@mmmbuto/nexuscrew 0.8.24 → 0.8.26
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/CHANGELOG.md +764 -0
- package/README.md +38 -7
- package/frontend/dist/assets/index-DkIPrmX6.js +91 -0
- package/frontend/dist/assets/index-DqG-mDnV.css +32 -0
- package/frontend/dist/index.html +2 -2
- package/frontend/dist/version.json +1 -1
- package/lib/cli/doctor.js +35 -0
- package/lib/fleet/catalogs/alibaba-token-plan-qwen3.8.json +51 -0
- package/lib/fleet/managed.js +106 -10
- package/lib/fleet/routes.js +4 -1
- package/lib/mcp/server.js +80 -19
- package/lib/mcp/tools.js +57 -8
- package/package.json +3 -2
- package/skills/alibaba-token-media/SKILL.md +133 -0
- package/skills/alibaba-token-media/agents/openai.yaml +4 -0
- package/skills/alibaba-token-media/references/api-contract.md +97 -0
- package/skills/alibaba-token-media/scripts/alibaba_token_media.py +550 -0
- package/frontend/dist/assets/index-77r8nzQf.css +0 -32
- package/frontend/dist/assets/index-CU9OOtL-.js +0 -91
package/README.md
CHANGED
|
@@ -140,10 +140,10 @@ The provider catalog is scoped to the selected CLI rather than to a machine-spec
|
|
|
140
140
|
|
|
141
141
|
| CLI | Built-in provider choices |
|
|
142
142
|
|---|---|
|
|
143
|
-
| Claude Code | Anthropic, OpenRouter, Kimi Code, Amazon Bedrock, Google Vertex AI, Microsoft Foundry, Ollama Cloud, local Ollama, Z.AI, custom Anthropic-compatible endpoint |
|
|
143
|
+
| Claude Code | Anthropic, Alibaba Token Plan Personal, OpenRouter, Kimi Code, Amazon Bedrock, Google Vertex AI, Microsoft Foundry, Ollama Cloud, local Ollama, Z.AI, custom Anthropic-compatible endpoint |
|
|
144
144
|
| Codex | OpenAI or ChatGPT login, OpenAI API, Ollama Cloud, local Ollama, LM Studio, custom OpenAI Responses endpoint |
|
|
145
|
-
| Codex-VL | OpenAI or ChatGPT login, OpenAI API, OpenRouter, Ollama Cloud, local Ollama, LM Studio, custom OpenAI Responses endpoint |
|
|
146
|
-
| Pi | Native default, Anthropic, OpenAI API, Codex OAuth, Gemini, GitHub Copilot, OpenRouter, Ollama, DeepSeek, Z.AI, custom provider |
|
|
145
|
+
| Codex-VL | OpenAI or ChatGPT login, OpenAI API, Alibaba Token Plan Personal, OpenRouter, Ollama Cloud, local Ollama, LM Studio, custom OpenAI Responses endpoint |
|
|
146
|
+
| Pi | Native default, Anthropic, OpenAI API, Alibaba Token Plan Personal, Codex OAuth, Gemini, GitHub Copilot, OpenRouter, Ollama, DeepSeek, Z.AI, custom provider |
|
|
147
147
|
|
|
148
148
|
Custom Codex-compatible endpoints use the real Responses wire API; NexusCrew does not silently
|
|
149
149
|
fall back to Chat Completions. Custom argv-based engines are also supported and are launched
|
|
@@ -160,6 +160,14 @@ uses `https://api.kimi.com/coding/`, and runs with an isolated Claude configurat
|
|
|
160
160
|
Anthropic account remains untouched. A Kimi Code membership key is not interchangeable with a
|
|
161
161
|
Moonshot pay-as-you-go API key.
|
|
162
162
|
|
|
163
|
+
Alibaba Token Plan Personal is a separate managed profile for Claude Code, Codex-VL and Pi. It
|
|
164
|
+
uses only `ALIBABA_CODE_API_KEY`, defaults to `qwen3.8-max-preview`, and has no OpenAI or
|
|
165
|
+
pay-as-you-go fallback. The npm package also includes the portable `alibaba-token-media` skill
|
|
166
|
+
for dry-run-first Wan image/edit and HappyHorse video workflows. Claude Code, Codex, Codex-VL
|
|
167
|
+
and Pi can invoke its dependency-free Python CLI directly; Pi is not assumed to support MCP
|
|
168
|
+
natively. The media skill requires Python 3. Media generation always requires explicit Credit
|
|
169
|
+
consent and never runs during installation, tests, or startup.
|
|
170
|
+
|
|
163
171
|
Permission handling is explicit per cell and engine:
|
|
164
172
|
|
|
165
173
|
- Claude engines can use standard permissions or `--dangerously-skip-permissions`.
|
|
@@ -323,11 +331,18 @@ It is intended for AI sessions running inside managed tmux cells.
|
|
|
323
331
|
| `nc_deck` | Discover owner-qualified decks containing the calling tmux session |
|
|
324
332
|
| `nc_cells` | List authorized active and inactive Fleet cells across visible nodes |
|
|
325
333
|
| `nc_send_cell` | Submit bounded text to one exact active cell returned by `nc_cells` |
|
|
334
|
+
| `nc_identity` | Read-only identity diagnostics; callable with no session and no token |
|
|
326
335
|
|
|
327
336
|
Cell delivery uses bracketed paste followed by a separate Enter. A `submitted` receipt confirms
|
|
328
337
|
delivery to the target TUI, not acceptance or completion by its model. There is no silent
|
|
329
338
|
offline queue.
|
|
330
339
|
|
|
340
|
+
`nc_identity` returns only non-sensitive data: the `source` the caller was resolved from
|
|
341
|
+
(`tmux`, `NEXUSCREW_MCP_SESSION`, or `missing`), boolean presence of the identity env vars,
|
|
342
|
+
a stable `code` (`OK`, `NEXUSCREW_MCP_IDENTITY_MISSING`, `NEXUSCREW_MCP_IDENTITY_INVALID`) and
|
|
343
|
+
a remediation hint. It never calls an HTTP API or reads the token, so it works even when the
|
|
344
|
+
identity is missing — use it to diagnose why the identity-gated tools fail closed.
|
|
345
|
+
|
|
331
346
|
Register the bridge in Claude Code:
|
|
332
347
|
|
|
333
348
|
```json
|
|
@@ -341,16 +356,32 @@ Register the bridge in Claude Code:
|
|
|
341
356
|
}
|
|
342
357
|
```
|
|
343
358
|
|
|
344
|
-
Or in Codex / Codex-VL
|
|
359
|
+
Or in Codex / Codex-VL (`env_vars` allowlists variable **names** only — no values are copied
|
|
360
|
+
into the CLI or config file):
|
|
345
361
|
|
|
346
362
|
```toml
|
|
347
363
|
[mcp_servers.nexuscrew]
|
|
348
364
|
command = "nexuscrew"
|
|
349
365
|
args = ["mcp"]
|
|
366
|
+
env_vars = ["NEXUSCREW_MCP_SESSION", "TMUX", "TMUX_PANE"]
|
|
367
|
+
```
|
|
368
|
+
|
|
369
|
+
The equivalent CLI form on a Codex-VL build that supports `--env-var` (allowlist by name, repeated):
|
|
370
|
+
|
|
371
|
+
```text
|
|
372
|
+
codex-vl mcp add nexuscrew \
|
|
373
|
+
--env-var NEXUSCREW_MCP_SESSION \
|
|
374
|
+
--env-var TMUX \
|
|
375
|
+
--env-var TMUX_PANE \
|
|
376
|
+
-- nexuscrew mcp
|
|
350
377
|
```
|
|
351
378
|
|
|
352
|
-
The caller is resolved from its tmux session
|
|
353
|
-
|
|
379
|
+
The caller is resolved, in order, from its tmux session (`tmux display-message -p '#S'`),
|
|
380
|
+
then from the `NEXUSCREW_MCP_SESSION` fallback, then not at all. Codex/Codex-VL launch MCP stdio
|
|
381
|
+
processes with a cleared environment, so those clients must explicitly allowlist the identity
|
|
382
|
+
env vars for the server to observe them; otherwise the identity-gated tools (`nc_ask`, `nc_send_file`,
|
|
383
|
+
`nc_deck`, `nc_send_cell`, `nc_inbox`) stay fail-closed with a stable
|
|
384
|
+
`NEXUSCREW_MCP_IDENTITY_*` code, while `nc_notify` degrades to an unknown sender.
|
|
354
385
|
|
|
355
386
|
## Configuration
|
|
356
387
|
|
|
@@ -403,7 +434,7 @@ See [CHANGELOG.md](CHANGELOG.md) for released changes.
|
|
|
403
434
|
|
|
404
435
|
## Status
|
|
405
436
|
|
|
406
|
-
The current stable release is **v0.8.
|
|
437
|
+
The current stable release is **v0.8.26** on npm and GitHub.
|
|
407
438
|
|
|
408
439
|
## License
|
|
409
440
|
|