dsh-deeppilot 0.9.7 → 0.9.9
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/COMPATIBILITY.md +74 -9
- package/README.md +33 -12
- package/README.zh-CN.md +21 -7
- package/bin/SHA256SUMS +6 -6
- package/bin/darwin-amd64/dsh-deeppilot-tunnel +0 -0
- package/bin/darwin-arm64/dsh-deeppilot-tunnel +0 -0
- package/bin/linux-amd64/dsh-deeppilot-tunnel +0 -0
- package/bin/linux-arm64/dsh-deeppilot-tunnel +0 -0
- package/bin/windows-amd64/dsh-deeppilot-tunnel.exe +0 -0
- package/bin/windows-arm64/dsh-deeppilot-tunnel.exe +0 -0
- package/docs/DSH_RELEASE_MEMORY.md +157 -8
- package/docs/NPM_PUBLISHING.md +29 -0
- package/lib/client.js +152 -4
- package/lib/client.js.map +1 -1
- package/lib/index.d.ts +90 -3
- package/lib/index.js +1077 -48
- package/lib/index.js.map +1 -1
- package/package.json +12 -12
package/COMPATIBILITY.md
CHANGED
|
@@ -1,8 +1,9 @@
|
|
|
1
1
|
# Compatibility
|
|
2
2
|
|
|
3
|
-
This branch targets exactly **
|
|
4
|
-
`latest
|
|
5
|
-
`0.2.1-alpha.5`, which the `alpha` channel has pointed at in turn
|
|
3
|
+
This branch targets exactly **five DSH releases**: `0.2.0-rc.2`, which the npm
|
|
4
|
+
`latest` channel points at; `0.2.1-alpha.3`, `0.2.1-alpha.4`, and
|
|
5
|
+
`0.2.1-alpha.5`, which the `alpha` channel has pointed at in turn; and
|
|
6
|
+
`0.2.1-rc.2`, which the `next` channel points at. All five are
|
|
6
7
|
named in both `engines.dsh` and `peerDependencies`; no other version is accepted.
|
|
7
8
|
DSH evaluates that declaration before composing a profile — on every start, not
|
|
8
9
|
only on install — so a newly published DSH release deactivates this plugin until
|
|
@@ -13,13 +14,14 @@ The decision and the reversal it represents are recorded in
|
|
|
13
14
|
[`docs/DSH_RELEASE_MEMORY.md`](./docs/DSH_RELEASE_MEMORY.md).
|
|
14
15
|
|
|
15
16
|
The development packages this branch builds and typechecks against stay pinned
|
|
16
|
-
to the exact release-candidate baseline, currently `0.2.0-rc.2`; the
|
|
17
|
-
|
|
17
|
+
to the exact release-candidate baseline, currently `0.2.0-rc.2`; the audited
|
|
18
|
+
`alpha` and `rc` hosts are covered by the isolated composes and live runs
|
|
19
|
+
recorded in the release memory.
|
|
18
20
|
|
|
19
21
|
| Component | Current contract | Evidence |
|
|
20
22
|
|---|---|---|
|
|
21
23
|
| Node.js | 22 or newer | Package engine and CI |
|
|
22
|
-
| DSH CLI and Host API | `engines.dsh` and `peerDependencies` both name `0.2.0-rc.2 \|\| 0.2.1-alpha.3 \|\| 0.2.1-alpha.4 \|\| 0.2.1-alpha.5` (dev pins `0.2.0-rc.2`) | Peer metadata, source typecheck, unit tests, bundle build, and config schema projection against a real 0.2.0-rc.2 CLI, plus isolated composes and live phone-protocol runs with the published 0.2.1-alpha.2, 0.2.1-alpha.3, and 0.2.1-alpha.4 CLIs,
|
|
24
|
+
| DSH CLI and Host API | `engines.dsh` and `peerDependencies` both name `0.2.0-rc.2 \|\| 0.2.1-alpha.3 \|\| 0.2.1-alpha.4 \|\| 0.2.1-alpha.5 \|\| 0.2.1-rc.2` (dev pins `0.2.0-rc.2`) | Peer metadata, source typecheck, unit tests, bundle build, and config schema projection against a real 0.2.0-rc.2 CLI, plus isolated composes and live phone-protocol runs with the published 0.2.1-alpha.2, 0.2.1-alpha.3, and 0.2.1-alpha.4 CLIs, an isolated 0.2.1-alpha.5 host where this plugin now composes, starts, and answers `/phone/health` with 200 on its own LAN listener, and a pre-widening 0.2.1-rc.2 host run where the unchanged declaration is refused at compose time (audits below) |
|
|
23
25
|
| iOS bridge | DeepPilot protocol v2 | Bridge protocol tests; device behavior must be checked with the running Host and app |
|
|
24
26
|
| Remote access | Tailscale Funnel on ports 443, 8443, or 10000 | Helper and supervisor tests |
|
|
25
27
|
|
|
@@ -71,6 +73,66 @@ carries the old top-level `debug: true` is accepted but that value is ignored:
|
|
|
71
73
|
verbose logging must be turned on again from the settings page. No other config
|
|
72
74
|
field changed shape, and the old key is not read anywhere in this version.
|
|
73
75
|
|
|
76
|
+
## 0.2.1-rc.2 audit outcome
|
|
77
|
+
|
|
78
|
+
The artifact-level audit of `dsh-v0.2.1-rc.2` against `dsh-v0.2.1-alpha.5` found
|
|
79
|
+
no change that reaches this plugin's call surface. Fifteen packages were compared
|
|
80
|
+
as whole trees rather than by `lib/` alone: all ten peer packages carry
|
|
81
|
+
**byte-identical `lib/` trees**, and across 137 modules **no export symbol is
|
|
82
|
+
added or removed**. The 245 dependency differences are spec-value moves only,
|
|
83
|
+
with zero structural difference: 78 `@deepseek-ai/dsh*` entries are pinned to
|
|
84
|
+
`0.2.1-rc.2`, and five vendor ranges are promoted from prerelease to GA —
|
|
85
|
+
`cordis ~4.0.5-alpha.2` → `~4.0.5`, `schemastery ~3.18.5-alpha.2` → `~3.18.5`,
|
|
86
|
+
plus `cordis-plugin-include`, `cordis-plugin-loader`, and `cordis-plugin-timer`.
|
|
87
|
+
No dependency package name is added or removed.
|
|
88
|
+
|
|
89
|
+
**One file outside `package.json` differs, and it is additive.**
|
|
90
|
+
`@deepseek-ai/dsh-web-app/cordis.patch.yml` gained one hunk / +3−0 that inserts
|
|
91
|
+
the `ui-deepseek-model-label-1017` bundle (the DeepSeek-V4.1-Flash-1017
|
|
92
|
+
adaptation named in the rc.2 release notes). It cannot reach this plugin: the
|
|
93
|
+
patch anchor this plugin targets, `id: directory-picker`, is byte-identical at
|
|
94
|
+
lines 113–114 on both sides, the new entry sits at lines 412–417, and the new
|
|
95
|
+
package name appears zero times in this plugin's `cordis.patch.yml` and
|
|
96
|
+
`package.json`.
|
|
97
|
+
|
|
98
|
+
That file is also why the audit method changed: **comparing only `lib/` would
|
|
99
|
+
have missed it**, and it is exactly the surface this plugin's own patch layer
|
|
100
|
+
acts on. Whole-tree comparison is now a required step of every future audit, not
|
|
101
|
+
a fallback.
|
|
102
|
+
|
|
103
|
+
The vendor peers need no widening either. `0.2.1-rc.2` vendors the same GA pair
|
|
104
|
+
as `0.2.1-alpha.5` — `cordis 4.0.5` and `schemastery 3.18.5` — and four
|
|
105
|
+
independent solver paths (host-tree and repository `semver`, each with and
|
|
106
|
+
without `includePrerelease`) agree that both already satisfy the existing
|
|
107
|
+
`~4.0.4 || ~4.0.5-alpha.1` / `~3.18.4 || ~3.18.5-alpha.1` ranges.
|
|
108
|
+
|
|
109
|
+
The activation verdict is the same shape as alpha.5: rc.2 resolves the
|
|
110
|
+
requirements path from a declared `engines.dsh` alone, so its refusal names that
|
|
111
|
+
single requirement rather than the ten peers, and `exempted` is false because no
|
|
112
|
+
profile exemption file exists on the audit machine. Before the widening, a real
|
|
113
|
+
rc.2 CLI refused the unchanged declaration —
|
|
114
|
+
|
|
115
|
+
```text
|
|
116
|
+
dsh: skipping profile bundle "dsh-deeppilot": … Plugin dsh-deeppilot@0.9.7 is incompatible with dsh 0.2.1-rc.2: requirements {"engines.dsh":"0.2.0-rc.2 || 0.2.1-alpha.3 || 0.2.1-alpha.4 || 0.2.1-alpha.5"}. … Exact-version exemption: not active.
|
|
117
|
+
```
|
|
118
|
+
|
|
119
|
+
— while the host itself still booted and served its web port. Three published
|
|
120
|
+
evaluators agree on all 14 candidates, and the widened declaration admits
|
|
121
|
+
`0.2.1-rc.2` while adding no other acceptance. **`0.2.1-rc.1` is deliberately not
|
|
122
|
+
added and remains refused**: the rc line reached this plugin one release later, so
|
|
123
|
+
admitting the earlier rc would extend support to a host that was never audited as
|
|
124
|
+
its own. `0.2.1-alpha.1`, `0.2.1-alpha.2`, `0.2.0-rc.1`, `0.2.0`, `0.2.1`,
|
|
125
|
+
`0.2.1-rc.3`, `0.3.0-rc.1`, `0.3.0`, `1.0.0`, and `0.1.7-rc.2` remain refused.
|
|
126
|
+
|
|
127
|
+
**Audit limits, stated rather than implied.** Loading this plugin on a real
|
|
128
|
+
isolated `0.2.1-rc.2` host — the positive half of the rc.2 verification — had
|
|
129
|
+
**not** been completed when this entry was written; the rc.2 evidence here is the
|
|
130
|
+
pre-widening refusal log plus the published evaluators' verdict on the widened
|
|
131
|
+
declaration. Until the positive load and the full release check set run against
|
|
132
|
+
that host, this release is recorded as audited at the declaration and
|
|
133
|
+
artifact-diff level only; no plugin release has been made from this tree, and
|
|
134
|
+
nothing here claims one.
|
|
135
|
+
|
|
74
136
|
## 0.2.1-alpha.5 audit outcome
|
|
75
137
|
|
|
76
138
|
The source-level audit of `dsh-v0.2.1-alpha.5` against `dsh-v0.2.1-alpha.4` found
|
|
@@ -210,9 +272,12 @@ needs a documentation update, not a peer-range widening.
|
|
|
210
272
|
> 0.2.1-alpha.3 outcome above and `docs/决策记录.md` (章节一). The paragraphs in
|
|
211
273
|
> this historical entry describe the policy as it stood on 2026-09-29.
|
|
212
274
|
> **This note is in turn superseded:** `0.2.1-alpha.4` was added the next day by
|
|
213
|
-
> the same deliberate audit step, `0.2.1-alpha.5` on 2026-10-16, and
|
|
214
|
-
>
|
|
215
|
-
>
|
|
275
|
+
> the same deliberate audit step, `0.2.1-alpha.5` on 2026-10-16, and
|
|
276
|
+
> `0.2.1-rc.2` on 2026-10-17 — each one its own audit, never an automatic
|
|
277
|
+
> consequence of a new DSH tag. The closed five-release list now lives in
|
|
278
|
+
> `engines.dsh` as well as in `peerDependencies`. `0.2.1-rc.1` shows zero code
|
|
279
|
+
> delta on this plugin's call surface but is deliberately left out, because it
|
|
280
|
+
> was never audited as its own host. See the 0.2.1-rc.2 outcome above.
|
|
216
281
|
|
|
217
282
|
## 0.2.0-rc.1 audit outcome
|
|
218
283
|
|
package/README.md
CHANGED
|
@@ -32,18 +32,19 @@ Host on your own computer and does not replace or modify the DSH Web UI.
|
|
|
32
32
|
|
|
33
33
|
## Install from npm
|
|
34
34
|
|
|
35
|
-
This source checkout targets Node.js 22+ and the
|
|
36
|
-
audited against — `0.2.0-rc.2` (the `latest
|
|
37
|
-
|
|
38
|
-
`web` profile. DSH checks a plugin's declared
|
|
39
|
-
profile, on every start rather than only on
|
|
40
|
-
refuses to activate this plugin. From DSH
|
|
41
|
-
`
|
|
42
|
-
range, so both name the same
|
|
35
|
+
This source checkout targets Node.js 22+ and the five DSH releases it has been
|
|
36
|
+
audited against — `0.2.0-rc.2` (the `latest` channel), the `alpha` channel
|
|
37
|
+
releases `0.2.1-alpha.3`, `0.2.1-alpha.4`, and `0.2.1-alpha.5`, and `0.2.1-rc.2`
|
|
38
|
+
(the `next` channel) — with a `web` profile. DSH checks a plugin's declared
|
|
39
|
+
requirements before composing a profile, on every start rather than only on
|
|
40
|
+
install, so any other host version refuses to activate this plugin. From DSH
|
|
41
|
+
`0.2.1-alpha.4` on, that check reads `engines.dsh` and ignores
|
|
42
|
+
`peerDependencies`; older hosts know only the peer range, so both name the same
|
|
43
|
+
five releases. The package includes Funnel helpers
|
|
43
44
|
for macOS, Linux, and Windows on amd64/arm64; trusted-LAN mode does not require
|
|
44
45
|
the helper.
|
|
45
46
|
|
|
46
|
-
The current working tree names exactly those
|
|
47
|
+
The current working tree names exactly those five versions in `engines.dsh` and
|
|
47
48
|
`peerDependencies` and is built/typechecked against `0.2.0-rc.2`. A previously
|
|
48
49
|
published plugin package may still carry its own older compatibility
|
|
49
50
|
declaration until a release is made from this tree.
|
|
@@ -124,10 +125,30 @@ system-reported name; saving an empty value restores the iPhone-reported
|
|
|
124
125
|
default.
|
|
125
126
|
|
|
126
127
|
The settings page's **Testing & troubleshooting** group has a `debug` toggle.
|
|
127
|
-
Turning it on prints
|
|
128
|
+
Turning it on prints per-frame and push diagnostics to the DSH host's terminal
|
|
128
129
|
output (for example, the actual capability bits sent in `s2c.welcome`); it
|
|
129
|
-
never prints pairing tokens, APNs tokens, or message content
|
|
130
|
-
|
|
130
|
+
never prints pairing tokens, APNs tokens, or message content. The toggle is
|
|
131
|
+
snapshotted per connection, so an already-open connection only picks up the new
|
|
132
|
+
value after it reconnects.
|
|
133
|
+
|
|
134
|
+
**Connection failures do not need that toggle.** The plugin always records
|
|
135
|
+
admission rejections, authentication failures (4401 distinguishes an unknown
|
|
136
|
+
device, a revoked device, a bad signature, an expired challenge and clock skew),
|
|
137
|
+
TLS handshake failures and pairing failures. They show up in the terminal as
|
|
138
|
+
`[deeppilot] <time> WARN <event> <reason> src=<source hash> attempt=<attempt ID>`
|
|
139
|
+
and are appended under `diagnostics/` in the plugin data directory (7 days,
|
|
140
|
+
5 MB cap). The settings page's **Connection diagnostics** group lists the most
|
|
141
|
+
recent failures with a copy button, and `/phone/health` additionally returns the
|
|
142
|
+
local transport phase and port, the data-plane phase, whether the TLS identity
|
|
143
|
+
was rebuilt, the start time and the last failure, so the phone can locate the
|
|
144
|
+
cause even while it cannot connect. Records contain only source hashes — never
|
|
145
|
+
raw IPs, pairing codes, keys, APNs tokens or message content.
|
|
146
|
+
|
|
147
|
+
Each connection attempt from the phone carries a random `attemptId` (the `?at=`
|
|
148
|
+
query parameter on the WebSocket upgrade, or the `X-DeepPilot-Attempt` header on
|
|
149
|
+
pairing and health checks). The plugin writes it into the log lines and
|
|
150
|
+
diagnostic events above, so a user reporting that ID lets you find the exact
|
|
151
|
+
attempt in the host log.
|
|
131
152
|
|
|
132
153
|
Offline push is optional. Relay mode sends only the target APNs device token
|
|
133
154
|
and a limited notification payload; full conversation history and live output
|
package/README.zh-CN.md
CHANGED
|
@@ -30,16 +30,17 @@
|
|
|
30
30
|
|
|
31
31
|
## 从 npm 安装
|
|
32
32
|
|
|
33
|
-
当前源码面向 Node.js 22+ 与带 `web` profile
|
|
34
|
-
`0.2.0-rc.2`(`latest
|
|
35
|
-
`0.2.1-alpha.4` 与 `0.2.1-alpha.5
|
|
33
|
+
当前源码面向 Node.js 22+ 与带 `web` profile 的**五个已审计 DSH 版本**:
|
|
34
|
+
`0.2.0-rc.2`(`latest` 通道)、`alpha` 通道的 `0.2.1-alpha.3`、
|
|
35
|
+
`0.2.1-alpha.4` 与 `0.2.1-alpha.5`,以及 `0.2.1-rc.2`(`next` 通道)。
|
|
36
|
+
DSH 在组合 profile 前会校验插件声明的
|
|
36
37
|
兼容要求,且**每次启动都会校验**(不只是安装时),因此其他宿主版本会拒绝
|
|
37
38
|
激活本插件。从 DSH `0.2.1-alpha.4` 起,该校验只读 `engines.dsh` 并忽略
|
|
38
|
-
`peerDependencies`;更早的宿主只认 peer
|
|
39
|
+
`peerDependencies`;更早的宿主只认 peer 范围,因此两处列出同一组五个版本。
|
|
39
40
|
npm 包内置 macOS、Linux、Windows 的 amd64/arm64 Funnel helper;可信局域网
|
|
40
41
|
模式不依赖 helper。
|
|
41
42
|
|
|
42
|
-
当前工作树的 `engines.dsh` 与 `peerDependencies`
|
|
43
|
+
当前工作树的 `engines.dsh` 与 `peerDependencies` 精确列出这五个版本,构建与
|
|
43
44
|
类型检查针对 `0.2.0-rc.2`。从此工作树发布新版本前,已发布的旧插件包仍可能
|
|
44
45
|
带有各自旧的兼容声明。
|
|
45
46
|
|
|
@@ -107,8 +108,21 @@ DeepPilot 设置页提供“每个公网来源的连接上限”,默认 `8`,
|
|
|
107
108
|
也不会被覆盖。留空保存可恢复 iPhone 上报的默认名称。
|
|
108
109
|
|
|
109
110
|
设置页的“测试与故障排查”分组下有一个 `debug` 开关,打开后会在 DSH host 的
|
|
110
|
-
|
|
111
|
-
不会打印配对令牌、APNs token
|
|
111
|
+
终端输出里打印逐帧与推送等诊断信息(例如 `s2c.welcome` 里实际发出的能力位),
|
|
112
|
+
不会打印配对令牌、APNs token 或消息内容。该开关按连接快照,已建立的连接需要
|
|
113
|
+
重连一次才按新值工作。
|
|
114
|
+
|
|
115
|
+
**连不上时不需要打开这个开关**:插件会无条件记录准入拒绝、认证失败(4401 会区分
|
|
116
|
+
设备不存在、已撤销、签名错、挑战过期、时钟偏差等原因)、TLS 握手失败与配对失败,
|
|
117
|
+
在终端输出为 `[deeppilot] <时间> WARN <事件> <原因> src=<来源哈希> attempt=<尝试 ID>`,
|
|
118
|
+
并落盘到插件数据目录的 `diagnostics/`(保留 7 天、上限 5 MB)。设置页“连接诊断”
|
|
119
|
+
分组可查看最近失败并一键复制;`/phone/health` 额外返回本地传输相位与端口、数据面
|
|
120
|
+
相位、是否重建过 TLS 身份、启动时间与最近一次失败,手机端在连不上时也能据此定位。
|
|
121
|
+
记录只含来源哈希,不含原始 IP、配对码、密钥、APNs token 或消息内容。
|
|
122
|
+
|
|
123
|
+
手机端每次连接尝试会带一个随机的 `attemptId`(WebSocket 升级的 `?at=` 查询参数,
|
|
124
|
+
或配对/健康检查的 `X-DeepPilot-Attempt` 请求头)。插件把它写进上面的日志与诊断
|
|
125
|
+
事件;用户反馈时报告这个 ID,就能在主机日志里定位到同一次尝试。
|
|
112
126
|
|
|
113
127
|
离线推送是可选功能。中继模式只发送目标 APNs 设备 Token 和有限的通知内容;
|
|
114
128
|
完整会话历史与实时输出不会经过中继。启用远程访问或推送前,请阅读
|
package/bin/SHA256SUMS
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
|
-
|
|
2
|
-
|
|
3
|
-
|
|
4
|
-
|
|
5
|
-
|
|
6
|
-
|
|
1
|
+
4e83f82d5c623115d121f58264ca84706e5fc935dadc0747b37ae886d102dc6a ./darwin-amd64/dsh-deeppilot-tunnel
|
|
2
|
+
7707999a5b745d11afeb4ab1868c85680e4ff5567202d97015d72d810817a134 ./darwin-arm64/dsh-deeppilot-tunnel
|
|
3
|
+
8a83351568cb223854efbe8f54ca9a403fc2b152382d4e7c952d372b0b0cb3e9 ./linux-amd64/dsh-deeppilot-tunnel
|
|
4
|
+
dfa4eb4dd50b386aadf40d9767ac0df8151308b620fa395d0a560498f01eb0cb ./linux-arm64/dsh-deeppilot-tunnel
|
|
5
|
+
9c734ea42bc6dd5763930991db52f910d5138e9b12df0e826d6964ceb271e215 ./windows-amd64/dsh-deeppilot-tunnel.exe
|
|
6
|
+
e7db4a0e62f04266d97a5bad38e1a21961474440e7231fd5b24980053589faa0 ./windows-arm64/dsh-deeppilot-tunnel.exe
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
@@ -3,10 +3,126 @@
|
|
|
3
3
|
> This is the durable audit record for DeepPilot's public DSH plugin. Re-audit
|
|
4
4
|
> the next DSH release instead of assuming that an rc upgrade is runtime-safe.
|
|
5
5
|
|
|
6
|
-
## Current decision — `dsh-v0.2.1-
|
|
6
|
+
## Current decision — `dsh-v0.2.1-rc.2`
|
|
7
|
+
|
|
8
|
+
- **Audited release:** `dsh-v0.2.1-rc.2`, published 2026-10-17T08:36:31.655Z. It is
|
|
9
|
+
**not** `latest`: `latest` still points at `0.2.0-rc.2`, and `alpha` still points
|
|
10
|
+
at `0.2.1-alpha.5`. It **is** the npm **`next`** channel release — this is the
|
|
11
|
+
first release on the rc line to enter the supported list.
|
|
12
|
+
- **Comparison baseline:** `dsh-v0.2.1-alpha.5` (the previous audited tag).
|
|
13
|
+
- **Plugin runtime verdict: no break, and no `src/` change was needed.** Fifteen
|
|
14
|
+
packages were compared **as whole trees, not by `lib/` alone**: all ten peer
|
|
15
|
+
packages ship a byte-identical `lib/` tree, and across 137 modules **no export
|
|
16
|
+
symbol is added or removed**, so every earlier fix still holds unchanged.
|
|
17
|
+
- **The one file outside `package.json` that differs is additive.**
|
|
18
|
+
`@deepseek-ai/dsh-web-app/cordis.patch.yml` gained one hunk / +3−0 inserting the
|
|
19
|
+
`ui-deepseek-model-label-1017` bundle (the DeepSeek-V4.1-Flash-1017 adaptation
|
|
20
|
+
named in the release notes). The patch anchor this plugin targets,
|
|
21
|
+
`id: directory-picker`, is byte-identical at lines 113–114 on both sides, the new
|
|
22
|
+
entry sits at lines 412–417, and the new package name appears zero times in this
|
|
23
|
+
plugin's `cordis.patch.yml` and `package.json`, so the change cannot reach any
|
|
24
|
+
patch this plugin applies.
|
|
25
|
+
- **Methodology change, now mandatory.** Comparing only `lib/` would have missed
|
|
26
|
+
exactly that file — and it is the surface this plugin's own patch layer acts on.
|
|
27
|
+
Whole-tree comparison is a required step of every future audit; do not fall back
|
|
28
|
+
to a `lib/`-only diff. See "Re-audit rule" below.
|
|
29
|
+
- **Activation verdict: the new release is admitted by both declarations, and the
|
|
30
|
+
earlier rc on the same line is deliberately left out.** This is the fourth
|
|
31
|
+
consecutive release on this line to trip the closed-list policy: rc.2 was refused
|
|
32
|
+
before the widening. rc.2 reads a declared `engines.dsh` alone, so its refusal
|
|
33
|
+
reported a **single** `engines.dsh` requirement rather than ten peers, and
|
|
34
|
+
`exempted: false` — the list is the only thing that admits a host, and no
|
|
35
|
+
exemption fallback exists on this machine (no profile `compatibility.json` under
|
|
36
|
+
`DSH_HOME`). Widening admits `0.2.1-rc.2` and nothing else. `0.2.1-rc.1` stays
|
|
37
|
+
refused even though it shows **zero code delta on this plugin's call surface**:
|
|
38
|
+
it was never audited as its own host, and the rc line reached this plugin one
|
|
39
|
+
release later.
|
|
40
|
+
- **Evidence: the published artifacts, the published evaluators, and a real rc.2
|
|
41
|
+
CLI run — all taken before any declaration changed.**
|
|
42
|
+
- **Change surface:** `.rc2-verify/CHANGE_SURFACE.md`. Raw logs:
|
|
43
|
+
`.rc2-verify/logs/lib-hash-compare.txt` and `lib-diff-exitcodes.txt` (10 of 10
|
|
44
|
+
peer `lib/` trees byte-identical), `whole-tree-compare.txt` (the 15-package
|
|
45
|
+
whole-tree fallback, which is what surfaced the `dsh-web-app` patch file),
|
|
46
|
+
`export-symbol-diff.out.json` (137 modules, zero symbol changes),
|
|
47
|
+
`pkg-manifest-delta.out.json` (245 spec-value differences, zero structural
|
|
48
|
+
differences), `webapp-cordis-patch-diff.txt` and `patch-anchor-verification.txt`
|
|
49
|
+
(the one differing file and its anchor), `rc1-delta.txt` (`0.2.1-rc.1` has zero
|
|
50
|
+
code delta on this plugin's call surface), and `evaluator-repro.out.json`
|
|
51
|
+
(three published evaluators × 14 candidates, all agreeing).
|
|
52
|
+
- **Activation decision, reproduced against rc.2's own published evaluator
|
|
53
|
+
before any edit:** `.rc2-verify/ACTIVATION.md` and `rc2-matrix.out.json`.
|
|
54
|
+
Unchanged, rc.2 was refused with a single `engines.dsh` requirement and
|
|
55
|
+
`exempted: false`; deleting `engines.dsh` from a copy of the manifest fell back
|
|
56
|
+
to all ten peers, confirming the short-circuit is unchanged. Projecting the
|
|
57
|
+
widening in memory admitted exactly `0.2.1-rc.2` with zero new refusals, while
|
|
58
|
+
`0.2.1-rc.3`, `0.2.1-rc.20`, `0.2.1-rc.2.1`, `0.2.1-RC.2`, and `0.2.1-rc.10`
|
|
59
|
+
stayed refused (`rc2-matrix-projected.out.json`).
|
|
60
|
+
- **Real CLI refusal, captured while the declaration was still unchanged:**
|
|
61
|
+
`.rc2-verify/PRECHANGE-BOOT.md`. Under an isolated `DSH_HOME` the real rc.2 CLI
|
|
62
|
+
printed `dsh: skipping profile bundle "dsh-deeppilot": … incompatible with dsh
|
|
63
|
+
0.2.1-rc.2: requirements {"engines.dsh":…}` plus `Exact-version exemption: not
|
|
64
|
+
active.`, `--dump-config` exited 0 with no plugin row in the composed tree, and
|
|
65
|
+
the host itself still booted (web port answered, while the plugin's own port
|
|
66
|
+
3098 had no listener). Raw logs: `PRECHANGE-dumpconfig.stderr.txt`,
|
|
67
|
+
`PRECHANGE-dumpconfig.stdout.txt`, `PRECHANGE-host.log`, `PRECHANGE-probes.txt`.
|
|
68
|
+
A control run of the same profile under the `0.2.1-alpha.5` CLI — which admits
|
|
69
|
+
`0.9.7` — loaded the plugin and answered `200` on port 3098
|
|
70
|
+
(`PRECHANGE-control-alpha5-host.log`, `PRECHANGE-control-alpha5-probes.txt`),
|
|
71
|
+
showing the negative probe has discriminating power.
|
|
72
|
+
- **Vendor pins: no range change needed.** rc.2 vendors the same GA pair as
|
|
73
|
+
alpha.5 — `cordis 4.0.5` and `schemastery 3.18.5` — and four independent solver
|
|
74
|
+
paths (host-tree and repository `semver`, each with and without
|
|
75
|
+
`includePrerelease`) agree both already satisfy the existing `~4.0.4 ||
|
|
76
|
+
~4.0.5-alpha.1` / `~3.18.4 || ~3.18.5-alpha.1` ranges
|
|
77
|
+
(`.rc2-verify/rc2-vendor.out.json`). Neither vendor published a new
|
|
78
|
+
per-DSH-release dist-tag for rc.2; their `latest` is already `4.0.5` / `3.18.5`.
|
|
79
|
+
- **Install verdict: the two declarations now agree on five releases.**
|
|
80
|
+
`engines.dsh` and all ten `@deepseek-ai/dsh*` peers name `0.2.0-rc.2 ||
|
|
81
|
+
0.2.1-alpha.3 || 0.2.1-alpha.4 || 0.2.1-alpha.5 || 0.2.1-rc.2`, mirrored in
|
|
82
|
+
`tests/compatibility-metadata.test.ts`, `package-lock.json` (root entry only),
|
|
83
|
+
both READMEs, `COMPATIBILITY.md`, `AGENTS.md`, and the local
|
|
84
|
+
`docs/决策记录.md` (章节一, 条目5). The tests now also pin `0.2.1-rc.1` and
|
|
85
|
+
`0.2.1-rc.3` as negative probes, so neither the skipped rc nor the next rc can be
|
|
86
|
+
admitted by accident. The dev pins and `DSH_BASELINE` stay at the exact
|
|
87
|
+
`0.2.0-rc.2` this branch builds and typechecks against.
|
|
88
|
+
- **Protocol verdict: no phone-protocol change.** Protocol v2, pairing state, and
|
|
89
|
+
device records are untouched by this release; no phone needs to re-pair.
|
|
90
|
+
- **The limits this entry originally stated are now closed.** The positive half of
|
|
91
|
+
the rc.2 verification ran before publication: on an isolated `0.2.1-rc.2` host
|
|
92
|
+
carrying the widened declaration, `--dump-config` wrote nothing to stderr, the
|
|
93
|
+
composed tree contained `# == dsh-deeppilot` (line 1233) together with
|
|
94
|
+
`# == @deepseek-ai/dsh-web-app, patched by dsh-deeppilot` (line 449), the running
|
|
95
|
+
host logged four `[deeppilot]` lines, and the plugin's own LAN listener on port
|
|
96
|
+
3098 answered `/phone/health` with `200` and `serverVersion` `0.9.8`. A tamper
|
|
97
|
+
control removed the `0.2.1-rc.2` token from a copy of `engines.dsh`: the host
|
|
98
|
+
refused again with the single `engines.dsh` requirement (`exempted: false`) and
|
|
99
|
+
accepted it once restored. Independent verification of the whole release tree —
|
|
100
|
+
`npm ci`, 581 tests (580 pass, 0 fail, 1 pre-existing TODO), typecheck, build
|
|
101
|
+
with a byte-identical regenerated `lib/`, `check:config-schema`, the Go helper
|
|
102
|
+
tests, and all six `bin/SHA256SUMS` entries — is recorded in
|
|
103
|
+
`.rc2-verify/VERIFICATION.md`; that pass also caught a real defect, a version
|
|
104
|
+
bump that had left `tests/apply-baseline.snapshot.json` stale.
|
|
105
|
+
- **Published as 0.9.8 on 2026-10-18.** The annotated tag `v0.9.8` dereferences to
|
|
106
|
+
`f4551b61602e5e60b11565ae31ccba3257d3dc9e` on `main`, CI and CodeQL were green on
|
|
107
|
+
that commit, and `publish.yml` ran on the tag (run `39048037503`, 24/24 steps).
|
|
108
|
+
The registry records `latest: 0.9.8`, published `2026-10-17T16:12:59.699Z`, with a
|
|
109
|
+
`https://slsa.dev/provenance/v1` attestation naming that commit as its source. The
|
|
110
|
+
published tarball's SHA1 `8288977711e04679d077b970f863e726631fc309` and its SHA512
|
|
111
|
+
match the registry, its 68 entries and 130,945,335 unpacked bytes match the
|
|
112
|
+
pre-publication pack, and every packaged file compared byte-identically with the
|
|
113
|
+
verified tree. A clean isolated profile then installed `dsh-deeppilot@0.9.8` from
|
|
114
|
+
npm, composed it on a real `0.2.1-rc.2` host, and answered `/phone/health` with
|
|
115
|
+
`200` and `serverVersion: "0.9.8"`; the GitHub Release was made public only after
|
|
116
|
+
that install verification. The same release shipped the mobile skills, host
|
|
117
|
+
commands, and catalog-cache work recorded below.
|
|
118
|
+
|
|
119
|
+
### Previous decision — `dsh-v0.2.1-alpha.5`
|
|
7
120
|
|
|
8
121
|
- **Audited release:** `dsh-v0.2.1-alpha.5`, the newest `alpha` channel release.
|
|
9
122
|
It is **not** `latest`; `latest`/`next` still point at `0.2.0-rc.2`.
|
|
123
|
+
*(Superseded 2026-10-17 by `dsh-v0.2.1-rc.2`: `next` now points at
|
|
124
|
+
`0.2.1-rc.2`. `alpha` still points at `0.2.1-alpha.5` and `latest` still points
|
|
125
|
+
at `0.2.0-rc.2`. See the current decision above.)*
|
|
10
126
|
- **Comparison baseline:** `dsh-v0.2.1-alpha.4` (the previous audited tag).
|
|
11
127
|
- **Plugin runtime verdict: no break, and no `src/` change was needed.** Nine of
|
|
12
128
|
the ten peer packages ship a byte-identical `lib/` tree, including
|
|
@@ -91,10 +207,12 @@
|
|
|
91
207
|
committed tree), `npm run check:config-schema` against the pinned
|
|
92
208
|
`0.2.0-rc.2` CLI, `go test ./...` for the helper, and all six
|
|
93
209
|
`bin/SHA256SUMS` entries verified.
|
|
94
|
-
- **
|
|
95
|
-
|
|
96
|
-
|
|
97
|
-
|
|
210
|
+
- **Published as 0.9.7 on 2026-10-17.** `npm view dsh-deeppilot dist-tags` reads
|
|
211
|
+
`latest: 0.9.7`, and the version carries an npm provenance attestation
|
|
212
|
+
(`https://slsa.dev/provenance/v1`). The release commit `bf18755` is on `main`,
|
|
213
|
+
the annotated tag `v0.9.7` dereferences to it, and the GitHub release was made
|
|
214
|
+
public only after a clean-profile install had been verified — see the publish
|
|
215
|
+
verification section below.
|
|
98
216
|
|
|
99
217
|
### Previous decision — `dsh-v0.2.1-alpha.4`
|
|
100
218
|
|
|
@@ -588,9 +706,13 @@ close that gap rather than assume it.
|
|
|
588
706
|
`npm run typecheck`, `npm run build` with a byte-identical regenerated `lib/`,
|
|
589
707
|
`npm run check:config-schema` against the pinned `0.2.0-rc.2` CLI, the Go helper
|
|
590
708
|
tests, and the six `bin/SHA256SUMS` entries.
|
|
591
|
-
- **
|
|
592
|
-
|
|
593
|
-
|
|
709
|
+
- **Release action completed — 0.9.7 published on 2026-10-17.** The annotated tag
|
|
710
|
+
`v0.9.7` points at `bf18755` on `main`; the publish workflow ran on that tag and
|
|
711
|
+
all of its steps passed, including the post-publish registry check whose retry
|
|
712
|
+
window this release widened. Registry `latest` reads `0.9.7` with provenance,
|
|
713
|
+
and a clean isolated profile installed `dsh-deeppilot@0.9.7` from npm, composed
|
|
714
|
+
it on a real `0.2.1-alpha.5` host, and answered `/phone/health` with `200` and
|
|
715
|
+
`serverVersion: "0.9.7"`.
|
|
594
716
|
- **The declaration mirrors have drifted historically.** `README.md`,
|
|
595
717
|
`README.zh-CN.md`, `COMPATIBILITY.md`, and `AGENTS.md` all said "three
|
|
596
718
|
releases" while `package.json` already named four; that contradiction was
|
|
@@ -1644,3 +1766,30 @@ For every future DSH release, compare the new tag against the last audited tag,
|
|
|
1644
1766
|
inspect the package/service surfaces actually imported by this repository,
|
|
1645
1767
|
classify install/runtime/protocol risk separately, and append a dated decision
|
|
1646
1768
|
here. Do not treat a release-note summary as proof of compatibility.
|
|
1769
|
+
|
|
1770
|
+
Compare **whole package trees**, not just `lib/`. The rc.2 audit found its only
|
|
1771
|
+
non-`package.json` difference in `@deepseek-ai/dsh-web-app/cordis.patch.yml`, a
|
|
1772
|
+
file outside every `lib/` directory and the exact surface this plugin's own patch
|
|
1773
|
+
layer acts on; a `lib/`-only diff would have reported "no difference" and missed
|
|
1774
|
+
it. Take the published registry tarballs as the source of truth, enumerate the
|
|
1775
|
+
package set explicitly (peers, plus every package raw `dsh` dependencies add or
|
|
1776
|
+
remove, plus any peer that is not hoisted into the host tree), and only then read
|
|
1777
|
+
release notes.
|
|
1778
|
+
|
|
1779
|
+
## 2026-10-15 — 移动端技能发现扩展
|
|
1780
|
+
|
|
1781
|
+
新增可选的 `skills` 手机能力,使用官方 `sessionSkillCatalog.list({sessionId}, AbortSignal)` 服务,已对照本机 `0.2.0-rc.2` 和 `0.2.1-alpha.3` 发行声明。缺少服务只关闭该特性,晚挂载服务会重新探测,不缓存缺席。未修改宿主版本范围、鉴权、包版本或命令派发契约。手机只接收元数据,技能注册表和会话预设、工作目录变化触发目录失效。完整文档清单与四类差异表见《技能提示对接说明》。此源码改动随 0.9.8(2026-10-18)发布;主机侧集成已在本机隔离真实宿主上验收,真机技能加载仍未验收。
|
|
1782
|
+
|
|
1783
|
+
## 2026-10-16 — 移动端系统指令扩展
|
|
1784
|
+
|
|
1785
|
+
1. 安装:本轮命令改动未调整依赖范围、包版本或发布状态。命令接口核对基于 `0.2.0-rc.2` 与 `0.2.1-alpha.3`;工作区并行的 `alpha.4` 兼容改动以其独立审计为准,不把本轮合成测试算作新版本真实业务验收。
|
|
1786
|
+
2. 运行:对照两版生成声明和实现,使用公开 `typertGateway.invoke` 的 `commands/list`、`commands/execute`、`fileUploads/upload`、`permissionPresets/catalog`、`sessionFeedback/record`;当前权限来自会话投影,日志复用官方共享 Fetch 归档路由。alpha.3 创建不参与审批的 owner Peer,rc.2 回退既有 operator。详情及上游全接口清单见[移动端命令对接说明](移动端命令对接说明.md)。
|
|
1787
|
+
3. 手机协议:v2 可选 `commands` 能力和五个新请求;读取使用 `sessions.read`,执行/反馈使用 `prompt.send`,小组件默认拒绝。旧手机忽略新字段,旧桥接不展示新增入口。导出分块不重放,接收方完整校验后分享。
|
|
1788
|
+
4. 验证:576 项测试通过、1 项原有待办;类型检查、构建和配置检查通过。新增逻辑为合成服务测试,未在两版真实 Host 会话执行系统指令,不据此声称运行验收或发布完成。
|
|
1789
|
+
|
|
1790
|
+
## 2026-10-16 — 移动端目录缓存
|
|
1791
|
+
|
|
1792
|
+
1. 安装与主机 API:本轮不改安装范围或依赖;读取沿用现有技能、命令和模型接口。工作区另有 alpha.4 兼容改动,命令对接已重新枚举该版 29 包、144 个接口,相关主机实现与 alpha.3 相同;不据此宣称真实业务验收。
|
|
1793
|
+
2. 手机协议:模型当前选择投影新增 `s2c.models.changed`,受 `sessions.read` 控制;原生权限执行可附加 `expectedDefinitionId`,重读目录核对来源,旧客户端字段缺省兼容。无新增周期轮询或第二层主机时间缓存。
|
|
1794
|
+
3. 客户端:内存五分钟、磁盘最长七天,重启旧目录先展示并校验,按配对身份/主机/会话隔离;当前模型、权限和执行结果不落盘。具体约定和格式重建见[移动端目录缓存](移动端目录缓存.md)。
|
|
1795
|
+
4. 验证:插件 577 项通过、1 项原有待办,类型检查、生成产物构建和使用本机固定版本的离线配置校验通过;两端缓存单元与模拟器证据见各自文档。随 0.9.8(2026-10-18)发布;真机与系统分享仍未验收。
|
package/docs/NPM_PUBLISHING.md
CHANGED
|
@@ -87,6 +87,35 @@
|
|
|
87
87
|
线性递增并封顶 20 秒(约 7 分钟),仍然只读元数据、不重复上传包。这样既避免
|
|
88
88
|
「上传成功 + 核对延迟」在 Actions 页留下误导性的红色结论,也不会把真正的失败
|
|
89
89
|
伪装成成功——超出窗口依旧抛错。
|
|
90
|
+
6. `0.9.7` 是放宽窗口后的第一次发布,也是首次全绿的发布:工作流在附注标签
|
|
91
|
+
`v0.9.7`(提交 `bf18755`)上运行,19 个步骤全部成功,包含此前两次失败的末尾
|
|
92
|
+
核对。发布后核验:注册表 `latest = 0.9.7`、来源证明类型为
|
|
93
|
+
`https://slsa.dev/provenance/v1`;在全新隔离 profile 中从 npm 安装该版本,并在
|
|
94
|
+
真实 `0.2.1-alpha.5` 宿主上启动,`/phone/health` 返回 200 且 `serverVersion`
|
|
95
|
+
为 `0.9.7`。公开 GitHub Release 之前先完成了这一步安装验证。
|
|
96
|
+
7. `0.9.8` 的本地发布清单在**待发布工作树**上执行并通过(详细数字见
|
|
97
|
+
`docs/RELEASE_NOTES-0.9.8.md` 第三节):`npm ci`、单元测试 581 项
|
|
98
|
+
(580 通过、0 失败、1 项既有待办)、类型检查、构建前后 `lib/` 逐文件
|
|
99
|
+
sha256 一致、`check:config-schema`(仍针对锁定的 `0.2.0-rc.2` CLI)、
|
|
100
|
+
helper 的 Go 测试、`bin/SHA256SUMS` 6/6。宿主侧本轮改用
|
|
101
|
+
**`0.2.1-rc.2`**:隔离 `DSH_HOME` 下改动前被拒绝激活(单键 `engines.dsh`
|
|
102
|
+
要求 + `Exact-version exemption: not active`),扩围后组合树含插件与
|
|
103
|
+
web-app 补丁、真实启动出现 4 条 `[deeppilot]`,插件自有端口 3098 的
|
|
104
|
+
`/phone/health` 返回 200 且 `serverVersion` 为 `0.9.8`。打包为 68 个条目。
|
|
105
|
+
本轮还确认了一个易漏点:升版本号必须同步重录
|
|
106
|
+
`tests/apply-baseline.snapshot.json`,否则 `npm test` 会失败——已写入
|
|
107
|
+
`AGENTS.md` 的发布流程第 1 步。
|
|
108
|
+
8. `0.9.8` 的实际发布结果(2026-10-18):附注标签 `v0.9.8` 指向
|
|
109
|
+
`f4551b61602e5e60b11565ae31ccba3257d3dc9e`,该提交的 CI 与 CodeQL 均为绿;
|
|
110
|
+
`publish.yml` 在该标签上运行(运行号 `39048037503`,24 个步骤全部成功),
|
|
111
|
+
注册表 `latest` 为 `0.9.8`。逐项核验:来源证明类型为
|
|
112
|
+
`https://slsa.dev/provenance/v1`,其 `gitCommit` 正是该提交;压缩包 SHA1
|
|
113
|
+
`8288977711e04679d077b970f863e726631fc309` 与 SHA512 与注册表一致;注册表
|
|
114
|
+
条目数 68、解包 130,945,335 字节与发布前的打包核对一致,包内逐个文件与已验证
|
|
115
|
+
的工作树逐字节相同。随后在全新隔离 profile 中从 npm 安装
|
|
116
|
+
`dsh-deeppilot@0.9.8`,在真实 `0.2.1-rc.2` 宿主上完成组合与启动,插件自有端口
|
|
117
|
+
3098 的 `/phone/health` 返回 200 且 `serverVersion` 为 `0.9.8`;GitHub Release
|
|
118
|
+
在这一步通过后才公开并标记 Latest。
|
|
90
119
|
|
|
91
120
|
## 五、官方资料
|
|
92
121
|
|