandCompact text — no new behavior, no config-surface change, no kernel/ touch. If CI is red, I'll triage it in this same lane (attempt 2/3).

---

八问（中文，供人工评审）:

1. 是什么问题:PR [PR] feat(dsh): keepCommandCompact — opt-in overlay scope keeping the explicit /compact command under bili (#4872) #4883 分支落后 master 1766 个提交,上游报 merge_conflict(dirty),无法合并。
2. 什么原因导致:分支自 2026-10-27 起未同步 master,期间 master 完成 [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 文档集中化(根级文档迁 docs/ + 重定向占位)及大量其他提交,与本 PR 触碰的 6 个文件产生文本冲突。
3. 影响范围与严重程度:仅本 PR 分支的合并状态,不影响任何已发布行为;默认 OFF 的 keepCommandCompact 零漂移,默认路径字节不变。
4. 怎么修复的:merge origin/master 进分支(非 rebase,沿用本分支既有 merge-guard 模式);8 处冲突全部按「保留 master 结构 + 重新注入 PR 的 keepCommandCompact 文本」解决,文档按 [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 布局落进 docs/ 而非根级。
5. 是否有回归问题:无。dsh-native.ts 两处冲突均逐字保留 master dsh Long-Session Drift Comprehensive Remediation: Sealing the Last Leak Window for Native Compression Local Deployment + | dsh 长会话漂移综合治理:堵死原生压缩本地落地的最后漏窗 + 会话身份分叉可见化与自愈 #5686 结构、仅插入 PR 既有文本;其余 5 个文件是文档占位/迁移,无行为语义。typecheck 干净、dsh-native 定向测试 76/76 绿。
6. 是否破坏缓存:不涉及——未触碰缓存/重试/游标/去重键任何路径。
7. 是否改了设计方向:不涉及——纯冲突消解,未改架构/边界/数据流。
8. 是否引入了新的配置:不涉及——keepCommandCompact 配置面是 PR 本身已声明的(非本次冲突消解新增),本次仅把它落进 reorg 后的 docs/ 位置。

判定:冲突已清(0 behind master),merge 已推(d61d472,干净 push 无 force),等远程 CI 在新 head 上重跑;CI 红则本车道 attempt 2/3 继续。

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Status (re-trigger, live-verified this run): This is a re-fire of the already-resolved conflict — no new work done, nothing re-pushed.

Live state as of this run (fresh git fetch + mirror cache):

• Conflict resolved: branch 2026-10-27_dsh-keep-command-compact is 0 behind / 12 ahead of master; master tip (33104b1) is an ancestor of HEAD. The mirror's mergeable_state reads behind (NOT dirty) — GitHub's async mergeability computation is just lagging the push; the authoritative git check confirms no conflict.
• CI running on the new head (d61d472, mirror fetched 08:38:23Z): 5 success / 13 skipped / 12 in progress / 0 failures. The 13 skips are the gated e2e/codex lanes (no gate env in this run) — normal.
• Remote head = d61d472 = my merge commit; no other agent moved the branch.

No further action this run. If CI lands red, this lane re-triggers (attempt 2/3) and I'll triage the failure; if green, the PR is mergeable for the owner. The full 七问 of the conflict-resolution work is in my previous comment.