severally-mcp 2.0.0 → 2.1.0
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/CONFIG.md +4 -2
- package/README.ja.md +28 -15
- package/README.md +29 -15
- package/docs/how-it-works.ja.md +7 -1
- package/docs/how-it-works.md +7 -1
- package/docs/install.ja.md +13 -1
- package/docs/install.md +16 -2
- package/package.json +2 -2
- package/plugins/severally/.claude-plugin/plugin.json +1 -1
- package/plugins/severally/.codex-plugin/plugin.json +4 -2
- package/plugins/severally/.mcp.json +2 -1
- package/plugins/severally/dist/severally-mcp.mjs +327 -94
- package/plugins/severally/skills/_template/SKILL.md.tmpl +5 -3
- package/plugins/severally/skills/_template/reference.md.tmpl +12 -6
- package/plugins/severally/skills/antigravity/severally/SKILL.md +5 -3
- package/plugins/severally/skills/antigravity/severally/reference.md +12 -6
- package/plugins/severally/skills/claude/severally/SKILL.md +5 -3
- package/plugins/severally/skills/claude/severally/reference.md +12 -6
- package/plugins/severally/skills/codex/severally/SKILL.md +5 -3
- package/plugins/severally/skills/codex/severally/reference.md +12 -6
- package/plugins/severally/skills/opencode/severally/SKILL.md +5 -3
- package/plugins/severally/skills/opencode/severally/reference.md +12 -6
- package/scripts/init-config.mjs +1 -1
- package/scripts/install.mjs +35 -5
- package/src/adapters/antigravity-sandbox.mjs +5 -1
- package/src/adapters/codex.mjs +17 -6
- package/src/brief.mjs +38 -16
- package/src/expose-paths.mjs +68 -5
- package/src/jobs.mjs +171 -45
- package/src/policy.mjs +5 -4
- package/src/redact.mjs +20 -1
- package/src/schema.mjs +5 -4
- package/src/server.mjs +44 -16
package/CONFIG.md
CHANGED
|
@@ -79,6 +79,8 @@ A name is lower-case letters, digits, `.` and `-`, at most 64 characters; `all`
|
|
|
79
79
|
| `enabled` | boolean | `true` | Turns this one consultant off, leaving the others on its CLI. |
|
|
80
80
|
| `note` | string | – | Why it is off, as for `clis`. |
|
|
81
81
|
|
|
82
|
+
### API keys and other model families, through OpenCode
|
|
83
|
+
|
|
82
84
|
OpenCode is the way to consult other model families. Any `provider/model` that `opencode models` lists can be
|
|
83
85
|
a consultant. Checked on 2026-10-17 with opencode 2.0.24: `opencode/deepseek-v4.1-flash-free` (DeepSeek) and
|
|
84
86
|
`opencode/nemotron-3-ultra-free#` (NVIDIA Nemotron, which has no effort variant) both answered as consultants,
|
|
@@ -198,8 +200,8 @@ server only from Claude Code and Antigravity.
|
|
|
198
200
|
| `max_rounds` | 5 (1 initial + 4 follow-ups) | 1–20 | `SEVERALLY_MAX_ROUNDS` | Rounds per consultation chain, the first one included. |
|
|
199
201
|
| `max_concurrent` | 9 | 1–9 | `SEVERALLY_MAX_CONCURRENT` | Consultations running at once, counted across every severally process that shares this home (one runs per client session); a request that would go over it is refused. A "consult everyone" fan-out takes one per consultant, and is refused if that is over the cap. |
|
|
200
202
|
| `max_jobs_retained` | 200 | 20–2000 | `SEVERALLY_MAX_JOBS_RETAINED` | Jobs the running server keeps in memory; older finished ones are dropped from memory, never from the history on disk. |
|
|
201
|
-
| `max_wait_ms` | 45000 | 0–600000 | `SEVERALLY_MAX_WAIT_MS` | Longest `wait_ms` a single `consult_get` may block for. Keep it under your client's tool-call timeout. A `consult_get` with no `wait_ms` waits for this long, or 45000 if this is higher. |
|
|
202
|
-
| `timeout_ms` |
|
|
203
|
+
| `max_wait_ms` | 45000 | 0–600000 | `SEVERALLY_MAX_WAIT_MS` | Longest `wait_ms` a single `consult_get` may block for. Keep it under your client's tool-call timeout. A `consult_get` with no `wait_ms` waits for this long, or 45000 if this is higher, unless the environment variable is set: it belongs to one client's registration, so it also sets that client's default wait (and takes precedence over this key). The plugin registrations for Claude Code and Codex (and the installer's `--manual` registration for Claude Code) set the environment variable to 600000: Claude Code waits on a tool call far longer, and the Codex plugin raises its tool-call timeout to 660 s. While it waits, `consult_get` sends a progress notification every 15 s when the client asks for them, and stops waiting if the client cancels the call. |
|
|
204
|
+
| `timeout_ms` | 1800000 | 1000–7200000 | `SEVERALLY_TIMEOUT_MS` | Wall-clock budget per consultation, for every consultant without its own `timeout_ms`. |
|
|
203
205
|
| `kill_grace_ms` | 5000 | 500–60000 | `SEVERALLY_KILL_GRACE_MS` | Time between asking a stopped consultant to exit and killing it; also how long the server waits for running consultants when it shuts down. |
|
|
204
206
|
| `max_auto_retries` | 2 | 0–3 | `SEVERALLY_MAX_AUTO_RETRIES` | How many times a consultation that failed with `rate_limited` or `cli_error` is run again automatically, inside its original time budget. `0` turns it off. |
|
|
205
207
|
| `retry_backoff_ms` | 5000 | 0–60000 | `SEVERALLY_RETRY_BACKOFF_MS` | Wait before the first automatic re-run; each later one waits three times longer. |
|
package/README.ja.md
CHANGED
|
@@ -6,16 +6,16 @@
|
|
|
6
6
|
</h1>
|
|
7
7
|
|
|
8
8
|
<p align="center">
|
|
9
|
-
<
|
|
9
|
+
<a href="https://github.com/coz-a/severally/actions/workflows/test.yml"><img src="https://github.com/coz-a/severally/actions/workflows/test.yml/badge.svg" alt="Test"></a>
|
|
10
10
|
</p>
|
|
11
11
|
|
|
12
12
|
<p align="center"><a href="README.md">English</a> | 日本語</p>
|
|
13
13
|
|
|
14
|
-
|
|
14
|
+
AI エージェントが、ほかのコーディングエージェントに独立した意見を求め、答えを相談相手ごとに受け取るためのプラグインです。
|
|
15
15
|
|
|
16
|
-
|
|
16
|
+
相談相手は Codex・Claude Code・Antigravity・OpenCode です。エージェントは、調査に行き詰まったときや、甲乙つけがたい選択に当たったときに、相談相手に意見を
|
|
17
17
|
求めます。返ってきた指摘は自分で確かめ、その結果をもとに作業を続けます。巻き戻しにくい変更のような
|
|
18
|
-
|
|
18
|
+
重要な判断では、承認を求める前に相談し、あなたが決めるべきことだけを訊きます。あなたから
|
|
19
19
|
「Codex に聞いて」と頼むこともできます。
|
|
20
20
|
|
|
21
21
|
severally は、独立した複数の相談相手に質問し、答えを 1 つにまとめずに相談相手ごとに返します。意見が割れたこと
|
|
@@ -26,6 +26,7 @@ severally は、独立した複数の相談相手に質問し、答えを 1 つ
|
|
|
26
26
|
- [使い方の例](#使い方の例)
|
|
27
27
|
- [仕組み](#仕組み)
|
|
28
28
|
- [インストール](#インストール)
|
|
29
|
+
- [必要なもの](#必要なもの)
|
|
29
30
|
- [更新](#更新)
|
|
30
31
|
- [アンインストール](#アンインストール)
|
|
31
32
|
- [使い方](#使い方)
|
|
@@ -118,13 +119,18 @@ flowchart TB
|
|
|
118
119
|
|
|
119
120
|
## インストール
|
|
120
121
|
|
|
121
|
-
|
|
122
|
-
|
|
122
|
+
### 必要なもの
|
|
123
|
+
|
|
124
|
+
- Node.js 20.10 以降
|
|
125
|
+
- インストールしてログインした CLI(4 つのうち少なくとも 1 つ)。1 つで足り、Claude Code だけでも、新しい
|
|
126
|
+
Claude Code のセッションに相談できます。
|
|
127
|
+
- macOS と Linux では、npm のグローバルの実行ディレクトリが `PATH` にあること。既定のインストールは、Codex が
|
|
128
|
+
起動する `severally-mcp` コマンドをそこに置きます。`--manual` なら要りません。
|
|
123
129
|
|
|
124
130
|
macOS、Linux、Windows(PowerShell またはコマンドプロンプト)のどれでも、次の 1 行で入れられます。
|
|
125
131
|
|
|
126
132
|
```bash
|
|
127
|
-
npx -y severally-mcp@2.
|
|
133
|
+
npx -y severally-mcp@2.1.0 install
|
|
128
134
|
```
|
|
129
135
|
|
|
130
136
|
インストーラは、`PATH` で見つかった CLI のそれぞれに severally を登録します。末尾に `--dry-run` を付けると、何も
|
|
@@ -143,7 +149,9 @@ npx -y severally-mcp@latest install
|
|
|
143
149
|
```
|
|
144
150
|
|
|
145
151
|
`@latest` を付けると、npx は毎回 npm に最新版を問い合わせます。手元に古い写しがあっても、それは使いません。
|
|
146
|
-
終わったら、クライアントを再起動してください。
|
|
152
|
+
終わったら、クライアントを再起動してください。Claude Code のプラグインマーケットプレイスから入れた場合は、
|
|
153
|
+
[docs/install.ja.md](docs/install.ja.md#更新) を見てください。版ごとの変更点は
|
|
154
|
+
[リリース](https://github.com/coz-a/severally/releases)にあります。
|
|
147
155
|
|
|
148
156
|
### アンインストール
|
|
149
157
|
|
|
@@ -172,12 +180,12 @@ npx -y severally-mcp@latest install
|
|
|
172
180
|
巻き戻しにくい変更の承認を求める前です。
|
|
173
181
|
|
|
174
182
|
相談が向いているのは、重要な設計判断、巻き戻しにくい選択、同じ不具合で 2 回以上失敗したときです。相談 1 回に
|
|
175
|
-
|
|
183
|
+
数分から 10 分ほどと実際の使用量がかかるので、小さな修正には向きません。
|
|
176
184
|
|
|
177
185
|
### 答えが返ったあと
|
|
178
186
|
|
|
179
|
-
|
|
180
|
-
|
|
187
|
+
相談を始めても、エージェントは答えを待たずに、答えに左右されない作業を進められます。答えはふつう 1〜10 分で
|
|
188
|
+
そろいます(いちばん時間がかかるのは Antigravity です)。
|
|
181
189
|
|
|
182
190
|
どの答えも同じ形で返ります。結論、根拠つきの指摘、決められなかったこと、判断が変わる条件、確かめ方です。
|
|
183
191
|
相談相手が失敗したときは、理由(`usage_limit`、`timeout` など)を付けて失敗として返します。「問題なし」として
|
|
@@ -195,7 +203,7 @@ npx -y severally-mcp@latest install
|
|
|
195
203
|
- `review`:計画を見せて、どこがまずいかを聞く(上の例はこれ)
|
|
196
204
|
- `debate`:計画と反対の主張を見せて、比べてもらう
|
|
197
205
|
|
|
198
|
-
最初の答えのあと、同じ相談相手に追加の質問を送れます。既定では 4 回まで、合わせて 5
|
|
206
|
+
最初の答えのあと、同じ相談相手に追加の質問を送れます。既定では 4 回まで、合わせて 5 往復です。全員に聞いた相談に追加の質問を送ると、全員にもう一度聞けます。それぞれの相談相手とのやりとりは別々に続きます。
|
|
199
207
|
|
|
200
208
|
### 履歴と記録
|
|
201
209
|
|
|
@@ -213,6 +221,10 @@ npx -y severally-mcp@latest install
|
|
|
213
221
|
すべて [CONFIG.md](CONFIG.md)(英語)にあり、[config.example.json](config.example.json) はそのまま写せる例です。
|
|
214
222
|
設定を変えたら、クライアントを再起動してください。
|
|
215
223
|
|
|
224
|
+
API キーで呼び出すモデル(OpenRouter や、OpenAI 互換のほかの API)やローカルのモデルに相談するには、OpenCode を入れて、
|
|
225
|
+
そのモデルを相談相手に足します。書き方は
|
|
226
|
+
[CONFIG.md](CONFIG.md#api-keys-and-other-model-families-through-opencode) にあります。
|
|
227
|
+
|
|
216
228
|
## 相談相手が読めるものとできること
|
|
217
229
|
|
|
218
230
|
相談では、ブリーフをその CLI のモデルに送ります。あなたが自分でその CLI を使うときと同じです。あなたの会話の履歴は
|
|
@@ -257,15 +269,16 @@ CLI ごとにどう守らせているかは [docs/how-it-works.ja.md](docs/how-i
|
|
|
257
269
|
|
|
258
270
|
`consult_start` の主な引数は次のとおりです。
|
|
259
271
|
|
|
260
|
-
- `context.expose_paths`:
|
|
272
|
+
- `context.expose_paths`: 相談相手に読み取り専用で写すファイルとディレクトリ(絶対パス)。git のリポジトリの中の
|
|
273
|
+
ディレクトリは、git が一覧するファイル(追跡中のものと、`.gitignore` で外していない未追跡のもの)だけを写す
|
|
261
274
|
- `context.expose_diff`: `{ repo, base? }`。変更中の差分をブリーフに入れ、差分が触るファイルを写す
|
|
262
275
|
- `network`: `"off"` にすると、その回のどの相談相手も Web の検索と閲覧ができない
|
|
263
|
-
- `followup_to`: 追加の質問のもとになる答えの `job_id`
|
|
276
|
+
- `followup_to`: 追加の質問のもとになる答えの `job_id`。全員に聞いた相談の `group_id` を渡すと、全員にもう一度聞く
|
|
264
277
|
|
|
265
278
|
## 貢献
|
|
266
279
|
|
|
267
280
|
質問と不具合の報告は [Issues](https://github.com/coz-a/severally/issues) へどうぞ。プルリクエストも歓迎します。
|
|
268
|
-
大きな変更は、先に Issue
|
|
281
|
+
大きな変更は、先に Issue で相談してください。参加する方は、[行動規範](CODE_OF_CONDUCT.md)に従ってください。
|
|
269
282
|
|
|
270
283
|
- `npm install` のあと `npm test` を実行します。テストは本物の代わりに用意した CLI を相手にオフラインで動き、
|
|
271
284
|
実際の API は呼びません。
|
package/README.md
CHANGED
|
@@ -6,17 +6,17 @@
|
|
|
6
6
|
</h1>
|
|
7
7
|
|
|
8
8
|
<p align="center">
|
|
9
|
-
<
|
|
9
|
+
<a href="https://github.com/coz-a/severally/actions/workflows/test.yml"><img src="https://github.com/coz-a/severally/actions/workflows/test.yml/badge.svg" alt="Test"></a>
|
|
10
10
|
</p>
|
|
11
11
|
|
|
12
12
|
<p align="center">English | <a href="README.ja.md">日本語</a></p>
|
|
13
13
|
|
|
14
|
-
A plugin that
|
|
14
|
+
A plugin that lets your AI agent ask other coding agents for independent opinions, returned separately.
|
|
15
15
|
|
|
16
|
-
When the agent gets stuck in an investigation, or faces two options that look equally good, it asks the
|
|
16
|
+
The consultants are Codex, Claude Code, Antigravity and OpenCode. When the agent gets stuck in an investigation, or faces two options that look equally good, it asks the
|
|
17
17
|
consultants for their opinions. It checks the findings that come back itself, and carries on from what it found.
|
|
18
|
-
For an important decision, such as a change that is hard to reverse, it
|
|
19
|
-
|
|
18
|
+
For an important decision, such as a change that is hard to reverse, it consults before asking you to approve it,
|
|
19
|
+
and asks you only about what is yours to decide. You can also ask yourself: "ask Codex".
|
|
20
20
|
|
|
21
21
|
The plugin puts the question to several independent consultants, and returns each one's answer on its own
|
|
22
22
|
instead of merging them into one, because a disagreement means something too. The agent checks why they disagree
|
|
@@ -27,6 +27,7 @@ before deciding how to take the answers.
|
|
|
27
27
|
- [Example](#example)
|
|
28
28
|
- [How it works](#how-it-works)
|
|
29
29
|
- [Install](#install)
|
|
30
|
+
- [Requirements](#requirements)
|
|
30
31
|
- [Updating](#updating)
|
|
31
32
|
- [Uninstall](#uninstall)
|
|
32
33
|
- [Usage](#usage)
|
|
@@ -121,13 +122,18 @@ What a consultant can read and do is under [What a consultant can see and do](#w
|
|
|
121
122
|
|
|
122
123
|
## Install
|
|
123
124
|
|
|
124
|
-
|
|
125
|
-
|
|
125
|
+
### Requirements
|
|
126
|
+
|
|
127
|
+
- Node.js 20.10 or newer
|
|
128
|
+
- At least one of the four CLIs, installed and logged in. One is enough: with only Claude Code, you can still ask a
|
|
129
|
+
fresh Claude Code session.
|
|
130
|
+
- On macOS and Linux, npm's global executable directory on `PATH`: the default install puts a `severally-mcp`
|
|
131
|
+
command there for Codex to launch. `--manual` does without it.
|
|
126
132
|
|
|
127
133
|
One line, on macOS, Linux and Windows (PowerShell or Command Prompt):
|
|
128
134
|
|
|
129
135
|
```bash
|
|
130
|
-
npx -y severally-mcp@2.
|
|
136
|
+
npx -y severally-mcp@2.1.0 install
|
|
131
137
|
```
|
|
132
138
|
|
|
133
139
|
The installer registers severally with each CLI it finds on `PATH`. Add `--dry-run` at the end to print what it
|
|
@@ -146,7 +152,9 @@ npx -y severally-mcp@latest install
|
|
|
146
152
|
```
|
|
147
153
|
|
|
148
154
|
With `@latest`, npx asks npm for the newest version every time, so an older copy it already has is not used.
|
|
149
|
-
Then restart the clients.
|
|
155
|
+
Then restart the clients. Installed from the Claude Code plugin marketplace: see
|
|
156
|
+
[docs/install.md](docs/install.md#updating). What changed in each version is in the
|
|
157
|
+
[releases](https://github.com/coz-a/severally/releases).
|
|
150
158
|
|
|
151
159
|
### Uninstall
|
|
152
160
|
|
|
@@ -175,12 +183,12 @@ Ask in your own words, in the chat with your agent:
|
|
|
175
183
|
good, and before asking you to approve a hard-to-reverse change.
|
|
176
184
|
|
|
177
185
|
It is worth it for important design decisions, hard-to-reverse choices, and two or more failed attempts at the
|
|
178
|
-
same bug. One consultation
|
|
186
|
+
same bug. One consultation takes minutes and real quota, so it is not for small fixes.
|
|
179
187
|
|
|
180
188
|
### When the answers come back
|
|
181
189
|
|
|
182
190
|
Starting a consultation does not make the agent wait: it can work on something that does not depend on the
|
|
183
|
-
answers, which usually arrive within
|
|
191
|
+
answers, which usually arrive within one to ten minutes (Antigravity is the slowest).
|
|
184
192
|
|
|
185
193
|
Every answer has the same shape: a bottom line, findings with their grounds, what the consultant could not
|
|
186
194
|
determine, conditions that would change its judgement, and how to check. A consultant that fails comes back as a
|
|
@@ -199,7 +207,7 @@ Codex").
|
|
|
199
207
|
- `debate`: shows the plan with the opposing claims, and asks the consultant to weigh them
|
|
200
208
|
|
|
201
209
|
After the first answer, the agent can send follow-ups to the same consultant: up to four by default, five rounds
|
|
202
|
-
in total.
|
|
210
|
+
in total. A follow-up to a fan-out asks every consultant in it again, each continuing its own exchange.
|
|
203
211
|
|
|
204
212
|
### History and records
|
|
205
213
|
|
|
@@ -219,6 +227,10 @@ becomes a consultant on its default model. To turn a CLI off or add consultants
|
|
|
219
227
|
[CONFIG.md](CONFIG.md), and [config.example.json](config.example.json) is a working example to copy from. Restart
|
|
220
228
|
the client after changing it.
|
|
221
229
|
|
|
230
|
+
To consult a model you reach with an API key (OpenRouter, or another OpenAI-compatible API) or a local model,
|
|
231
|
+
install OpenCode and add the model as a consultant: see
|
|
232
|
+
[CONFIG.md](CONFIG.md#api-keys-and-other-model-families-through-opencode).
|
|
233
|
+
|
|
222
234
|
## What a consultant can see and do
|
|
223
235
|
|
|
224
236
|
A consultation sends the brief to that CLI's model, the same way using the CLI yourself does. Your conversation
|
|
@@ -262,15 +274,17 @@ it exposes, go to each consultant's model. Every argument is described in the to
|
|
|
262
274
|
|
|
263
275
|
The arguments of `consult_start` that matter most:
|
|
264
276
|
|
|
265
|
-
- `context.expose_paths`: files and directories, by absolute path, copied read-only for the consultant
|
|
277
|
+
- `context.expose_paths`: files and directories, by absolute path, copied read-only for the consultant; a
|
|
278
|
+
directory in a git repository copies only the files git lists (nothing `.gitignore` excludes)
|
|
266
279
|
- `context.expose_diff`: `{ repo, base? }`, a change in progress: the diff goes in the brief and the files it touches are copied
|
|
267
280
|
- `network`: `"off"` takes web search and browsing away from every consultant in the round
|
|
268
|
-
- `followup_to`: the `job_id` of the answer a follow-up continues
|
|
281
|
+
- `followup_to`: the `job_id` of the answer a follow-up continues, or a fan-out's `group_id` to ask every consultant in it again
|
|
269
282
|
|
|
270
283
|
## Contributing
|
|
271
284
|
|
|
272
285
|
Questions and bug reports go to [Issues](https://github.com/coz-a/severally/issues). Pull requests are welcome; for
|
|
273
|
-
a large change, open an issue first so we can talk it over.
|
|
286
|
+
a large change, open an issue first so we can talk it over. Everyone taking part is expected to follow the
|
|
287
|
+
[Code of Conduct](CODE_OF_CONDUCT.md).
|
|
274
288
|
|
|
275
289
|
- `npm install`, then `npm test`. The tests run offline against stand-in CLIs and make no real API calls.
|
|
276
290
|
- After changing the source or the Skill template, run `npm run build`. The built server in
|
package/docs/how-it-works.ja.md
CHANGED
|
@@ -134,7 +134,9 @@ Antigravity にはツールを制限するフラグが無いので、Antigravity
|
|
|
134
134
|
質問がリポジトリのコードについてなら、主担当は最初の依頼で、関係するファイルを絶対パスで指定して渡します
|
|
135
135
|
(`expose_paths`)。テストの出力やログのようにファイルでないものはブリーフに入れます。ブリーフに貼った抜粋は、
|
|
136
136
|
ファイルの代わりではなく、ファイルに添えるものです。指定したファイルとディレクトリは、読み取り専用で相談相手の作業
|
|
137
|
-
ディレクトリ(`./workspace` の下)に写します。上限は、合わせて 20 項目、500 ファイル、5 MB
|
|
137
|
+
ディレクトリ(`./workspace` の下)に写します。上限は、合わせて 20 項目、500 ファイル、5 MB です。git の作業ツリーの中のディレクトリは、
|
|
138
|
+
git が管理するファイル(追跡しているものと、`.gitignore` で外していない未追跡のもの)だけを写すので、リポジトリのルートも
|
|
139
|
+
上限に収まります。git の外では `.git` と `node_modules` を飛ばします。シンボリックリンクは
|
|
138
140
|
たどらずに飛ばし、テキストファイルはブリーフと同じように認証情報を伏せ字にして、バイナリファイルはそのまま写します。
|
|
139
141
|
|
|
140
142
|
変更中のコードを見せるときは、代わりにリポジトリのパスを `expose_diff` で渡せます。差分(`HEAD`、または `base` に
|
|
@@ -151,3 +153,7 @@ Antigravity にはツールを制限するフラグが無いので、Antigravity
|
|
|
151
153
|
今の内容で写し直します。前の回から変わったかどうかを項目ごとに SHA-256 で示し、無くなったパスは飛ばして名前を挙げます。前の回の
|
|
152
154
|
`expose_diff` は、差分を取り直さず、差分が触ったファイルだけを写し直します。主担当がその呼び出しで指定しない限り、
|
|
153
155
|
何も写し直しません。
|
|
156
|
+
|
|
157
|
+
全員に聞いた相談への追加の質問(`followup_to` に `group_id` を渡す)は、全員に同じ新しい質問を送ります。それぞれの
|
|
158
|
+
相談相手は、自分のやりとりの最新の答えから続けるので、ブリーフにはその相談相手自身の前の回が入ります。まだ動いている
|
|
159
|
+
相談相手、答えの無い相談相手、回数を使い切った相談相手には送らず、`consult_start` の応答の `skipped` に理由とともに挙げます。
|
package/docs/how-it-works.md
CHANGED
|
@@ -136,7 +136,9 @@ When the question is about code in your repository, the lead hands the files ove
|
|
|
136
136
|
absolute paths (`expose_paths`); what is not a file there, such as a test's output or a log, goes in the brief,
|
|
137
137
|
and an excerpt in the brief goes alongside the files rather than instead of them. Those files and directories are
|
|
138
138
|
copied read-only into the consultant's working directory (under `./workspace`): at most 20 entries, 500 files
|
|
139
|
-
and 5 MB in total
|
|
139
|
+
and 5 MB in total. A directory inside a git work tree copies only the files git lists, tracked or untracked but
|
|
140
|
+
not ignored, so a repository root fits; outside git, `.git` and `node_modules` are skipped. Symlinks are skipped
|
|
141
|
+
rather than followed; text files are credential-masked exactly as the
|
|
140
142
|
brief is, binary files are copied unchanged. For a change under review the lead can pass `expose_diff` with the
|
|
141
143
|
repository's path instead: the diff (against `HEAD`, or a `base` revision) goes in the brief, cut to fit the
|
|
142
144
|
brief's budget (20,000 characters at most), and every file it touches, untracked ones included, is copied under
|
|
@@ -150,3 +152,7 @@ directory with no files. When it needs the earlier round's files, the lead passe
|
|
|
150
152
|
what that round exposed is copied again as it is now: each entry is marked changed or unchanged by its SHA-256, a
|
|
151
153
|
path that no longer exists is left out and named, and an earlier `expose_diff` brings back the files it touched
|
|
152
154
|
rather than a new diff. Nothing is copied again unless the lead asks for it in that call.
|
|
155
|
+
|
|
156
|
+
A follow-up to a fan-out (`followup_to` its `group_id`) asks every member the same new question, each continuing
|
|
157
|
+
its own chain from its latest answered round, so each brief carries that consultant's own earlier rounds. A member
|
|
158
|
+
that is still running, never answered or has used its rounds is not asked, and the `consult_start` response lists it under `skipped` with the reason.
|
package/docs/install.ja.md
CHANGED
|
@@ -106,7 +106,7 @@ Codex、Antigravity、OpenCode からも使うなら、クローンしてイン
|
|
|
106
106
|
|
|
107
107
|
## 外に何も送らないツールの許可の確認を省く
|
|
108
108
|
|
|
109
|
-
`consult_start` 以外の 5 つのツールは手元で完結します。`consult_get
|
|
109
|
+
`consult_start` 以外の 5 つのツールは手元で完結します。`consult_get`(相談が動いている間、繰り返し呼ぶ)、`consult_list`、
|
|
110
110
|
`consult_cancel`、`consult_record`、`consult_export` です。どのクライアントにも許可リストに
|
|
111
111
|
ツールを足すコマンドが無いので、インストーラは触りません。手で足してください。`consult_start` は確認を残します。
|
|
112
112
|
ほかのベンダーに材料を送るのはこの呼び出しだからです。
|
|
@@ -132,6 +132,18 @@ Codex、Antigravity、OpenCode からも使うなら、クローンしてイン
|
|
|
132
132
|
- **OpenCode** は、自分の `permission` の設定で変えていなければ、MCP ツールを確認なしで使います。
|
|
133
133
|
- **Antigravity**: MCP ツールを 1 つずつ許可する規則の書き方を確かめていないので、ここでは扱いません。
|
|
134
134
|
|
|
135
|
+
### 相談の開始の確認も省く
|
|
136
|
+
|
|
137
|
+
エージェントは自分から相談します(相談するかどうかを訊かないよう Skill が指示しています)。そのため `consult_start` の
|
|
138
|
+
確認を残すと、エージェントが始める相談のたびに 1 回あなたの確認を待ちます。この確認は、材料が手元を離れる前の
|
|
139
|
+
唯一の歯止めなので、既定では残します。省くときは、上のツールと同じ書き方で `consult_start` も足します。
|
|
140
|
+
Claude Code なら許可リストに `mcp__plugin_severally_severally__consult_start` を、Codex なら `config.toml` に
|
|
141
|
+
`[mcp_servers.severally.tools.consult_start]` の表を書きます(`--manual` で登録した場合。プラグイン方式では確かめていません)。
|
|
142
|
+
|
|
143
|
+
そうすると、どの相談でも次のものが、あなたに訊かずに各相談相手のモデルへ送られます。ブリーフ、渡したファイル、
|
|
144
|
+
相談相手が実行中に読んだものです([相談相手が読めるものとできること](../README.ja.md#相談相手が読めるものとできること))。
|
|
145
|
+
エージェントが触れるものがすべてそのベンダーに渡ってよい環境でだけ、省いてください。
|
|
146
|
+
|
|
135
147
|
## 更新
|
|
136
148
|
|
|
137
149
|
`npx -y severally-mcp@latest install` を実行します。
|
package/docs/install.md
CHANGED
|
@@ -107,8 +107,8 @@ Windows. Pass `--manual` to use manual mode on macOS/Linux too.
|
|
|
107
107
|
|
|
108
108
|
## Skipping the permission prompt for tools that send nothing out
|
|
109
109
|
|
|
110
|
-
Every tool except `consult_start` stays on your machine: `consult_get` (called
|
|
111
|
-
consultation), `consult_list`, `consult_cancel`, `consult_record` and `consult_export`. None of the clients has a command that adds tools to its allowlist, so the installer
|
|
110
|
+
Every tool except `consult_start` stays on your machine: `consult_get` (called repeatedly while a
|
|
111
|
+
consultation runs), `consult_list`, `consult_cancel`, `consult_record` and `consult_export`. None of the clients has a command that adds tools to its allowlist, so the installer
|
|
112
112
|
leaves this to you. Keep `consult_start` prompting: it is the point where material leaves for another vendor.
|
|
113
113
|
|
|
114
114
|
- **Claude Code**: the Skill already allows these five tools (its `allowed-tools`) for the turn it is invoked in, so
|
|
@@ -132,6 +132,20 @@ leaves this to you. Keep `consult_start` prompting: it is the point where materi
|
|
|
132
132
|
- **OpenCode** allows MCP tools without asking unless your own `permission` settings say otherwise.
|
|
133
133
|
- **Antigravity**: not covered here; the rule syntax for a single MCP tool has not been checked.
|
|
134
134
|
|
|
135
|
+
### Letting consultations start without the prompt too
|
|
136
|
+
|
|
137
|
+
The agent consults on its own (the Skill tells it not to ask whether to), so with `consult_start` prompting, every
|
|
138
|
+
consultation it starts waits for you once. That prompt is the only check before material leaves your machine, so it
|
|
139
|
+
stays by default. To drop it, add `consult_start` the same way as the tools above:
|
|
140
|
+
`mcp__plugin_severally_severally__consult_start` in Claude Code's allow list, or a
|
|
141
|
+
`[mcp_servers.severally.tools.consult_start]` table in Codex's `config.toml` (for a `--manual` registration; not
|
|
142
|
+
checked for the plugin install).
|
|
143
|
+
|
|
144
|
+
From then on, each consultation sends these to every consultant's model without asking you: the brief, the files
|
|
145
|
+
it exposes, and whatever the consultant reads while it runs (see
|
|
146
|
+
[What a consultant can see and do](../README.md#what-a-consultant-can-see-and-do)). Do it only where everything the
|
|
147
|
+
agent can reach may go to those vendors.
|
|
148
|
+
|
|
135
149
|
## Updating
|
|
136
150
|
|
|
137
151
|
Run `npx -y severally-mcp@latest install`, or pull the clone and run `npm install` and
|
package/package.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "severally-mcp",
|
|
3
|
-
"version": "2.
|
|
4
|
-
"description": "A plugin that
|
|
3
|
+
"version": "2.1.0",
|
|
4
|
+
"description": "A plugin that lets your AI agent ask other coding agents for independent opinions, returned separately.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"license": "Apache-2.0",
|
|
7
7
|
"author": "coz.a (https://github.com/coz-a)",
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
{
|
|
2
2
|
"$schema": "https://anthropic.com/claude-code/plugin.schema.json",
|
|
3
3
|
"name": "severally",
|
|
4
|
-
"version": "2.
|
|
4
|
+
"version": "2.1.0",
|
|
5
5
|
"description": "Ask Codex, Claude Code, Antigravity or OpenCode for an independent opinion, review or structured debate, in a sandboxed child session that can read files and search the web but cannot edit files, change the system, or start other agents.",
|
|
6
6
|
"author": {
|
|
7
7
|
"name": "coz.a",
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "severally",
|
|
3
|
-
"version": "2.
|
|
3
|
+
"version": "2.1.0",
|
|
4
4
|
"description": "Ask Codex, Claude Code, Antigravity or OpenCode for an independent opinion, review or structured debate, in a sandboxed child session that can read files and search the web but cannot edit files, change the system, or start other agents.",
|
|
5
5
|
"author": {
|
|
6
6
|
"name": "coz.a",
|
|
@@ -20,7 +20,9 @@
|
|
|
20
20
|
"mcpServers": {
|
|
21
21
|
"severally": {
|
|
22
22
|
"type": "stdio",
|
|
23
|
-
"command": "severally-mcp"
|
|
23
|
+
"command": "severally-mcp",
|
|
24
|
+
"env": { "SEVERALLY_MAX_WAIT_MS": "600000" },
|
|
25
|
+
"tool_timeout_sec": 660
|
|
24
26
|
}
|
|
25
27
|
},
|
|
26
28
|
"interface": {
|