@seemseam/ccb 7.4.2 → 7.4.4

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.
Files changed (4) hide show
  1. package/README.md +152 -116
  2. package/README_zh.md +148 -109
  3. package/VERSION +1 -1
  4. package/package.json +2 -2
package/README.md CHANGED
@@ -1,136 +1,63 @@
1
1
  <div align="center">
2
2
 
3
- # CCB - Visible, Controllable Multi-Agent CLI Workspace
3
+ # CCB
4
+
5
+ **Visible, controllable multi-agent CLI workspace for Codex, Claude, Gemini, OpenCode, and more.**
4
6
 
5
7
  <p>
6
- <img src="https://img.shields.io/badge/v7-multi--agent--workspace-0B7285?style=for-the-badge" alt="v7 multi-agent workspace">
7
- <img src="https://img.shields.io/badge/terminal-tmux-2F9E44?style=for-the-badge" alt="tmux">
8
- <img src="https://img.shields.io/badge/providers-Codex%20%7C%20Claude%20%7C%20Gemini%20%7C%20OpenCode%20%7C%20Antigravity-CF1322?style=for-the-badge" alt="providers">
8
+ <img src="https://img.shields.io/badge/version-7.4.4-orange.svg" alt="version">
9
+ <img src="https://img.shields.io/badge/platform-Linux%20%7C%20macOS%20%7C%20WSL-lightgrey.svg" alt="platform">
10
+ <img src="https://img.shields.io/badge/providers-Codex%20%7C%20Claude%20%7C%20Gemini%20%7C%20OpenCode%20%7C%20Antigravity-0B7285.svg" alt="providers">
9
11
  </p>
10
12
 
11
- [![Platform](https://img.shields.io/badge/platform-Linux%20%7C%20macOS%20%7C%20WSL-lightgrey.svg)]()
12
- [![Python](https://img.shields.io/badge/python-3.10%2B-blue.svg)]()
13
- [![Version](https://img.shields.io/badge/version-7.4.2-orange.svg)]()
14
- [![Release](https://img.shields.io/badge/install-release--first-orange.svg)]()
15
-
16
13
  **English** | [中文](README_zh.md)
17
14
 
18
- [Why Multi Agents](#why-multi-agents) · [Comparison](#which-multi-agent-approach-should-you-use) · [v7 UI](#v7-ui-tour) · [Quick Start](#quick-start) · [tmux Basics](#tmux-basics) · [Configure Agents](#configure-your-agent-team) · [Install](#install-and-update)
15
+ [Quick Start](#quick-start) · [v7 UI](#v7-ui-tour) · [Configure Agents](#configure-your-agent-team) · [User Guide](docs/manuals/user-guide/) · [Developer Guide](docs/manuals/developer-guide/)
16
+
17
+ <p align="center">
18
+ <img src="assets/readme_v7/ccb-hero-en.png" alt="CCB v7 visible multi-agent CLI workspace" width="960">
19
+ </p>
19
20
 
20
21
  </div>
21
22
 
22
23
  ---
23
24
 
24
- ## Why Multi Agents
25
-
26
- A single agent is enough for small tasks. Once work needs planning, parallel edits, review, testing, and handoff, multi agents help separate roles, context, models, and execution. CCB focuses on putting multiple real CLI agents into one visible terminal workspace.
27
-
28
- | Value | Plain meaning |
29
- | :--- | :--- |
30
- | Role separation | `main` plans, `worker` implements, `reviewer` checks risk. |
31
- | Parallel progress | One agent can edit while another reads docs, validates, or reviews. |
32
- | Model and context layering | Different agents can use different providers, models, APIs, worktrees, and memory. |
33
-
34
- <details>
35
- <summary><b>Why one agent starts to struggle</b></summary>
36
-
37
- - Mixed roles reduce context focus: one conversation tries to architect, edit, test, and review itself.
38
- - Complex task execution has a ceiling: long work needs split points, handoffs, checks, and rollback boundaries.
39
- - Cost pressure is higher: if every step needs the strongest model, even simple sub-tasks become expensive.
40
- - Tool and skill management becomes harder: a "does everything" agent also accumulates too much authority and instruction load.
41
- - Serial waiting is inefficient: when one agent is reading logs or running tests, other independent work cannot naturally continue.
42
-
43
- </details>
44
-
45
- ## Which Multi-Agent Approach Should You Use?
46
-
47
- Multi-agent systems are not one fixed shape. Use the short table first; expand the details only if you are comparing tradeoffs.
25
+ ## What You Get
48
26
 
49
- | Approach | One-line summary | Best fit |
27
+ | See the work | Mix providers | Keep control |
50
28
  | :--- | :--- | :--- |
51
- | [Claude Code native subagents](https://code.claude.com/docs/en/sub-agents) / [agent teams](https://code.claude.com/docs/en/agent-teams) | Native delegation inside Claude Code. | You mostly stay in Claude Code and want more coordination handled by a Claude lead. |
52
- | [Hive / OpenHive](https://github.com/aden-hive/hive) | Production-oriented multi-agent workflow harness. | You need state, recovery, observability, cost controls, and graph workflows. |
53
- | CCB | Visible, controllable local CLI-agent workspace with mixed providers. | You want Codex, Claude, Gemini, OpenCode, Antigravity, and other real CLIs in one project terminal. |
29
+ | Every agent is a real terminal pane, not a hidden background worker. | Run Codex, Claude, Gemini, OpenCode, Antigravity, and related CLIs together. | Start, attach, delegate, review, rebuild, update, and stop the project workspace explicitly. |
54
30
 
55
- <details>
56
- <summary><b>Details: model choice, control, context, and complex workflows</b></summary>
57
-
58
- | Question | Claude Code native | Hive / OpenHive | CCB |
59
- | :--- | :--- | :--- | :--- |
60
- | Different model vendors? | Can choose Claude models for teammates/subagents; overall path is still Claude Code. | LiteLLM route covers many hosted and local providers. | Choose Codex, Claude, Gemini, OpenCode, Droid, Antigravity, and per-agent model/key/url. |
61
- | Is the process visible? | In-process or split panes depending on mode. | Runtime observability and dashboard-style control. | Real tmux panes by default; users can click, type, copy, and inspect each CLI. |
62
- | Is topology controllable? | Natural-language teammate setup, with much coordination handled by the lead. | Goal-generated graph-like topology, harness oriented. | Config explicitly defines agents, windows, panes, worktrees, and sidebar behavior. |
63
- | Is context manageable? | Subagents/teammates have separate contexts; teams have task and message state. | Role memory, durable state, and recovery are core design points. | Each CLI keeps its provider session; shared project memory and per-agent memory are optional. |
64
- | Best landing zone | Fast delegation inside Claude Code. | Business automation, long-running workflows, production reliability. | Local development with visible cross-provider CLI agents. |
65
-
66
- CCB also supports complex workflows, but it is not an automatic DAG generator. You design complexity explicitly through `.ccb/ccb.config`, windows, role memory, worktrees, model/API settings, and ask/callback routes.
67
-
68
- </details>
69
-
70
- ## What Is CCB?
71
-
72
- CCB is a project-level agent CLI workspace. It uses tmux to manage multiple real CLI agents and unifies startup, restore, communication, configuration, windows, and runtime state for one project.
73
-
74
- - **Real CLI sessions, not fake panels**: every agent pane runs the actual provider CLI.
75
- - **Visible collaboration**: the sidebar shows windows, agents, status, and communication; users can switch panes by mouse.
76
- - **Mixed providers**: one project can run Codex, Claude, Gemini, OpenCode, Droid, and Antigravity (`agy`) together.
77
- - **Project config**: `.ccb/ccb.config` defines the team, layout, windows, worktrees, model, key, and url.
78
- - **Roles**: a new role packaging model that lets specialized agents carrying
79
- "heavy weapons" such as independent skills, memory, and tool dependencies
80
- instantly land in a target project as hot-loadable, removable agents, while
81
- leaving the main environment, user global config, and project runtime state
82
- unchanged.
83
- - **Recoverable runtime**: CCB supervises agent panes and supports attach, restore, and project-scoped cleanup.
84
- - **Explicit collaboration channel**: agents can delegate through `/ask`, `$ask`, callback, and silence routes.
85
-
86
- ## v7 UI Tour
87
-
88
- This screenshot is a real dark terminal session from the `ccb_test2` project. The labels explain the regions; you do not need to memorize every shortcut first.
89
-
90
- <p align="center">
91
- <img src="assets/readme_v7/ccb-test2-terminal-annotated-en.png" alt="CCB v7 terminal workspace region guide" width="960">
92
- </p>
93
-
94
- | Region | Purpose |
95
- | :--- | :--- |
96
- | Sidebar | Shows the current window, agent list, provider labels, selected agent, and status hints. |
97
- | Comms | Shows ask/callback communication and collaboration status. |
98
- | Agent pane | Each pane is a real CLI session, such as Codex or Claude. |
99
- | Current input target | The status bar and pane border show where your input goes. |
100
- | Status bar | Shows project name, current agent, CCB version, date, and mouse/keyboard hints. |
101
- | Window grouping | v7 `[windows]` can group agents into main, work, review, research, or other workflow windows. |
102
-
103
- The sidebar implementation uses ideas from [tmux-agent-sidebar](https://github.com/hiroppy/tmux-agent-sidebar). Thanks to that project.
31
+ Blank projects include `ccb_self`, CCB's built-in self-understanding expert for usage guidance, layout explanation, config design, runtime diagnostics, recovery, and workflow repair.
104
32
 
105
33
  ## Quick Start
106
34
 
107
35
  ### 1. Install or update
108
36
 
109
- New users can install CCB from npm:
37
+ New installs should use the npm package:
110
38
 
111
39
  ```bash
112
40
  npm install -g @seemseam/ccb
113
41
  ```
114
42
 
115
- The npm package downloads and verifies the matching GitHub Release package for
116
- your platform, then exposes `ccb`, `ask`, `autonew`, and `ctx-transfer`.
117
-
118
- You can also install from a release package directly. Download the matching package from [Releases](https://github.com/SeemSeam/claude_codex_bridge/releases), then install it:
43
+ After CCB is installed, use CCB's updater:
119
44
 
120
45
  ```bash
121
- tar -xzf ccb-*.tar.gz
122
- cd ccb-*
123
- ./install.sh install
46
+ ccb update
124
47
  ```
125
48
 
126
- If CCB is already installed:
49
+ <details>
50
+ <summary><b>GitHub release package and source install fallbacks</b></summary>
51
+
52
+ If npm is not available in your environment, download the matching package from [Releases](https://github.com/SeemSeam/claude_codex_bridge/releases):
127
53
 
128
54
  ```bash
129
- ccb update
55
+ tar -xzf ccb-*.tar.gz
56
+ cd ccb-*
57
+ ./install.sh install
130
58
  ```
131
59
 
132
- <details>
133
- <summary><b>Source install is for development or fallback use</b></summary>
60
+ Source install is for development or temporary fallback use:
134
61
 
135
62
  ```bash
136
63
  git clone https://github.com/SeemSeam/claude_codex_bridge.git
@@ -138,7 +65,7 @@ cd claude_codex_bridge
138
65
  ./install.sh install
139
66
  ```
140
67
 
141
- Source installs link global `ccb` / `ask` back to the checkout. Regular users should prefer a stable release install or update.
68
+ Source installs link global `ccb` / `ask` back to the checkout. Regular users should prefer the npm package.
142
69
 
143
70
  </details>
144
71
 
@@ -167,7 +94,7 @@ tips_height = "35%"
167
94
  comms_limit = 3
168
95
  ```
169
96
 
170
- If you are not sure how to group windows, how many workers you need, which agents should use worktrees, or which agents need separate models or API routes, ask `ccb_self` to design the config with its built-in `ccb-config` skill. Blank projects include `ccb_self`; existing custom configs can add it with `ccb roles add agentroles.ccb_self:codex`.
97
+ If you are not sure how to group windows, how many workers you need, which agents should use worktrees, or which agents need separate models or API routes, ask `ccb_self`. It is CCB's built-in self-agent: it understands CCB commands, config authority, roles, windows, reload behavior, and common recovery paths, and can design the config with its private `ccb-config` skill. Blank projects include `ccb_self`; existing custom configs can add it with `ccb roles add agentroles.ccb_self:codex`.
171
98
 
172
99
  Validate the config:
173
100
 
@@ -189,6 +116,84 @@ Type directly in an agent pane, or route work between agents:
189
116
  /ask reviewer review the latest parser changes and list blocking issues.
190
117
  ```
191
118
 
119
+ ## v7 UI Tour
120
+
121
+ The hero screenshot above is a real dark terminal session from the `ccb_test2` project. The labels explain the regions; you do not need to memorize every shortcut first.
122
+
123
+ | Region | Purpose |
124
+ | :--- | :--- |
125
+ | Sidebar | Shows the current window, agent list, provider labels, selected agent, and status hints. |
126
+ | Comms | Shows ask/callback communication and collaboration status. |
127
+ | Agent pane | Each pane is a real CLI session, such as Codex or Claude. |
128
+ | Current input target | The status bar and pane border show where your input goes. |
129
+ | Status bar | Shows project name, current agent, CCB version, date, and mouse/keyboard hints. |
130
+ | Window grouping | v7 `[windows]` can group agents into main, work, review, research, or other workflow windows. |
131
+
132
+ The sidebar implementation uses ideas from [tmux-agent-sidebar](https://github.com/hiroppy/tmux-agent-sidebar). Thanks to that project.
133
+
134
+ ## What Is CCB?
135
+
136
+ CCB is a project-level agent CLI workspace. It uses tmux to manage multiple real CLI agents and unifies startup, restore, communication, configuration, windows, and runtime state for one project.
137
+
138
+ - **Real CLI sessions, not fake panels**: every agent pane runs the actual provider CLI.
139
+ - **Visible collaboration**: the sidebar shows windows, agents, status, and communication; users can switch panes by mouse.
140
+ - **Mixed providers**: one project can run Codex, Claude, Gemini, OpenCode, Droid, and Antigravity (`agy`) together.
141
+ - **Project config**: `.ccb/ccb.config` defines the team, layout, windows, worktrees, model, key, and url.
142
+ - **Built-in CCB expert**: blank projects include `ccb_self`, a self-maintenance agent with deep CCB knowledge for usage guidance, config design, diagnostics, recovery, and workflow repair.
143
+ - **Roles**: a new role packaging model that lets specialized agents carrying
144
+ "heavy weapons" such as independent skills, memory, and tool dependencies
145
+ instantly land in a target project as hot-loadable, removable agents, while
146
+ leaving the main environment, user global config, and project runtime state
147
+ unchanged.
148
+ - **Recoverable runtime**: CCB supervises agent panes and supports attach, restore, and project-scoped cleanup.
149
+ - **Explicit collaboration channel**: agents can delegate through `/ask`, `$ask`, callback, and silence routes.
150
+
151
+ ## Why Multi Agents
152
+
153
+ A single agent is enough for small tasks. Once work needs planning, parallel edits, review, testing, and handoff, multi agents help separate roles, context, models, and execution. CCB focuses on putting multiple real CLI agents into one visible terminal workspace.
154
+
155
+ | Value | Plain meaning |
156
+ | :--- | :--- |
157
+ | Role separation | `main` plans, `worker` implements, `reviewer` checks risk. |
158
+ | Parallel progress | One agent can edit while another reads docs, validates, or reviews. |
159
+ | Model and context layering | Different agents can use different providers, models, APIs, worktrees, and memory. |
160
+
161
+ <details>
162
+ <summary><b>Why one agent starts to struggle</b></summary>
163
+
164
+ - Mixed roles reduce context focus: one conversation tries to architect, edit, test, and review itself.
165
+ - Complex task execution has a ceiling: long work needs split points, handoffs, checks, and rollback boundaries.
166
+ - Cost pressure is higher: if every step needs the strongest model, even simple sub-tasks become expensive.
167
+ - Tool and skill management becomes harder: a "does everything" agent also accumulates too much authority and instruction load.
168
+ - Serial waiting is inefficient: when one agent is reading logs or running tests, other independent work cannot naturally continue.
169
+
170
+ </details>
171
+
172
+ ## Which Multi-Agent Approach Should You Use?
173
+
174
+ Multi-agent systems are not one fixed shape. Use the short table first; expand the details only if you are comparing tradeoffs.
175
+
176
+ | Approach | One-line summary | Best fit |
177
+ | :--- | :--- | :--- |
178
+ | [Claude Code native subagents](https://code.claude.com/docs/en/sub-agents) / [agent teams](https://code.claude.com/docs/en/agent-teams) | Native delegation inside Claude Code. | You mostly stay in Claude Code and want more coordination handled by a Claude lead. |
179
+ | [Hive / OpenHive](https://github.com/aden-hive/hive) | Production-oriented multi-agent workflow harness. | You need state, recovery, observability, cost controls, and graph workflows. |
180
+ | CCB | Visible, controllable local CLI-agent workspace with mixed providers. | You want Codex, Claude, Gemini, OpenCode, Antigravity, and other real CLIs in one project terminal. |
181
+
182
+ <details>
183
+ <summary><b>Details: model choice, control, context, and complex workflows</b></summary>
184
+
185
+ | Question | Claude Code native | Hive / OpenHive | CCB |
186
+ | :--- | :--- | :--- | :--- |
187
+ | Different model vendors? | Can choose Claude models for teammates/subagents; overall path is still Claude Code. | LiteLLM route covers many hosted and local providers. | Choose Codex, Claude, Gemini, OpenCode, Droid, Antigravity, and per-agent model/key/url. |
188
+ | Is the process visible? | In-process or split panes depending on mode. | Runtime observability and dashboard-style control. | Real tmux panes by default; users can click, type, copy, and inspect each CLI. |
189
+ | Is topology controllable? | Natural-language teammate setup, with much coordination handled by the lead. | Goal-generated graph-like topology, harness oriented. | Config explicitly defines agents, windows, panes, worktrees, and sidebar behavior. |
190
+ | Is context manageable? | Subagents/teammates have separate contexts; teams have task and message state. | Role memory, durable state, and recovery are core design points. | Each CLI keeps its provider session; shared project memory and per-agent memory are optional. |
191
+ | Best landing zone | Fast delegation inside Claude Code. | Business automation, long-running workflows, production reliability. | Local development with visible cross-provider CLI agents. |
192
+
193
+ CCB also supports complex workflows, but it is not an automatic DAG generator. You design complexity explicitly through `.ccb/ccb.config`, windows, role memory, worktrees, model/API settings, and ask/callback routes.
194
+
195
+ </details>
196
+
192
197
  ## Daily Operation
193
198
 
194
199
  | Goal | Command |
@@ -301,12 +306,13 @@ bound role locks against the current installed roles; interactive starts ask
301
306
  whether to refresh stale project locks in place, and non-interactive starts
302
307
  print a warning only.
303
308
 
304
- `ccb_self` is strongly recommended for CCB projects because it owns CCB config
305
- maintenance, runtime diagnostics, guarded recovery, and single-agent restart
306
- assistance without taking over product work. Blank projects include it in the
307
- built-in default. Existing projects, and projects with user or project config
308
- that replace the built-in default, should add it explicitly where they want
309
- that maintenance agent:
309
+ `ccb_self` is strongly recommended for CCB projects because it is the built-in
310
+ CCB expert agent. It carries CCB-specific knowledge about project config,
311
+ command usage, role binding, reload boundaries, runtime diagnostics, guarded
312
+ recovery, workflow repair, and single-agent restart assistance without taking
313
+ over product work. Blank projects include it in the built-in default. Existing
314
+ projects, and projects with user or project config that replace the built-in
315
+ default, should add it explicitly where they want that maintenance agent:
310
316
 
311
317
  ```bash
312
318
  ccb roles add agentroles.ccb_self:codex
@@ -416,6 +422,8 @@ Do not commit real API keys to a public repository. `key` / `url` are agent-loca
416
422
 
417
423
  The full `ccb-config` skill belongs to the `agentroles.ccb_self` role. It is not a globally inherited skill for every agent. CCB installs or refreshes this Role Pack by default, and blank projects include `ccb_self` in the built-in default config. Existing projects, or projects with a user/project config that replaces the built-in default, should bind it where they want the maintenance assistant.
418
424
 
425
+ `ccb_self` is more than a config helper: it is designed as CCB's self-understanding agent. Use it when you need help using CCB, explaining the active layout, choosing an agent topology, migrating `.ccb/ccb.config`, diagnosing project runtime state, or repairing a CCB workflow.
426
+
419
427
  If you do not want to hand-write `.ccb/ccb.config`, ask `ccb_self` and describe your project goal, parallelism, window grouping, worktree isolation, provider/model/API preferences. `ccb_self` uses its built-in `ccb-config` skill to discuss the shape with you and propose a complete config.
420
428
 
421
429
  Example:
@@ -487,33 +495,29 @@ CCB does not require leaving your editor. A common setup is: editor for code, CC
487
495
 
488
496
  ### Requirements
489
497
 
498
+ - Node.js and npm for the recommended npm install path
490
499
  - Python 3.10+
491
500
  - `tmux`
492
501
  - At least one agent CLI you plan to use, such as Codex, Claude, Gemini, OpenCode, Droid, or Antigravity
493
502
  - Linux, macOS, or WSL
494
- - Node.js 18+ when installing through npm
495
503
 
496
504
  Current v7 / newer versions do not claim native Windows support. Native Windows support only applies to the v5 line. If you are on Windows and want current versions, use WSL and keep both `ccb` and agent CLIs inside WSL.
497
505
 
498
506
  ### npm first
499
507
 
508
+ For first install, prefer npm:
509
+
500
510
  ```bash
501
511
  npm install -g @seemseam/ccb
502
512
  ```
503
513
 
504
- The npm package is a lightweight installer wrapper around the official GitHub
505
- Release artifacts. It verifies `SHA256SUMS` before extracting the platform
506
- package.
507
-
508
- ### Release package fallback
509
-
510
- For first install, prefer a package from [GitHub Releases](https://github.com/SeemSeam/claude_codex_bridge/releases). For existing installs:
514
+ For later updates:
511
515
 
512
516
  ```bash
513
517
  ccb update
514
518
  ```
515
519
 
516
- Source checkout install is for development, fix validation, or temporary fallback when a release package is not available.
520
+ [GitHub Releases](https://github.com/SeemSeam/claude_codex_bridge/releases) remain available for environments where npm is unavailable. Source checkout install is for development, fix validation, or temporary fallback.
517
521
 
518
522
  ### Uninstall
519
523
 
@@ -581,6 +585,38 @@ v7 highlights:
581
585
  - Hardened tmux, Ghostty, release helper, Codex trust, and provider session restore paths.
582
586
 
583
587
  <details open>
588
+ <summary><b>v7.4.4</b> - Claude End-Turn And npm Release Surface</summary>
589
+
590
+ - Completes Claude pane-backed asks promptly when a primary assistant response
591
+ emits `stop_reason=end_turn` with an observed request anchor and non-empty
592
+ reply, avoiding the previous 900-second timeout path.
593
+ - Treats empty session-boundary terminal events with no prior assistant reply
594
+ as `incomplete/task_complete_empty_reply` with empty-provider diagnostics.
595
+ - Restores the `@seemseam/ccb` npm release surface with package metadata, CLI
596
+ runner wrappers, and tag-triggered Trusted Publishing after GitHub release
597
+ assets are available.
598
+ - Refreshes the v7 README homepage around canonical hero assets, npm-first
599
+ install, and clearer `ccb_self` guidance.
600
+
601
+ </details>
602
+
603
+ <details>
604
+ <summary><b>v7.4.3</b> - PR #225 Reliability Follow-Up</summary>
605
+
606
+ - Restores the Claude launcher contract: inline `--settings` now reflects the
607
+ materialized settings overlay without injecting provider env into settings
608
+ JSON.
609
+ - Fixes Claude WSL Windows-executable environment forwarding so path variables
610
+ use `/p` translation while `ANTHROPIC_*` API values pass through as raw env
611
+ names.
612
+ - Hardens Antigravity resume lookup for SQLite `bytes`, `str`, and
613
+ `memoryview` metadata and falls back to `--continue` if lookup fails.
614
+ - Adds regression tests for the Claude settings contract, WSL API env
615
+ forwarding, and AGY resume fallback behavior.
616
+
617
+ </details>
618
+
619
+ <details>
584
620
  <summary><b>v7.4.2</b> - Self-Supervision And Empty Reply Guards</summary>
585
621
 
586
622
  - Hardens CCB self-supervision with bounded provider-runtime snapshots,
package/README_zh.md CHANGED
@@ -1,133 +1,63 @@
1
1
  <div align="center">
2
2
 
3
- # CCB - 可见、可控的多 Agent CLI 工作台
3
+ # CCB
4
+
5
+ **可见、可控的多 Agent CLI 工作台，用一个项目级 tmux 会话同时管理真实 CLI。**
4
6
 
5
7
  <p>
6
- <img src="https://img.shields.io/badge/v7-multi--agent--workspace-0B7285?style=for-the-badge" alt="v7 multi-agent workspace">
7
- <img src="https://img.shields.io/badge/terminal-tmux-2F9E44?style=for-the-badge" alt="tmux">
8
- <img src="https://img.shields.io/badge/providers-Codex%20%7C%20Claude%20%7C%20Gemini%20%7C%20OpenCode%20%7C%20Antigravity-CF1322?style=for-the-badge" alt="providers">
8
+ <img src="https://img.shields.io/badge/version-7.4.4-orange.svg" alt="version">
9
+ <img src="https://img.shields.io/badge/platform-Linux%20%7C%20macOS%20%7C%20WSL-lightgrey.svg" alt="platform">
10
+ <img src="https://img.shields.io/badge/providers-Codex%20%7C%20Claude%20%7C%20Gemini%20%7C%20OpenCode%20%7C%20Antigravity-0B7285.svg" alt="providers">
9
11
  </p>
10
12
 
11
- [![Platform](https://img.shields.io/badge/platform-Linux%20%7C%20macOS%20%7C%20WSL-lightgrey.svg)]()
12
- [![Python](https://img.shields.io/badge/python-3.10%2B-blue.svg)]()
13
- [![Version](https://img.shields.io/badge/version-7.4.2-orange.svg)]()
14
- [![Release](https://img.shields.io/badge/install-release--first-orange.svg)]()
15
-
16
13
  **中文** | [English](README.md)
17
14
 
18
- [为什么需要多 agents](#为什么需要多-agents) · [方案对比](#多-agents-方案怎么选) · [v7 界面](#v7-界面速览) · [快速开始](#快速开始) · [tmux 常规操作](#tmux-常规操作) · [配置团队](#配置-agent-团队) · [安装更新](#安装和更新)
15
+ [快速开始](#快速开始) · [v7 界面](#v7-界面速览) · [配置团队](#配置-agent-团队) · [使用文档](docs/manuals/user-guide/) · [开发文档](docs/manuals/developer-guide/)
16
+
17
+ <p align="center">
18
+ <img src="assets/readme_v7/ccb-hero-zh.png" alt="CCB v7 可见多 Agent CLI 工作台" width="960">
19
+ </p>
19
20
 
20
21
  </div>
21
22
 
22
23
  ---
23
24
 
24
- ## 为什么需要多 agents
25
-
26
- 小任务用单 agent 就够了；一旦任务需要规划、并行实现、审查、测试和交接，多 agents 的价值就变成：把角色、上下文、模型和执行过程拆开管理。CCB 的重点是把多个真实 CLI agent 放进同一个可见终端工作台。
27
-
28
- | 价值 | 直观理解 |
29
- | :--- | :--- |
30
- | 角色分离 | `main` 负责任务拆分，`worker` 负责实现，`reviewer` 负责审查。 |
31
- | 并行推进 | 一个 agent 写代码时，另一个 agent 可以读文档、跑验证或审查风险。 |
32
- | 模型和上下文分层 | 不同 agent 可以用不同 provider、model、API、worktree 和记忆。 |
33
-
34
- <details>
35
- <summary><b>展开：单 agent 为什么会吃力？</b></summary>
36
-
37
- - 角色混杂会影响上下文集中度：同一个会话既要做架构，又要写代码，还要自我审查，容易在长任务里丢掉重点。
38
- - 可执行复杂度有上限：任务越长，越需要拆分、交接、核对和回滚点。
39
- - 成本压力更高：如果所有步骤都依赖一个最强模型，简单任务也会变贵。
40
- - 工具和 skill 集中会变难管理：什么都给一个 agent，等于把权限、说明和工具复杂度全部堆在一起。
41
- - 串行等待效率低：一个 agent 在跑测试、读日志或长时间思考时，其他可并行工作无法自然展开。
25
+ ## 三件事
42
26
 
43
- </details>
44
-
45
- ## 多 agents 方案怎么选
46
-
47
- 多 agents 不是一种固定形态。先用下面这张表判断大方向，细节可以展开看。
48
-
49
- | 方案 | 一句话 | 更适合你如果 |
27
+ | 看得见 | 混合 provider | 项目级控制 |
50
28
  | :--- | :--- | :--- |
51
- | [Claude Code 原生 subagents](https://code.claude.com/docs/en/sub-agents) / [agent teams](https://code.claude.com/docs/en/agent-teams) | Claude Code 内部的原生分工。 | 你主要留在 Claude Code，并接受更多协调由 Claude lead 处理。 |
52
- | [Hive / OpenHive](https://github.com/aden-hive/hive) | 面向生产工作流的多 agent harness。 | 你要状态、恢复、观测、成本控制和图式工作流。 |
53
- | CCB | 可见、可控、混合 provider 的本地 CLI agent 工作台。 | 你要把 Codex、Claude、Gemini、OpenCode、Antigravity 等真实 CLI 放到一个项目终端里操作。 |
54
-
55
- <details>
56
- <summary><b>展开：模型、可控性、上下文和复杂工作流怎么区别？</b></summary>
29
+ | 每个 agent 都是真实终端 pane，不是隐藏后台 worker。 | 同时运行 Codex、Claude、Gemini、OpenCode、Antigravity 等真实 CLI。 | 显式启动、attach、委派、审查、重建、更新和停止项目工作台。 |
57
30
 
58
- | 关键问题 | Claude Code 原生 | Hive / OpenHive | CCB |
59
- | :--- | :--- | :--- | :--- |
60
- | 能否使用不同家的模型 | 可给 teammate / subagent 指定 Claude 模型；整体仍在 Claude Code 体系内。 | 通过 LiteLLM 路线支持大量 hosted / local provider。 | 按 agent 选择 Codex、Claude、Gemini、OpenCode、Droid、Antigravity 等，并可设置独立 model / key / url。 |
61
- | 过程是否可见 | in-process 或 split panes，取决于模式和终端。 | 强调 runtime observability 和控制台视角。 | 默认就是 tmux 可见 pane，用户能直接点击、输入、复制、观察每个 CLI。 |
62
- | 拓扑是否可控 | 可自然语言指定队友，但运行时协调较多交给 lead。 | 由目标生成图式拓扑，偏 harness。 | 配置文件显式定义 agent、窗口、pane、worktree 和 sidebar。 |
63
- | 上下文是否可管理 | subagent / teammate 有独立上下文；team 有任务和消息状态。 | 角色记忆、状态持久化、恢复能力是核心卖点。 | 每个 CLI 保留自己的 provider 会话；项目共享记忆和 per-agent 记忆可选。 |
64
- | 最适合的落点 | Claude Code 内部的快速委派和团队模式。 | 业务流程自动化、长期运行和生产可靠性。 | 本地开发、代码协作、跨 provider CLI agent 可视化工作台。 |
65
-
66
- CCB 也支持复杂工作流，但它不是自动生成 DAG 的 harness；复杂度主要通过 `.ccb/ccb.config`、多 window、角色记忆、worktree、model/API 配置和 ask/callback 路由显式设计。
67
-
68
- </details>
69
-
70
- ## CCB 是什么
71
-
72
- CCB 是一个项目级 agent CLI 工作台。它用 tmux 管理多个真实 CLI agent，把启动、恢复、通信、配置、窗口和运行态聚合在一个项目里。
73
-
74
- - **真实 CLI，不是模拟面板**：每个 agent pane 都运行对应 provider 的真实 CLI。
75
- - **可见协作**：sidebar 展示窗口、agent 状态和通信区；用户可以用鼠标直接切 pane。
76
- - **混合 provider**：一个项目里可以同时跑 Codex、Claude、Gemini、OpenCode、Droid 和 Antigravity（`agy`）。
77
- - **项目级配置**：`.ccb/ccb.config` 决定团队、布局、窗口、worktree、model、key、url。
78
- - **Roles**：全新的角色封装概念；它让携带“重武器”（独立 skills、记忆和
79
- 工具依赖等）的专业角色瞬间“降临”到目标项目中，成为一个可以快速热加载和
80
- 卸载的独立 agent，同时保持主环境、用户全局配置和项目运行状态不发生改变。
81
- - **可恢复运行态**：CCB 后台守护 agent pane，支持 attach、恢复和项目级清理。
82
- - **显式协作通道**：agent 可以通过 `/ask`、`$ask`、callback 和 silence 进行委派与交接。
83
-
84
- ## v7 界面速览
85
-
86
- 下面截图来自 `ccb_test2` 项目的真实深色终端会话。图片上的标注只解释区域，不代表必须记住所有快捷键。
87
-
88
- <p align="center">
89
- <img src="assets/readme_v7/ccb-test2-terminal-annotated.png" alt="CCB v7 终端工作台区域说明" width="960">
90
- </p>
91
-
92
- | 区域 | 作用 |
93
- | :--- | :--- |
94
- | Sidebar | 显示当前 window、agent 列表、provider 标签、当前选中 agent 和状态提示。 |
95
- | Comms | 显示 ask/callback 等协作消息和通信状态。 |
96
- | Agent pane | 每个 pane 是一个真实 CLI 会话，例如 Codex 或 Claude。 |
97
- | 当前输入目标 | 状态栏和 pane 边框会提示当前输入会发给哪个 agent。 |
98
- | 状态栏 | 显示项目名、当前 agent、CCB 版本、日期和常用鼠标/键盘提示。 |
99
- | Window 分组 | v7 可用 `[windows]` 把 agent 按 main、work、review、research 等窗口分组。 |
100
-
101
- Sidebar 相关实现使用并借鉴了 [tmux-agent-sidebar](https://github.com/hiroppy/tmux-agent-sidebar) 的思路，在此表示感谢。
31
+ 空白项目默认包含 `ccb_self`，它是 CCB 内置自理解专家，可帮助使用 CCB、解释当前布局、设计配置、诊断运行态、恢复和修复工作流。
102
32
 
103
33
  ## 快速开始
104
34
 
105
35
  ### 1. 安装或更新
106
36
 
107
- 新用户可以先用 npm 安装：
37
+ 新安装推荐使用 npm 包：
108
38
 
109
39
  ```bash
110
40
  npm install -g @seemseam/ccb
111
41
  ```
112
42
 
113
- npm 包会下载并校验与你的平台匹配的 GitHub Release 包，然后提供 `ccb`、`ask`、`autonew` 和 `ctx-transfer`。
114
-
115
- 也可以直接使用 release 包。到 [Releases](https://github.com/SeemSeam/claude_codex_bridge/releases) 下载与你的平台匹配的包，解压后安装：
43
+ 安装完成后，后续更新直接使用 CCB 自带 updater：
116
44
 
117
45
  ```bash
118
- tar -xzf ccb-*.tar.gz
119
- cd ccb-*
120
- ./install.sh install
46
+ ccb update
121
47
  ```
122
48
 
123
- 如果你已经装过 CCB：
49
+ <details>
50
+ <summary><b>GitHub release 包和源码安装兜底</b></summary>
51
+
52
+ 如果当前环境不方便使用 npm，可以到 [Releases](https://github.com/SeemSeam/claude_codex_bridge/releases) 下载与你的平台匹配的包，解压后安装：
124
53
 
125
54
  ```bash
126
- ccb update
55
+ tar -xzf ccb-*.tar.gz
56
+ cd ccb-*
57
+ ./install.sh install
127
58
  ```
128
59
 
129
- <details>
130
- <summary><b>源码安装只建议开发或临时兜底使用</b></summary>
60
+ 源码安装只建议开发或临时兜底使用：
131
61
 
132
62
  ```bash
133
63
  git clone https://github.com/SeemSeam/claude_codex_bridge.git
@@ -135,7 +65,7 @@ cd claude_codex_bridge
135
65
  ./install.sh install
136
66
  ```
137
67
 
138
- 源码安装会让全局 `ccb` / `ask` 链接回当前 checkout。普通用户更建议安装或更新到稳定 release。
68
+ 源码安装会让全局 `ccb` / `ask` 链接回当前 checkout。普通用户更建议使用 npm 包。
139
69
 
140
70
  </details>
141
71
 
@@ -164,7 +94,7 @@ tips_height = "35%"
164
94
  comms_limit = 3
165
95
  ```
166
96
 
167
- 如果你不确定应该如何分组、要几个 worker、哪些 agent 用 worktree、哪些 agent 需要独立模型或 API，可以让 `ccb_self` 使用它内置的 `ccb-config` 和你讨论并生成配置方案。空白项目默认包含 `ccb_self`；已有自定义配置可用 `ccb roles add agentroles.ccb_self:codex` 添加。
97
+ 如果你不确定应该如何分组、要几个 worker、哪些 agent 用 worktree、哪些 agent 需要独立模型或 API，可以直接问 `ccb_self`。它是 CCB 内置的 self-agent，理解 CCB 命令、配置权威层、roles、windows、reload 边界和常见恢复路径，并能用私有 `ccb-config` skill 和你讨论后生成配置方案。空白项目默认包含 `ccb_self`；已有自定义配置可用 `ccb roles add agentroles.ccb_self:codex` 添加。
168
98
 
169
99
  验证配置：
170
100
 
@@ -186,6 +116,82 @@ ccb
186
116
  /ask reviewer review the latest parser changes and list blocking issues.
187
117
  ```
188
118
 
119
+ ## v7 界面速览
120
+
121
+ 上方 hero 图来自 `ccb_test2` 项目的真实深色终端会话。图片上的标注只解释区域，不代表必须记住所有快捷键。
122
+
123
+ | 区域 | 作用 |
124
+ | :--- | :--- |
125
+ | Sidebar | 显示当前 window、agent 列表、provider 标签、当前选中 agent 和状态提示。 |
126
+ | Comms | 显示 ask/callback 等协作消息和通信状态。 |
127
+ | Agent pane | 每个 pane 是一个真实 CLI 会话，例如 Codex 或 Claude。 |
128
+ | 当前输入目标 | 状态栏和 pane 边框会提示当前输入会发给哪个 agent。 |
129
+ | 状态栏 | 显示项目名、当前 agent、CCB 版本、日期和常用鼠标/键盘提示。 |
130
+ | Window 分组 | v7 可用 `[windows]` 把 agent 按 main、work、review、research 等窗口分组。 |
131
+
132
+ Sidebar 相关实现使用并借鉴了 [tmux-agent-sidebar](https://github.com/hiroppy/tmux-agent-sidebar) 的思路，在此表示感谢。
133
+
134
+ ## CCB 是什么
135
+
136
+ CCB 是一个项目级 agent CLI 工作台。它用 tmux 管理多个真实 CLI agent，把启动、恢复、通信、配置、窗口和运行态聚合在一个项目里。
137
+
138
+ - **真实 CLI，不是模拟面板**：每个 agent pane 都运行对应 provider 的真实 CLI。
139
+ - **可见协作**：sidebar 展示窗口、agent 状态和通信区；用户可以用鼠标直接切 pane。
140
+ - **混合 provider**：一个项目里可以同时跑 Codex、Claude、Gemini、OpenCode、Droid 和 Antigravity（`agy`）。
141
+ - **项目级配置**：`.ccb/ccb.config` 决定团队、布局、窗口、worktree、model、key、url。
142
+ - **内置 CCB 专家**：空白项目默认包含 `ccb_self`，它是具备 CCB 自理解能力的自维护 agent，可帮助使用 CCB、设计配置、诊断运行态、恢复工作流。
143
+ - **Roles**：全新的角色封装概念；它让携带“重武器”（独立 skills、记忆和
144
+ 工具依赖等）的专业角色瞬间“降临”到目标项目中，成为一个可以快速热加载和
145
+ 卸载的独立 agent，同时保持主环境、用户全局配置和项目运行状态不发生改变。
146
+ - **可恢复运行态**：CCB 后台守护 agent pane，支持 attach、恢复和项目级清理。
147
+ - **显式协作通道**：agent 可以通过 `/ask`、`$ask`、callback 和 silence 进行委派与交接。
148
+
149
+ ## 为什么需要多 agents
150
+
151
+ 小任务用单 agent 就够了；一旦任务需要规划、并行实现、审查、测试和交接，多 agents 的价值就变成：把角色、上下文、模型和执行过程拆开管理。CCB 的重点是把多个真实 CLI agent 放进同一个可见终端工作台。
152
+
153
+ | 价值 | 直观理解 |
154
+ | :--- | :--- |
155
+ | 角色分离 | `main` 负责任务拆分，`worker` 负责实现，`reviewer` 负责审查。 |
156
+ | 并行推进 | 一个 agent 写代码时，另一个 agent 可以读文档、跑验证或审查风险。 |
157
+ | 模型和上下文分层 | 不同 agent 可以用不同 provider、model、API、worktree 和记忆。 |
158
+
159
+ <details>
160
+ <summary><b>展开：单 agent 为什么会吃力？</b></summary>
161
+
162
+ - 角色混杂会影响上下文集中度：同一个会话既要做架构，又要写代码，还要自我审查，容易在长任务里丢掉重点。
163
+ - 可执行复杂度有上限：任务越长，越需要拆分、交接、核对和回滚点。
164
+ - 成本压力更高：如果所有步骤都依赖一个最强模型，简单任务也会变贵。
165
+ - 工具和 skill 集中会变难管理：什么都给一个 agent，等于把权限、说明和工具复杂度全部堆在一起。
166
+ - 串行等待效率低：一个 agent 在跑测试、读日志或长时间思考时，其他可并行工作无法自然展开。
167
+
168
+ </details>
169
+
170
+ ## 多 agents 方案怎么选
171
+
172
+ 多 agents 不是一种固定形态。先用下面这张表判断大方向，细节可以展开看。
173
+
174
+ | 方案 | 一句话 | 更适合你如果 |
175
+ | :--- | :--- | :--- |
176
+ | [Claude Code 原生 subagents](https://code.claude.com/docs/en/sub-agents) / [agent teams](https://code.claude.com/docs/en/agent-teams) | Claude Code 内部的原生分工。 | 你主要留在 Claude Code，并接受更多协调由 Claude lead 处理。 |
177
+ | [Hive / OpenHive](https://github.com/aden-hive/hive) | 面向生产工作流的多 agent harness。 | 你要状态、恢复、观测、成本控制和图式工作流。 |
178
+ | CCB | 可见、可控、混合 provider 的本地 CLI agent 工作台。 | 你要把 Codex、Claude、Gemini、OpenCode、Antigravity 等真实 CLI 放到一个项目终端里操作。 |
179
+
180
+ <details>
181
+ <summary><b>展开：模型、可控性、上下文和复杂工作流怎么区别？</b></summary>
182
+
183
+ | 关键问题 | Claude Code 原生 | Hive / OpenHive | CCB |
184
+ | :--- | :--- | :--- | :--- |
185
+ | 能否使用不同家的模型 | 可给 teammate / subagent 指定 Claude 模型；整体仍在 Claude Code 体系内。 | 通过 LiteLLM 路线支持大量 hosted / local provider。 | 按 agent 选择 Codex、Claude、Gemini、OpenCode、Droid、Antigravity 等，并可设置独立 model / key / url。 |
186
+ | 过程是否可见 | in-process 或 split panes，取决于模式和终端。 | 强调 runtime observability 和控制台视角。 | 默认就是 tmux 可见 pane，用户能直接点击、输入、复制、观察每个 CLI。 |
187
+ | 拓扑是否可控 | 可自然语言指定队友，但运行时协调较多交给 lead。 | 由目标生成图式拓扑，偏 harness。 | 配置文件显式定义 agent、窗口、pane、worktree 和 sidebar。 |
188
+ | 上下文是否可管理 | subagent / teammate 有独立上下文；team 有任务和消息状态。 | 角色记忆、状态持久化、恢复能力是核心卖点。 | 每个 CLI 保留自己的 provider 会话；项目共享记忆和 per-agent 记忆可选。 |
189
+ | 最适合的落点 | Claude Code 内部的快速委派和团队模式。 | 业务流程自动化、长期运行和生产可靠性。 | 本地开发、代码协作、跨 provider CLI agent 可视化工作台。 |
190
+
191
+ CCB 也支持复杂工作流，但它不是自动生成 DAG 的 harness；复杂度主要通过 `.ccb/ccb.config`、多 window、角色记忆、worktree、model/API 配置和 ask/callback 路由显式设计。
192
+
193
+ </details>
194
+
189
195
  ## 日常操作
190
196
 
191
197
  | 目标 | 命令 |
@@ -292,10 +298,11 @@ ccb roles update agentroles.archi
292
298
  role；如果锁已经落后，交互式启动会询问是否就地刷新项目锁，非交互启动只打印
293
299
  提醒。
294
300
 
295
- 强烈建议 CCB 项目保留 `ccb_self`，因为它负责 CCB 配置维护、运行诊断、受保护
296
- 恢复、工作链修复和单 agent 重启辅助，同时不接管业务任务。空白项目的内置默认配置
297
- 已经包含它；已有项目，或使用用户配置/项目配置替换内置默认的项目，需要该维护
298
- agent 时应显式把它作为 window leaf 加进去：
301
+ 强烈建议 CCB 项目保留 `ccb_self`，因为它是 CCB 内置专家 agent，携带 CCB
302
+ 项目配置、命令使用、role 绑定、reload 边界、运行诊断、受保护恢复、工作链修复和
303
+ 单 agent 重启辅助等专用知识，同时不接管业务任务。空白项目的内置默认配置已经
304
+ 包含它；已有项目，或使用用户配置/项目配置替换内置默认的项目，需要该维护 agent
305
+ 时应显式把它作为 window leaf 加进去：
299
306
 
300
307
  ```bash
301
308
  ccb roles add agentroles.ccb_self:codex
@@ -405,6 +412,8 @@ model = "sonnet"
405
412
 
406
413
  完整的 `ccb-config` skill 属于 `agentroles.ccb_self` 角色，不再作为所有 agent 都继承的公共 skill。CCB 默认会安装或刷新这个 Role Pack，空白项目的内置默认配置也会包含 `ccb_self`。已有项目，或使用用户配置/项目配置替换内置默认的项目，需要维护助手时应显式绑定它。
407
414
 
415
+ `ccb_self` 不只是配置助手，它被设计成 CCB 的自理解 agent。使用 CCB 时遇到布局解释、团队拓扑选择、`.ccb/ccb.config` 迁移、运行态诊断、恢复路径或工作流修复问题，都可以先问它。
416
+
408
417
  如果你不想手写 `.ccb/ccb.config`，可以直接询问 `ccb_self`，再用自然语言描述项目目标、并行程度、窗口分组、worktree 隔离、provider/model/API 偏好。`ccb_self` 会使用它内置的 `ccb-config` 和你讨论后提出完整配置方案。
409
418
 
410
419
  示例：
@@ -474,31 +483,29 @@ CCB 不要求你离开编辑器。常见方式是：编辑器负责写代码，C
474
483
 
475
484
  ### 环境要求
476
485
 
486
+ - 推荐 npm 安装路径需要 Node.js 和 npm
477
487
  - Python 3.10+
478
488
  - `tmux`
479
489
  - 至少一个你要使用的 agent CLI，例如 Codex、Claude、Gemini、OpenCode、Droid 或 Antigravity
480
490
  - Linux、macOS 或 WSL
481
- - 通过 npm 安装时需要 Node.js 18+
482
491
 
483
492
  当前 v7 / 新版本不声明原生 Windows 支持。原生 Windows 只支持到 v5 线；如果你在 Windows 上使用新版本，推荐使用 WSL，并让 `ccb` 与 agent CLI 都运行在 WSL 内。
484
493
 
485
494
  ### npm 优先
486
495
 
496
+ 首次安装推荐使用 npm：
497
+
487
498
  ```bash
488
499
  npm install -g @seemseam/ccb
489
500
  ```
490
501
 
491
- npm 包是官方 GitHub Release artifacts 的轻量安装 wrapper，会先校验 `SHA256SUMS`，再解压平台包。
492
-
493
- ### Release 包兜底
494
-
495
- 首次安装推荐使用 [GitHub Releases](https://github.com/SeemSeam/claude_codex_bridge/releases) 的 release 包；已安装用户推荐：
502
+ 后续更新直接使用：
496
503
 
497
504
  ```bash
498
505
  ccb update
499
506
  ```
500
507
 
501
- 源码 checkout 安装只适合开发、验证修复或 release 包暂不可用时临时使用。
508
+ [GitHub Releases](https://github.com/SeemSeam/claude_codex_bridge/releases) 仍作为不方便使用 npm 时的备选路径。源码 checkout 安装只适合开发、验证修复或临时兜底。
502
509
 
503
510
  ### 卸载
504
511
 
@@ -566,6 +573,38 @@ v7 线重点：
566
573
  - 加固 tmux、Ghostty、release helper、Codex trust 和 provider 会话恢复路径。
567
574
 
568
575
  <details open>
576
+ <summary><b>v7.4.4</b> - Claude end_turn 与 npm 发布面修复</summary>
577
+
578
+ - Claude pane-backed ask 在 primary assistant response 带
579
+ `stop_reason=end_turn`、已看到请求 anchor 且回复非空时，会立即产生
580
+ `TURN_BOUNDARY(reason=assistant_end_turn)` 并正常完成，不再等到 900 秒
581
+ reliability timeout。
582
+ - 空的 session-boundary terminal event 如果之前没有 assistant 回复证据，会
583
+ 终止为 `incomplete/task_complete_empty_reply`，并带
584
+ `empty_provider_reply` 诊断。
585
+ - 恢复 `@seemseam/ccb` npm 发布面：补回 package metadata、CLI runner
586
+ wrappers，以及等待 GitHub release assets 后再发布 npm 包的 Trusted
587
+ Publishing workflow。
588
+ - 刷新 v7 README 首页：使用 canonical hero assets，默认 npm-first 安装，并
589
+ 更明确说明 `ccb_self` 是 CCB 内置的使用、配置、诊断和恢复专家。
590
+
591
+ </details>
592
+
593
+ <details>
594
+ <summary><b>v7.4.3</b> - PR #225 可靠性跟进修复</summary>
595
+
596
+ - 恢复 Claude launcher contract：inline `--settings` 只反映 materialized
597
+ settings overlay，不再把 provider env 注入 settings JSON。
598
+ - 修复 Claude 在 WSL 调 Windows executable 时的环境透传：路径变量使用 `/p`
599
+ 转换，`ANTHROPIC_*` API 值作为 raw env 名透传。
600
+ - 加固 Antigravity resume lookup，兼容 SQLite 返回的 `bytes`、`str` 和
601
+ `memoryview` metadata，并在查找失败时降级为 `--continue`。
602
+ - 新增 Claude settings contract、WSL API env forwarding 和 AGY resume
603
+ fallback 的回归测试。
604
+
605
+ </details>
606
+
607
+ <details>
569
608
  <summary><b>v7.4.2</b> - Self-supervision 与空回复防护</summary>
570
609
 
571
610
  - 通过有界 provider-runtime snapshot、project-view 活动证据、suspicion
package/VERSION CHANGED
@@ -1 +1 @@
1
- 7.4.2
1
+ 7.4.4
package/package.json CHANGED
@@ -1,8 +1,8 @@
1
1
  {
2
2
  "name": "@seemseam/ccb",
3
- "version": "7.4.2",
3
+ "version": "7.4.4",
4
4
  "description": "Visible, controllable multi-agent CLI workspace for Codex, Claude, Gemini, OpenCode, and Antigravity.",
5
- "license": "MIT",
5
+ "license": "AGPL-3.0-only",
6
6
  "homepage": "https://github.com/SeemSeam/claude_codex_bridge#readme",
7
7
  "repository": {
8
8
  "type": "git",