dsh-packycode-api 1.4.7 → 1.4.9

package/CHANGELOG.md

@@ -1,5 +1,33 @@
 # Changelog
 
+## 1.4.9
+
+- **Auto review asks `jev-1.14.0`.** PackyCode retired the `jev-latest` alias: on 2026-11-13 `/api/pricing` lists `jev-1.14.0` and `jev-1.13.0` in `typesafe-officially` and no `jev-latest`, and a `/v1/systemone` request naming `jev-latest` is refused before any upstream (`400 model_not_found`, `当前令牌未覆盖供应商`), while `jev-1.14.0` answered in about 1.3 s. The reviewer, the classifier descriptor whose group coverage the settings read, the denial reasons and every failure sentence now use `jev-1.14.0`.
+- **A saved `jev-latest` choice keeps working.** Taking the alias out of the Config schema would make the plugin's Config fail to parse for anyone who had picked it. The schema keeps it as a hidden value, and the preference reader maps it to `jev-1.14.0` (`RETIRED_AUTO_REVIEW_MODELS` in `src/auto-review.js`). The card offers only the versioned reviewer and refuses the alias on write.
+- Validation, 2026-11-13:
+  - **Offline:** 570 tests (new: a saved `jev-latest` parses through the plugin Config, stays out of the offered choices, reads as `jev-1.14.0` in the preference, the status and the `llm/stream` listener, and is refused on write; Jev's request body, denial reason and coverage sentence name `jev-1.14.0`; the classifier descriptor, the client's coverage line and the settings card use it), both rebuilt bundles and the versioned tarball.
+  - **Live, `scripts/live-check.mjs` with the test key:** the group reading reached `jev-1.14.0`. The official review request was answered by `jev-1.14.0` with allow → allow (low risk) in 0.46 s and deny → deny (high risk) in 0.47 s, and by DeepSeek Flash with 1.28 s and 1.50 s. All fourteen catalog models answered through the route, and key rotation, the rejected-key classification, the RPC projections and the Claude CLI route were unchanged. A direct `/v1/systemone` request named `jev-1.14.0` answered; one naming `jev-latest` was refused.
+  - **Not checked:** the settings card in the desktop app (its rendering is covered by the built-bundle test, not by a look at the app). The id is pinned, so the next time PackyCode retires a Jev version the constant needs the same change.
+
+## 1.4.8
+
+- **Claude Fable 5.2 joins the catalog.** On 2026-11-11 `/api/pricing` lists `claude-fable-5-2` with the `anthropic` endpoint alone, in `cc`, `cc-sale`, `cc-expensive`, `claude-officially` and `aws-q`. It takes the route of the other Claude models: `anthropic-messages` at `https://cf.api.fan`, text and image input, 1,000,000 context and 128,000 output, first in the picker's Claude models. Anthropic's [Claude Fable 5.2 overview](https://platform.claude.com/docs/en/models/fable-5-2/overview) gives the 1M context window (its default and maximum) and 128K output, thinking that is always on and adaptive, `output_config.effort` `low` to `max` as the only depth control, and no `temperature`; so it offers `low` / `medium` / `high` / `xhigh` / `max` and cannot be turned off, like the 5.5 models. A key in `cc` runs it through the Claude Code CLI as the agent; the other Claude groups use HTTP, where it asks for progress-update thinking (`display: "updates"` with the `thinking-display-updates-2026-08-18` beta) as Opus and Sonnet 5.5 do. Its session titles run on `claude-haiku-5-5`. The settings page counts it in each key's Claude coverage, and `scripts/claude-check.mjs` and `scripts/claude-agent-check.mjs` include it; the six-hour `scripts/claude-stability.mjs` window still samples the three 5.5 models, which its request budget is sized for.
+- **MiniMax H3 Image joins the image engines.** On 2026-11-11 `/api/pricing` lists `MiniMax-H3-Image` and `MiniMax-H3-Image-Highspeed` in the `image` group, with the `image-generation` endpoint (`/v1/images/generations` and `/v1/images/edits`). Both are offered in the settings card and the `packycode_image_generate` tool. MiniMax's [text-to-image](https://platform.minimax.io/docs/api-reference/image-generation-h3-generations) and [image edit](https://platform.minimax.io/docs/api-reference/image-generation-h3-edits) references give these paths their own fields, so the request differs from GPT Image's (`requestFields` in `src/images.js`):
+  - There is no `quality` or `background`: the only quality is `auto`, and neither field is sent.
+  - `output_format: "png"` and `response_format: "b64_json"` are sent, because MiniMax answers with a JPEG link by default and the plugin stores the PNG it is given.
+  - `size: auto` goes with `area: "2K"`: the model picks the shape from the prompt or the references, at about 4 million pixels. MiniMax renders its 4K tier when neither a size nor an area is given, at four times 2K's pixels and price and past the size the plugin stores. An explicit size is sent without `area` (MiniMax refuses the two together) and keeps the existing limits (3840 per edge, 8,294,400 pixels), so the 4K tier is not offered.
+  - A transparent background and a mask, which MiniMax does not support, are refused before the request instead of being dropped. A tool call that names a MiniMax engine while the saved quality belongs to another engine runs at `auto`; an explicit quality it does not have is still refused.
+- **The image card shows each image key's groups and offers only the engines the key in use is served.** PackyCode lists the MiniMax engines in `image` and not in `image-sale`, and nothing in the card said which group an image key was in. The image key ring now gets the conversation keys' group reader (`createGroupReader` with `modelIds`, new RPC `images/groups`): the gateway's refusal of a model id it cannot route names the key's groups, `/v1/models` lists what the key reaches, and the public `/api/pricing` lists each engine's groups. None is billed, and the browser receives group names, their descriptions and per-engine booleans only. Each key row shows its groups. The model list leaves out engines the key in use is not served, and a note names them. A saved engine the key is not served stays in the list, marked, with a warning and the group it needs; the setting is never changed to another engine. With no reading, nothing is hidden. `/api/pricing` answers by where the request comes from (42 models and 13 groups to a direct connection from mainland China, with no GPT Image or Claude models, against 85 and 26 through a proxy), so an engine is hidden only when neither the groups nor the key's own model list reach it.
+- **The group refusal no longer invites a model switch.** The gateway's routing refusal (`error.code: "model_not_found"`, `分组 image-sale 下模型 MiniMax-H3-Image 无可用渠道`) read as "check model, quality, and image key permissions". In the desktop app, DeepSeek Flash answered a first wording of the new message ("choose another model or use an image key…") by retrying with another engine on its own. The message now says the image key's group does not serve the selected engine, not to retry with another model, and to have the user pick a served engine in the settings or add a key in a group that lists it. The gateway's text is still never shown.
+- `scripts/image-check.mjs` runs an engine without quality levels at `auto`, takes `--size` (`auto` or `WIDTHxHEIGHT`, `1024x1024` by default), and with `--all` reports an engine that fails and goes on to the next; the edit runs on the last engine that produced an image.
+- Validation, 2026-11-11:
+  - **Offline:** 569 tests (new for Fable: its catalog entry, protocol, efforts and wire dialect; its native efforts and client envelope over HTTP; `updates` thinking with the beta header, and an explicit `summarized` kept; direct and prepared titles on Haiku; an HTTP turn on a `cc-sale` key; pricing and coverage for its groups. New for MiniMax: the exact generation body for both engines at `auto` (with `area`) and at an explicit size (without it); a quality, a transparent background, a mask and an out-of-range size refused before any request; a multipart edit with two references, `area` and no `quality`, `background` or `mask`; the group refusal's message, without the gateway text and telling the model not to switch engines; the saved quality not inherited by a MiniMax call; a real PNG mask refused and the same edit sent without it; both engines as saved choices with `auto` alone, in the Config and in the built settings card. New for image key groups: the reader over an `image-sale`, an `image` and a rejected key with the image engines as its models; the `images/groups` RPC's projection, its failure message and its invalidation on a key added or removed; in the built card, the group tag on the key row, engines hidden with their note, the saved engine kept and marked with its warning, everything offered again for an `image` key or an unread one, and a rejected key), both rebuilt bundles and the versioned tarball. The suite also passed on Node 22.23.1.
+  - **Live, `scripts/claude-check.mjs --transport cli --model claude-fable-5-2` with the test key (`cc` among its groups), Claude Code 2.1.295:** 7 of 7 checks passed in 16 requests with no client-anomaly refusal: a read-only lookup, an approved edit, a denied delete, plan mode through DSH's plan review, a question, an image named `red`, and a title from `claude-haiku-5-5`. `scripts/live-check.mjs --model claude-fable-5-2` read the key's coverage as reaching it and answered the order-total scenario through the registered route.
+  - **Live, `scripts/image-check.mjs` with the image key:** while the key was in `image-sale`, both MiniMax engines were refused by the gateway before any upstream (HTTP 503 `model_not_found`) and no image was made. After the user moved it to `image`, 6 of 6 requests returned PNGs with no retry: `MiniMax-H3-Image` generated and edited at `1024x1024` in 11 and 16 s, `MiniMax-H3-Image-Highspeed` in 8 and 10 s, and `MiniMax-H3-Image` at `auto` returned 2048×2048 for both in 15 and 26 s. The generated teapot was blue and the edited one orange. A direct request with `size: auto`, `area: "1K"` and a wide landscape prompt returned 1568×672, so the gateway forwards `area` and the model picks the shape. `gpt-image-2.6-flare` at `low` still generated (34 s) and edited (49 s), reporting `quality: low`; both PNGs came back 1254×1254 for a `1024x1024` request, where the same request returned 1024×1024 on 2026-11-05.
+  - **Live, the image key's groups:** the probe was refused with `分组 image 下模型 dsh-packycode-group-probe 无可用渠道`, `/v1/models` listed seven image models, and the built reader took 2.7 s and marked all five engines served, with no key material in the reading.
+  - **Not checked:** the settings card in the desktop app (its rendering is covered by the built-bundle test, not by a look at the app), MiniMax's 4K tier, several references in one edit, and a request longer than the plugin's 5-minute timeout (MiniMax's own guidance is 600 s).
+  - **Not checked live:** Fable's HTTP route. No stored key is in `cc-sale`, `cc-expensive` or `claude-officially`, and a `cc` key over HTTP is refused. Anthropic enforces preserved thinking's history check on Claude Fable 5.2 for every account, where the 5.5 models enforce it only for accounts created since 2026-08-31, so a session whose earlier messages, system prompt or tools change between turns can be refused there with `The block is bound to a different conversation`.
+
 ## 1.4.7
 
 - **Images move to GPT Image 2.6.** PackyCode added `gpt-image-2.6-flare` and `gpt-image-2.6-sunburst`: on 2026-11-05 `/api/pricing` lists them in the `image` group beside the 2.5 pair, with the `image-generation` endpoint. They replace `gpt-image-2.5-flare` and `gpt-image-2.5-sunburst` in the settings card, the `packycode_image_generate` tool and `scripts/image-check.mjs`, whose default is now 2.6 Sunburst. OpenAI's [image generation guide](https://developers.openai.com/api/docs/guides/image-generation) gives "GPT Image 2.6 and 2.5 Sunburst and Flare" one quality list, `low` to `max`, and its [Sunburst](https://developers.openai.com/api/docs/models/gpt-image-2.6-sunburst) and [Flare](https://developers.openai.com/api/docs/models/gpt-image-2.6-flare) pages list `/v1/images/edits`, so both keep `auto`, `low`, `medium`, `high`, `xhigh` and `max`, and requests, sizes and edits are sent as before. `gpt-image-2` is unchanged.

package/README.en.md

@@ -32,7 +32,7 @@ The `PackyCode` model group ships with the plugin. No YAML edits and no separate
 | **Claude route by group** | The `cc` group only serves the official client, so Claude sessions on `cc` keys are carried out by this device's Claude Code itself, with its own system prompt and tools, while DSH shows every tool call and handles permissions, plan review and questions; `cc-sale`, `cc-expensive`, `claude-officially` and other groups use the plugin's HTTP route with DSH's own agent |
 | **Web search through PackyCode** | Optional: DSH's `web_search` searches through PackyCode, with a DeepSeek or GPT backend, for every request or for PackyCode models only |
 | **Image generation and editing** | Separate image keys, GPT Image 2 / 2.6 Flare / 2.6 Sunburst, model and quality settings, reference edits and original PNG storage; generation and editing verified against the gateway |
-| **Auto review on your model** | With DSH's built-in “Auto authorization review” switched on, the permission menu's Auto review can review each tool call with jev-latest or DeepSeek Flash; the rules and outcomes stay the official ones |
+| **Auto review on your model** | With DSH's built-in “Auto authorization review” switched on, the permission menu's Auto review can review each tool call with jev-1.14.0 or DeepSeek Flash; the rules and outcomes stay the official ones |
 | **Credentials stay on the machine** | Keys live in the DSH credential file. The browser receives labels, state, and balances only |
 
 <img width="774" height="455" alt="截屏2026-10-15 10 34 28" src="https://github.com/user-attachments/assets/c5eaa535-dd1e-4e2f-be9f-c181404d4dd6" />
@@ -48,10 +48,10 @@ You need **DeepSeek Harness** and a **PackyCode (PackyAPI) API key**. For DSH se
 Open the desktop application's **Plugins** page, click **Add plugin**, and enter the package name **with a version** as the **package name or address**:
 
 ```text
-dsh-packycode-api@1.4.7
+dsh-packycode-api@1.4.9
 ```
 
-Choose the **official** install source, then click **Install**. `1.4.7` was the latest release when this was written; the [npm page](https://www.npmjs.com/package/dsh-packycode-api) shows the current one.
+Choose the **official** install source, then click **Install**. `1.4.9` was the latest release when this was written; the [npm page](https://www.npmjs.com/package/dsh-packycode-api) shows the current one.
 
 > **Always include the version.** With a bare `dsh-packycode-api`, the install prompt shows the newest version on npm, but DSH's bundled pnpm by default installs only versions published at least 24 hours ago, so you can end up several releases behind with no sign of it in the prompt. With an exact version, pnpm installs that version and adds it to the profile's `minimumReleaseAgeExclude` itself. Afterwards, check **Current version** under **Settings → PackyCode → Plugin version**.
 
@@ -72,7 +72,7 @@ Choose a model from the `PackyCode` group. Of the thirteen models, Claude runs t
 
 The **Plugins** page shows the installed state and version, and is also where you enable, disable, or uninstall.
 
-**For non-desktop profiles only:** `dsh plugin --profile <name> add dsh-packycode-api@1.4.7`. Do not use it for `desktop` — the CLI refuses it (`profile "desktop" is managed exclusively by the Electron application`).
+**For non-desktop profiles only:** `dsh plugin --profile <name> add dsh-packycode-api@1.4.9`. Do not use it for `desktop` — the CLI refuses it (`profile "desktop" is managed exclusively by the Electron application`).
 
 For a local checkout, keep the `link:` prefix: `dsh plugin --profile <name> add link:/path/to/dsh-packycode-api`. **Do not pass a bare path or `file:`** — that install copies the whole checkout, `node_modules` included, into the profile, where it then shadows the packages the DSH installation itself provides and the plugin looks uninstalled; see the FAQ. If you already installed that way, uninstall first and reinstall as above.
 
@@ -163,9 +163,13 @@ Tool callers can set `model` and `quality` to `settings` to follow the saved pre
 
 Image 2 accepts `auto / low / medium / high`; both 2.6 models also accept `xhigh / max`. A saved 2.5 Flare or Sunburst setting is read as the 2.6 engine of the same name, with its quality kept. These are Images API `quality` values, separate from text reasoning effort. There is no automatic model substitution, quality downgrade or paid retry. Each call returns one PNG; original bytes are saved under `~/.dsh/packycode-api/images/` (following `DSH_HOME` when set), alongside the chat preview. References and optional PNG masks must belong to the current session; file paths cannot replace attachment references.
 
+`MiniMax-H3-Image` and `MiniMax-H3-Image-Highspeed` (faster, at a higher price) use the same two endpoints with different parameters. They have no quality levels: the setting offers only Auto and no `quality` is sent. They do not support transparent backgrounds or masks; a request with either is refused before it is sent, never rendered without it. MiniMax bills by output pixels and renders 4K when no size is given, so for `auto` the plugin adds `area: "2K"`: the model picks the shape from the description or the references, at about 4 million pixels. A size can also be given in the conversation (for example `2720x1536`), within the same limits as GPT Image (3840 per edge, 8.29 million pixels).
+
+The image card shows each image key's groups. The model list offers only the engines that the key in use is served, and a note names the ones hidden because its groups do not serve them. A saved engine the key is not served stays in the list, marked as not served, with a warning and the group it needs; it is never changed to another engine for you. PackyCode lists the two MiniMax engines in the `image` group only, for example, so a key in `image-sale` does not see them. When the groups could not be read, nothing is hidden.
+
 Click a generated image to open the in-app large preview. **Download original**, immediately left of Close, opens a save-location picker for the complete preserved PNG. Cancelling is harmless; a moved or deleted original reports a retryable error instead of downloading the chat thumbnail. Saved conversations also support this as long as their original file is still on this host.
 
-After adding the image key, `node scripts/image-check.mjs --all --edit` checks all three engines and one edit. Use `--model <id> --edit` for generation and editing on a specific engine, adding `--quality xhigh` or `--quality max` to check a quality. This consumes image quota, saves outputs under `.artifacts/`, and prints only a masked key tail plus the gateway-reported model/quality.
+After adding the image key, `node scripts/image-check.mjs --all --edit` checks every engine and one edit (an engine the key's group does not serve reports its own error and the others still run). Use `--model <id> --edit` for generation and editing on a specific engine, adding `--quality xhigh` or `--quality max` to check a quality. This consumes image quota, saves outputs under `.artifacts/`, and prints only a masked key tail plus the gateway-reported model/quality.
 
 Live checks on 2026-11-05 passed generation and editing at `low`, `xhigh` and `max` on both 2.6 Flare and 2.6 Sunburst, with matching quality values in the responses; `gpt-image-2` passed generation at `low`, and the gateway reported Sunburst's `auto` as `low`. All 15 requests succeeded without retries and returned 1024×1024 PNGs; visual inspection confirmed the requested color edits with composition preserved. Generate/edit times were about 14/15 seconds at xhigh and 24/26 seconds at max on Flare, and 51/64 and 96/97 seconds on Sunburst. These are built-client gateway checks; no desktop conversation was created to verify the full chat flow in this run. See the [protocol research and gateway results](docs/research/image-api.md) for scope and evidence.
 
@@ -181,10 +185,12 @@ By default DSH reviews with the conversation's own model. This plugin leaves the
 
 | Reviewer | Notes | Group needed |
 | --- | --- | --- |
-| jev-latest (default) | TypeSafe's decision classifier: about 0.4 s per review, billed by input only; a denial names only the risk level | `typesafe-officially` |
+| jev-1.14.0 (default) | TypeSafe's decision classifier: about 0.4 s per review, billed by input only; a denial names only the risk level | `typesafe-officially` |
 | DeepSeek Flash | About 2 s per review, and a denial says why | `deepseek-officially` / `deepseek-sale` |
 | Conversation model | Leave the review alone, as DSH does by default | — |
 
+A saved `jev-latest` still works and reviews with `jev-1.14.0`.
+
 Steps:
 
 1. Switch on “Auto authorization review” on DSH's Plugins page in the sidebar.
@@ -202,17 +208,17 @@ The card shows whether DSH's plugin is on, how many reviews this plugin answered
   - and it expects one `{"risk","decision","reason"?}` object back.
 - Every model request passes DSH's `llm/stream` waterfall. The plugin recognizes only this request there; conversation turns and everything else go on untouched.
 - **DeepSeek Flash** gets the same request, re-addressed to the PackyCode route; DSH still reads its answer.
-- **jev-latest** cannot write that object. The plugin passes the four sections and the official policy text to System One as the state and asks two questions: the risk, and whether you explicitly asked for this action.
+- **jev-1.14.0** cannot write that object. The plugin passes the four sections and the official policy text to System One as the state and asks two questions: the risk, and whether you explicitly asked for this action.
   - It allows a call when the probability of "low risk, or medium risk you asked for" is at least 0.6 and that of high risk is below 0.2.
-  - A denial says that jev-latest judged the call medium or high risk.
-  - jev-latest has a small context, so a long history is cut starting from the oldest entries; your instructions are kept first.
-- What is reviewed, the rules, and what an allow or a denial does all stay DSH's. DSH refuses a call whose review failed (for example, no key reaches jev-latest), and the plugin keeps that.
+  - A denial says that jev-1.14.0 judged the call medium or high risk.
+  - jev-1.14.0 has a small context, so a long history is cut starting from the oldest entries; your instructions are kept first.
+- What is reviewed, the rules, and what an allow or a denial does all stay DSH's. DSH refuses a call whose review failed (for example, no key reaches jev-1.14.0), and the plugin keeps that.
 - Measured on 37 scenarios with the official request format and the official parser: both reviewers matched all 35 scenarios that have a clear answer.
-  - jev-latest's median was 0.35 s, DeepSeek Flash's 2.3 s.
+  - jev-1.14.0's median was 0.35 s, DeepSeek Flash's 2.3 s.
 - If a later DSH changes the review request:
   - a request the plugin no longer recognizes goes on untouched, reviewed by the conversation model;
-  - a format jev-latest cannot read fails that review, which shows in the card's counts.
-- The description under Auto review in the menu says the same model reviews; that is DSH's text. With jev-latest or DeepSeek Flash chosen, that model reviews.
+  - a format jev-1.14.0 cannot read fails that review, which shows in the card's counts.
+- The description under Auto review in the menu says the same model reviews; that is DSH's text. With jev-1.14.0 or DeepSeek Flash chosen, that model reviews.
 
 </details>
 
@@ -230,13 +236,14 @@ The card shows whether DSH's plugin is on, how many reviews this plugin answered
 | `hy4` (Hy4, Tencent Hunyuan) | HTTP, Responses, text/image input; 1024K context (at most 960K input), 64K output | Can be off; `high` only | `hunyuan-officially` |
 | `gemini-4-argon` (Gemini 4 Argon) | HTTP, native Gemini streaming, text/image input; 1,048,576 context, 65,536 output | Cannot be off; `low` / `medium` / `high` | `gemini-slb` |
 | `gemini-3.8-flash` (Gemini 3.8 Flash) | HTTP, native Gemini streaming, text/image input; 1,048,576 input, 65,536 output | Cannot be off; `low` / `medium` / `high` | `gemini-slb` |
-| `claude-opus-5-5` | Anthropic Messages, images and native tools: this device's Claude Code CLI as the agent for `cc` keys, HTTP otherwise | Cannot be off; `low` / `medium` / `high` / `xhigh` / `max` | Claude-group coverage reported in Settings |
+| `claude-fable-5-2` (Claude Fable 5.2) | Anthropic Messages, images and native tools: this device's Claude Code CLI as the agent for `cc` keys, HTTP otherwise; 1M context, 128K output | Cannot be off; `low` / `medium` / `high` / `xhigh` / `max` | Claude-group coverage reported in Settings |
+| `claude-opus-5-5` | Same | Same | Same |
 | `claude-sonnet-5-5` | Same | Same | Same |
 | `claude-haiku-5-5` | Same; also generates Claude session titles | Same | Same |
 
 A model in the list does not mean the current token group allows it. When changing groups, do not drop the group that serves the model you still want.
 
-`jev-latest` (TypeSafe Jev) is not in the model picker: it is a decision model that answers classification questions and cannot hold a conversation, so it only serves [Auto review](#auto-review). The plugin calls `/v1/systemone` host-side; it needs a key in the `typesafe-officially` group, and each key's Jev entry under Accounts and keys shows whether it can.
+`jev-1.14.0` (TypeSafe Jev) is not in the model picker: it is a decision model that answers classification questions and cannot hold a conversation, so it only serves [Auto review](#auto-review). PackyCode lists it by version and no longer offers the `jev-latest` alias, so the plugin sends `jev-1.14.0` itself. It calls `/v1/systemone` host-side and needs a key in the `typesafe-officially` group; each key's Jev entry under Accounts and keys shows whether it can.
 
 Of the thirteen models in the current catalog, all but DeepSeek V4 Pro declare text and image input. DeepSeek Flash follows the [official vision guide](https://api-docs.deepseek.com/guides/vision/); images are read through DSH's attachment service and sent with the Responses request. DeepSeek V4 Pro takes text only, per DeepSeek's [models and pricing page](https://api-docs.deepseek.com/quick_start/pricing/): DSH does not let you attach an image to it, and an image already in the session (say, one switched over from Flash) is sent as a one-line "image omitted" note. PackyCode's gateway does not refuse an image sent to V4 Pro; it drops it.
 
@@ -268,16 +275,16 @@ DeepSeek Flash uses the gateway's Responses endpoint (`/api/pricing` lists `open
 - Claude turns on both routes retain the full DSH tool loop, permissions and plan mode, using host search and image tools. Claude titles use Haiku on the same route as the conversation; GPT-6 Astra and GPT-5.6 Sol titles use GPT-6.1 Sol, and DeepSeek V4 Pro titles use DeepSeek Flash.
 - The 1.2 `claudeCliMode` / `claudeCliCwd` settings do not return: Claude Code's permissions follow the session's DSH permission preset, and its working directory is the session's. The tool named `claude_code` is now the mirror tool that shows Claude Code's tool calls in cc sessions; only the plugin uses it, and it is not the 1.2 delegation tool. `claudeCliPath` applies again and names the CLI (empty means the usual install places, then PATH). 1.3.4's `claudeTransport` is removed, and a value left in a profile is ignored. Separately installed Claude Code is only modified when you click **Update Claude Code CLI** in the CLI card.
 - Codex Responses requests still need `prompt_cache_key`; the plugin supplies a missing session identity. Other Responses families omit it.
-- After building, `node scripts/claude-check.mjs` checks all three Claude models on order totals, support triage, replenishment, read-only inventory tool roundtrips, product color labels, Chinese titles and plan mode (read-only lookup → plan submitted with `exit_plan_mode` → approval), validating the business answers; `--transport cli` runs the cc route's checks instead (`scripts/claude-agent-check.mjs`): in a scratch directory Claude Code reads a file and computes, edits one (approved), deletes one (refused), plans and then implements after review, asks a question, reads an image and names a session, and the files and answers are checked, with the gateway's client-anomaly and anti-abuse refusals counted. `node scripts/claude-stability.mjs --start --hours 6 --key-label <label>` starts a deadline- and request-budget-limited window. Invoke the same script on a schedule to sample again; `--status` reads the record. Each round records its build and scenario version. A test-message policy refusal fails that check while the other scheduled checks continue; real credential or quota failures still stop the window. Checks never change stored credentials or start DSH.
+- After building, `node scripts/claude-check.mjs` checks all four Claude models on order totals, support triage, replenishment, read-only inventory tool roundtrips, product color labels, Chinese titles and plan mode (read-only lookup → plan submitted with `exit_plan_mode` → approval), validating the business answers; `--transport cli` runs the cc route's checks instead (`scripts/claude-agent-check.mjs`): in a scratch directory Claude Code reads a file and computes, edits one (approved), deletes one (refused), plans and then implements after review, asks a question, reads an image and names a session, and the files and answers are checked, with the gateway's client-anomaly and anti-abuse refusals counted. `node scripts/claude-stability.mjs --start --hours 6 --key-label <label>` starts a deadline- and request-budget-limited window. Invoke the same script on a schedule to sample again; `--status` reads the record. Each round records its build and scenario version. A test-message policy refusal fails that check while the other scheduled checks continue; real credential or quota failures still stop the window. Checks never change stored credentials or start DSH.
 - For continuous dialogs, use `node scripts/claude-conversation-check.mjs --model claude-haiku-5-5 --rounds 3 --key-label <label> --output .artifacts/claude-conversation.json`. One plugin adapter handles independent sessions; each group retains six user turns and full tool history, validates arithmetic, final JSON and unexpected characters, and stops on a failure without discarding evidence. Requests default to 65 seconds and 2048 output tokens. Diagnostic overrides use `--timeout-ms` (5000–300000) and `--max-tokens` (1024–128000). `--capture-wire` records the synthetic business requests, non-authentication headers and partial stream events for exact replay, without keys. Reports include scenario and adapter lifecycle, configured limits, response-header latency and total response time. These options do not alter the plugin's runtime settings.
 
 </details>
 
 ## Update and uninstall
 
-> **Upgrading from 1.3.2 or earlier:** install `dsh-packycode-api@1.4.7` on the Plugins page instead of clicking **Update plugin** in the old settings card. The old button bypasses pnpm's release-age policy: if the version it installs is less than 24 hours old, the profile refuses every other plugin install until that version is 24 hours old. After the install and a restart, the button is the fixed one, and later updates can use it.
+> **Upgrading from 1.3.2 or earlier:** install `dsh-packycode-api@1.4.9` on the Plugins page instead of clicking **Update plugin** in the old settings card. The old button bypasses pnpm's release-age policy: if the version it installs is less than 24 hours old, the profile refuses every other plugin install until that version is 24 hours old. After the install and a restart, the button is the fixed one, and later updates can use it.
 
-**Update:** open **Settings → PackyCode** and, when a newer version is listed, click **Update plugin**. It installs that exact version into the owning profile with DSH's bundled pnpm and reads `package.json` back before reporting success. The install keeps pnpm's release-age policy instead of overriding it: by default pnpm installs the version at once and records it in the profile's `minimumReleaseAgeExclude`; under a stricter pnpm policy the button says when the update can run. The newer version is read from npm, while pnpm downloads it from the profile's own install source: right after a release that source (a regional mirror in particular) may not have it yet, and the button asks you to update again in a few minutes; when the source cannot be reached, it asks you to check the network or proxy. A local `link:` checkout does not get the button. Then quit and reopen the desktop application. A page refresh does not reload the Host. Reinstalling from the **Plugins** page works too, again with the version in the package name (for example `dsh-packycode-api@1.4.7`); without it you may get a version that is merely the newest one published at least 24 hours ago.
+**Update:** open **Settings → PackyCode** and, when a newer version is listed, click **Update plugin**. It installs that exact version into the owning profile with DSH's bundled pnpm and reads `package.json` back before reporting success. The install keeps pnpm's release-age policy instead of overriding it: by default pnpm installs the version at once and records it in the profile's `minimumReleaseAgeExclude`; under a stricter pnpm policy the button says when the update can run. The newer version is read from npm, while pnpm downloads it from the profile's own install source: right after a release that source (a regional mirror in particular) may not have it yet, and the button asks you to update again in a few minutes; when the source cannot be reached, it asks you to check the network or proxy. A local `link:` checkout does not get the button. Then quit and reopen the desktop application. A page refresh does not reload the Host. Reinstalling from the **Plugins** page works too, again with the version in the package name (for example `dsh-packycode-api@1.4.9`); without it you may get a version that is merely the newest one published at least 24 hours ago.
 
 **Coming back from a local-path install, uninstall first.** An update is a `pnpm add`: it rewrites the files inside the package directory and leaves a `node_modules` that an earlier local-path install left there. That older tree keeps shadowing the packages the DSH installation provides, so the plugin row fails to load with no visible reason. **Uninstall** runs `pnpm remove`, which deletes the package directory outright — only a reinstall after that is clean. `node scripts/install-check.mjs` reports whether a profile is in that state.
 
@@ -293,7 +300,7 @@ ls ~/.dsh/profiles/desktop/node_modules/dsh-packycode-api/node_modules
 
 Output means a leftover tree: a registry install ships no `node_modules`, so this one usually comes from an earlier local-path install (`file:`, or a bare path instead of `link:`). It sits inside the profile directory, and DSH resolves a plugin's imports from inside the profile first, so the plugin gets those old copies — for example `@deepseek-ai/schemastery` 3.18.2, which has no `Schema.volatile()`, so the module throws while it is imported. When an import fails the Loader creates no fiber for that row and no client bundle is composed either, so the model route, the RPC channel, and the settings panel all stay absent.
 
-Recovery: **uninstall** the plugin on the **Plugins** page, **install** it again with the version in the package name (for example `dsh-packycode-api@1.4.7`), then quit and reopen the desktop application. Uninstall runs `pnpm remove` and deletes the whole package directory; reinstalling or clicking **Update** only rewrites files and keeps the leftover tree. Deleting that `node_modules` by hand and restarting works too.
+Recovery: **uninstall** the plugin on the **Plugins** page, **install** it again with the version in the package name (for example `dsh-packycode-api@1.4.9`), then quit and reopen the desktop application. Uninstall runs `pnpm remove` and deletes the whole package directory; reinstalling or clicking **Update** only rewrites files and keeps the leftover tree. Deleting that `node_modules` by hand and restarting works too.
 
 **Right after a first install, every card in Settings shows `transport failure for /packycode-api/…: HTTP 405`?** The installed version is 1.1.6 or earlier, usually because only the bare package name was entered: the install prompt shows the newest version, but pnpm installs the newest one published at least 24 hours ago (see **Install the plugin**). Those older versions can register their settings channel only after DSH restarts and rebuilds its `connection` service. So right after a first install `/packycode-api` does not exist yet, requests fall through to the static page, and it answers 405; the model route and the settings panel itself load normally. From 1.1.7 on, a first install needs no restart. Check which version was installed:
 

package/README.md

@@ -32,7 +32,7 @@
 | **Claude 按分组选通道** | `cc` 分组只接受官方客户端，这些 Key 的 Claude 会话由本机 Claude Code 自己完成：它用自己的系统提示词和工具做事，DSH 显示每次工具调用，并负责授权、计划审阅和提问；`cc-sale`、`cc-expensive`、`claude-officially` 等分组走插件内 HTTP，由 DSH 自己的 agent 执行 |
 | **网页搜索可走 PackyCode** | 可选：DSH 的 `web_search` 改用 PackyCode 搜索，后端在 DeepSeek 与 GPT 之间切换；可只对 PackyCode 模型生效 |
 | **图片生成与编辑** | 独立图片 Key，支持 GPT Image 2 / 2.6 Flare / 2.6 Sunburst、型号与质量设置、参考图编辑和 PNG 原图保存；已通过网关生图与编辑实测 |
-| **自动审查可换模型** | 开启 DSH 自带的「自动授权审查」后，权限菜单里的「自动审查」可改用 jev-latest 或 DeepSeek Flash 审查每次工具调用，审查规则和处理方式保持官方原样 |
+| **自动审查可换模型** | 开启 DSH 自带的「自动授权审查」后，权限菜单里的「自动审查」可改用 jev-1.14.0 或 DeepSeek Flash 审查每次工具调用，审查规则和处理方式保持官方原样 |
 | **凭据不离开本机** | Key 只存在 DSH 凭据文件里。浏览器只能看到备注、状态和余额，拿不到 Key |
 
 <img width="774" height="455" alt="截屏2026-10-15 10 34 28" src="https://github.com/user-attachments/assets/060c4192-00ea-40e6-9b85-bbec605b5c31" />
@@ -48,10 +48,10 @@
 打开桌面应用的**插件**页，点击**添加插件**，在**包名或地址**中填入**带版本号**的包名：
 
 ```text
-dsh-packycode-api@1.4.7
+dsh-packycode-api@1.4.9
 ```
 
-在**安装源**中选择**官方源**，然后点击**安装**。`1.4.7` 是写这份说明时的最新版，当前最新版本号以 [npm 页面](https://www.npmjs.com/package/dsh-packycode-api) 为准。
+在**安装源**中选择**官方源**，然后点击**安装**。`1.4.9` 是写这份说明时的最新版，当前最新版本号以 [npm 页面](https://www.npmjs.com/package/dsh-packycode-api) 为准。
 
 > **一定要带版本号。** 只填 `dsh-packycode-api` 时，安装前的提示显示的是 npm 上的最新版，但 DSH 自带的 pnpm 默认只安装发布满 24 小时的版本，实际装上的可能是好几个版本之前的旧版，而安装提示里看不出这一点。填写精确版本时，pnpm 会直接安装这个版本，并自动把它加入 profile 的 `minimumReleaseAgeExclude`。装好后可以在 **设置 → PackyCode → 插件版本** 核对「当前版本」。
 
@@ -72,7 +72,7 @@ dsh-packycode-api@1.4.7
 
 桌面应用的**插件**页可以看到已安装状态与版本，也可以在那里启用、停用或卸载。
 
-**仅用于非 desktop profile：** `dsh plugin --profile <name> add dsh-packycode-api@1.4.7`。不要对 `desktop` 使用它 —— CLI 会直接拒绝（`profile "desktop" is managed exclusively by the Electron application`）。
+**仅用于非 desktop profile：** `dsh plugin --profile <name> add dsh-packycode-api@1.4.9`。不要对 `desktop` 使用它 —— CLI 会直接拒绝（`profile "desktop" is managed exclusively by the Electron application`）。
 
 本地开发要加 `link:` 前缀：`dsh plugin --profile <name> add link:/path/to/dsh-packycode-api`。**不要直接写路径（或 `file:`）** —— 那样的安装会把整个仓库目录（连同 `node_modules`）复制进 profile，之后它会在 profile 内部挡住 DSH 自带的包，插件看起来像没装，见「常见问题」。已经那样装过的话，先卸载再按上面的方式重装。
 
@@ -90,7 +90,7 @@ dsh-packycode-api@1.4.7
 
 **看余额。** 「账户与 Key」里每个 Key 显示它自己的剩余、已用与额度（额度 = 剩余 + 已用），不限额的 Key 显示「不限额度」；多个 Key 不相加成合计。API Key 只能读到自己的额度，读不到 PackyCode 账号的总余额，账号余额请到 PackyCode 控制台查看。读失败的 Key 会在它那一行注明原因；某个 Key 这次读失败但之前读到过，会继续显示上次的余额并注明读取时间（输入框里显示为 `≈US$…`），重启 DSH 后才清空。输入框右侧只在选中 PackyCode 模型时显示当前 Key 的剩余额度。「刷新」会跳过约 60 秒的缓存。默认模型接口是 `https://cf.api.fan`，余额接口是 `https://www.packyapi.ai`，可在「高级：接口地址」里改。旧默认值 `https://www.packyapi.com` 在国内网络下 DNS 被污染、直连会 `fetch failed`，即使设置里保存的是它，插件也会先试 `.ai`。
 
-**用 Claude。** 直接选择三个 Claude 模型之一。Key 在 `cc` 分组时，这个会话由本机的官方 Claude Code CLI 完成，需要先在 **设置 → PackyCode → Claude Code CLI** 一键安装（或自行 `npm install -g @anthropic-ai/claude-code --allow-scripts=@anthropic-ai/claude-code`），版本过低或有新版本时在同一处点「更新 Claude Code CLI」；`cc-sale`、`cc-expensive`、`claude-officially` 等分组直接走插件内 HTTP，不需要 CLI。「账户与 Key」里每个能用 Claude 的 Key 会标出「Claude 经官方 CLI」或「Claude 经 HTTP」，通道由分组决定，不能手动切换；分组还没读到时先走 CLI。经 HTTP 时工具由 DSH 执行，计划模式、提问、待办、子代理、附件、思考档位和历史都和其它模型一样；经 CLI 时由 Claude Code 自己执行，差别见下方「Claude Code CLI 通道」。
+**用 Claude。** 直接选择四个 Claude 模型之一。Key 在 `cc` 分组时，这个会话由本机的官方 Claude Code CLI 完成，需要先在 **设置 → PackyCode → Claude Code CLI** 一键安装（或自行 `npm install -g @anthropic-ai/claude-code --allow-scripts=@anthropic-ai/claude-code`），版本过低或有新版本时在同一处点「更新 Claude Code CLI」；`cc-sale`、`cc-expensive`、`claude-officially` 等分组直接走插件内 HTTP，不需要 CLI。「账户与 Key」里每个能用 Claude 的 Key 会标出「Claude 经官方 CLI」或「Claude 经 HTTP」，通道由分组决定，不能手动切换；分组还没读到时先走 CLI。经 HTTP 时工具由 DSH 执行，计划模式、提问、待办、子代理、附件、思考档位和历史都和其它模型一样；经 CLI 时由 Claude Code 自己执行，差别见下方「Claude Code CLI 通道」。
 
 **网页搜索。** DSH 的 `web_search` 默认调用 DeepSeek 官方接口（用 `DEEPSEEK_API_KEY` 计费）。在 **设置 → PackyCode → 网页搜索** 里可以改走 PackyCode：
 
@@ -163,9 +163,13 @@ Claude 与其它模型一样，通过 DSH 的 `packycode_image_generate` 工具
 
 `gpt-image-2` 的质量为 `auto / low / medium / high`；两款 2.6 另支持 `xhigh / max`。设置里原来选的 2.5 Flare / Sunburst 会按同名的 2.6 型号读取，质量不变，无需重新设置。质量按 Images API 的 `quality` 原样发送，不作为文字推理强度；不会自动换型号、降档或重试付费请求。输出先采用 PNG，每次一张，聊天预览之外的原图保存在 `~/.dsh/packycode-api/images/`（设置了 `DSH_HOME` 时跟随该目录）。参考图与可选 PNG mask 必须来自当前会话；不能用路径替代附件引用。
 
+`MiniMax-H3-Image` 和 `MiniMax-H3-Image-Highspeed`（出图更快、单价更高）走同样的两个端点，但参数不同：没有质量档位（设置里只有「自动」，请求不带 `quality`），不支持透明背景和 mask，带这两项的请求会在发出前被拒绝，不会丢掉后照常生图。MiniMax 按输出像素计费，不给尺寸时默认出 4K；插件在尺寸为 `auto` 时附带 `area: "2K"`，由模型按描述或参考图选比例、约 400 万像素出图，也可以在对话里指定尺寸（如 `2720x1536`），上限与 GPT Image 相同（单边 3840、829 万像素）。
+
+图片卡片会在每个图片 Key 下显示它的分组。型号列表只列出当前使用的 Key 所在分组支持的型号，并在下方说明哪些型号因分组不支持而隐藏；已保存的型号若不被当前 Key 支持，会保留在列表里并标注「当前 Key 不支持」，同时给出警告和所需分组，不会自动改成别的型号。例如 MiniMax 两个型号在 PackyCode 只开给 `image` 分组，`image-sale` 分组的 Key 看不到它们。分组没读到时不隐藏任何型号。
+
 点击生成图可在应用内查看大图，右上角关闭按钮左侧的「下载原图」可选择保存位置，另存本机保留的完整 PNG。取消保存不会报错；若原图已被移走或删除，会提示失败，不会改为下载聊天缩略图。已有会话也可使用此功能，前提是记录对应的本机原图仍存在。
 
-用户填入图片 Key 后，可运行 `node scripts/image-check.mjs --all --edit` 验证三个型号与一次编辑；用 `--model <型号> --edit` 单独验证某个型号的生图和编辑，搭配 `--quality xhigh` 或 `--quality max` 验证质量。测试会消耗图片额度，输出写入 `.artifacts/`，日志只显示 Key 的掩码尾号，并记录网关回报的型号/质量。
+用户填入图片 Key 后，可运行 `node scripts/image-check.mjs --all --edit` 验证全部型号与一次编辑（Key 所在分组不提供的型号会单独报错，不影响其它型号）；用 `--model <型号> --edit` 单独验证某个型号的生图和编辑，搭配 `--quality xhigh` 或 `--quality max` 验证质量。测试会消耗图片额度，输出写入 `.artifacts/`，日志只显示 Key 的掩码尾号，并记录网关回报的型号/质量。
 
 2026-11-05 实测：2.6 Flare 和 2.6 Sunburst 都通过 `low`、`xhigh`、`max` 的生图与编辑，响应回报的 quality 与请求一致；`gpt-image-2` 通过 `low` 生图，Sunburst 的 `auto` 被网关回报为 `low`。合计 15 次请求全部成功，无重试，均返回 1024×1024 PNG；编辑正确改色并保留构图。Flare 的 xhigh 生图/编辑约 14/15 秒、max 约 24/26 秒；Sunburst 的 xhigh 约 51/64 秒、max 约 96/97 秒。这是已构建图片 client 的网关验证；本轮未新建 DSH 对话验证完整聊天流程。完整范围和记录见[协议核查与网关实测](docs/research/image-api.md)。
 
@@ -181,10 +185,12 @@ DSH 自带一个实验性插件「自动授权审查」，默认关闭。开启
 
 | 审查模型 | 说明 | 需要的分组 |
 | --- | --- | --- |
-| jev-latest（默认） | TypeSafe 的决策分类模型，一次约 0.4 秒，只按输入计费；拒绝时只给出风险等级 | `typesafe-officially` |
+| jev-1.14.0（默认） | TypeSafe 的决策分类模型，一次约 0.4 秒，只按输入计费；拒绝时只给出风险等级 | `typesafe-officially` |
 | DeepSeek Flash | 一次约 2 秒，拒绝时写明原因 | `deepseek-officially` / `deepseek-sale` |
 | 跟随对话模型 | 不接管，按官方默认审查 | — |
 
+以前保存的 `jev-latest` 仍然有效，会按 `jev-1.14.0` 审查。
+
 使用步骤：
 
 1. 在 DSH 左侧「插件」页开启「自动授权审查」。
@@ -202,17 +208,17 @@ DSH 自带一个实验性插件「自动授权审查」，默认关闭。开启
   - 要求返回一个 `{"risk","decision","reason"?}` 对象。
 - 所有模型请求都经过 DSH 的 `llm/stream` 流水线。本插件在这里只认这一种请求，对话等其它请求原样放过。
 - **DeepSeek Flash**：同一个请求原样改发到 PackyCode 线路，结果仍由官方解析。
-- **jev-latest**：它写不出 JSON。插件把四段内容和官方规则原文作为 System One 的 state，问两个问题：风险等级，以及你是否明确要求过这个操作。
+- **jev-1.14.0**：它写不出 JSON。插件把四段内容和官方规则原文作为 System One 的 state，问两个问题：风险等级，以及你是否明确要求过这个操作。
   - 放行条件：「低风险，或中风险且你要求过」的概率不低于 0.6，且高风险概率低于 0.2。
-  - 拒绝时，理由里写明是 jev-latest 判定的中风险或高风险。
-  - jev-latest 上下文较小，过长的历史会从最旧的开始截掉，你的指令优先保留。
-- 看什么、按什么规则判、放行或拒绝后怎么处理，都由官方插件决定。官方对审查失败（例如 Key 调不了 jev-latest）的处理是拒绝这次调用，本插件不改变这一点。
+  - 拒绝时，理由里写明是 jev-1.14.0 判定的中风险或高风险。
+  - jev-1.14.0 上下文较小，过长的历史会从最旧的开始截掉，你的指令优先保留。
+- 看什么、按什么规则判、放行或拒绝后怎么处理，都由官方插件决定。官方对审查失败（例如 Key 调不了 jev-1.14.0）的处理是拒绝这次调用，本插件不改变这一点。
 - 实测方式：37 个场景，用官方的请求格式和官方解析器。其中 35 个有明确答案，两个模型都全部符合预期。
-  - jev-latest 中位 0.35 秒，DeepSeek Flash 中位 2.3 秒。
+  - jev-1.14.0 中位 0.35 秒，DeepSeek Flash 中位 2.3 秒。
 - 如果以后 DSH 改了审查请求的格式：
   - 认不出这个请求时，原样放过，由对话模型审查；
-  - jev-latest 读不懂新格式时，这次审查失败，卡片计数里会看到。
-- 菜单里「自动审查」的说明写的是「由同一模型审查」，那是 DSH 的文字。选了 jev-latest 或 DeepSeek Flash 时，实际由所选模型审查。
+  - jev-1.14.0 读不懂新格式时，这次审查失败，卡片计数里会看到。
+- 菜单里「自动审查」的说明写的是「由同一模型审查」，那是 DSH 的文字。选了 jev-1.14.0 或 DeepSeek Flash 时，实际由所选模型审查。
 
 </details>
 
@@ -230,13 +236,14 @@ DSH 自带一个实验性插件「自动授权审查」，默认关闭。开启
 | `hy4`（Hy4，腾讯混元） | HTTP，Responses，可带图；1024K 上下文（输入最多 960K），64K 输出 | 可关；只有 `high` | `hunyuan-officially` |
 | `gemini-4-argon`（Gemini 4 Argon） | HTTP，Gemini 原生流式，可带图；1,048,576 上下文，65,536 输出 | 不能关；`low` / `medium` / `high` | `gemini-slb` |
 | `gemini-3.8-flash`（Gemini 3.8 Flash） | HTTP，Gemini 原生流式，可带图；1,048,576 输入，65,536 输出 | 不能关；`low` / `medium` / `high` | `gemini-slb` |
-| `claude-opus-5-5` | Anthropic Messages，可带图和原生工具：`cc` 分组由本机 Claude Code CLI 当 agent，其它分组走 HTTP | 不能关；`low` / `medium` / `high` / `xhigh` / `max` | 以设置页读取的 Claude 分组覆盖为准 |
+| `claude-fable-5-2`（Claude Fable 5.2） | Anthropic Messages，可带图和原生工具：`cc` 分组由本机 Claude Code CLI 当 agent，其它分组走 HTTP；1M 上下文，128K 输出 | 不能关；`low` / `medium` / `high` / `xhigh` / `max` | 以设置页读取的 Claude 分组覆盖为准 |
+| `claude-opus-5-5` | 同上 | 同上 | 同上 |
 | `claude-sonnet-5-5` | 同上 | 同上 | 同上 |
 | `claude-haiku-5-5` | 同上；Claude 会话标题固定走它 | 同上 | 同上 |
 
 列表里看得到模型，不代表当前令牌分组一定放行。换分组时不要把对应分组一起去掉。
 
-`jev-latest`（TypeSafe Jev）不在模型选择器里：它是只回答分类问题的决策模型，不能对话，只用于[自动审查](#自动审查)。插件在宿主侧直接调用 `/v1/systemone`，需要 Key 在 `typesafe-officially` 分组；「账户与 Key」里每个 Key 的「Jev」一项会显示能否使用。
+`jev-1.14.0`（TypeSafe Jev）不在模型选择器里：它是只回答分类问题的决策模型，不能对话，只用于[自动审查](#自动审查)。PackyCode 按版本号列出它，不再提供 `jev-latest` 别名，所以插件直接发送 `jev-1.14.0`。插件在宿主侧直接调用 `/v1/systemone`，需要 Key 在 `typesafe-officially` 分组；「账户与 Key」里每个 Key 的「Jev」一项会显示能否使用。
 
 当前目录的 13 个模型中，除 DeepSeek V4 Pro 外都声明了文字和图像输入。DeepSeek Flash 的识图能力按 [官方图像理解文档](https://api-docs.deepseek.com/zh-cn/guides/vision/) 配置；图片通过 DSH 附件服务读取，再随 Responses 请求发送。DeepSeek V4 Pro 按 [官方模型与价格页](https://api-docs.deepseek.com/zh-cn/quick_start/pricing/) 只收文字：DSH 不让给它附图，会话里已有的图片（比如从 Flash 切过来的会话）会换成一行「图片已省略」的说明再发送。PackyCode 网关收到发给 V4 Pro 的图片不报错，而是直接丢掉。
 
@@ -268,16 +275,16 @@ DeepSeek Flash 走网关的 Responses 接口（`/api/pricing` 列出 `openai-res
 - 两条通道的 Claude 对话都保留完整 DSH 工具循环、权限及计划模式；网页搜索和生图使用宿主工具。会话标题由 `claude-haiku-5-5` 生成（与对话走同一条通道）；GPT-6 Astra 与 GPT-5.6 Sol 的标题交给 GPT-6.1 Sol，DeepSeek V4 Pro 的标题交给 DeepSeek Flash。
 - 1.2.x 的 `claudeCliMode`、`claudeCliCwd` 设置不再恢复：Claude Code 的权限由会话的 DSH 权限预设决定，工作目录就是会话目录。名为 `claude_code` 的工具现在是 cc 会话里显示 Claude Code 工具调用的镜像工具，只由插件使用，不是 1.2.x 的委派工具。`claudeCliPath` 重新生效，用于指定 CLI 位置（留空则按常见安装位置和 PATH 查找）。1.3.4 的 `claudeTransport` 已移除，profile 里残留的值会被忽略。除非你在 CLI 卡片里点「更新 Claude Code CLI」，用户独立安装的 Claude Code 不会被改动。
 - Codex 的 Responses 请求仍需要 `prompt_cache_key`；插件补上缺失的会话标识，其它 Responses 模型省略该字段。
-- 构建后可用 `node scripts/claude-check.mjs` 验证三个 Claude 模型的订单核算、工单归类、库存补货、只读库存查询工具回传、商品色卡标注、中文标题和计划模式（只读查询 → `exit_plan_mode` 提交计划 → 审批），并核对业务答案；加 `--transport cli` 则改跑 cc 通道的检查（`scripts/claude-agent-check.mjs`）：在临时目录里让 Claude Code 自己读文件计算、改文件（授权批准）、删文件（授权拒绝）、计划审阅后实施、提问、看图和起标题，核对文件和答案，并统计网关的「客户端异常」和「反滥用」拒绝次数。`node scripts/claude-stability.mjs --start --hours 6 --key-label <备注名>` 开始有截止时间和请求上限的采样；定时执行同一脚本继续采样，用 `--status` 查看记录。业务场景版本和每轮构建分别记录；测试消息拒绝将对应检查记为失败，其余预定检查继续，真实凭据或额度错误仍会停止整个窗口。测试不改变已保存的凭据，也不启动 DSH。
+- 构建后可用 `node scripts/claude-check.mjs` 验证四个 Claude 模型的订单核算、工单归类、库存补货、只读库存查询工具回传、商品色卡标注、中文标题和计划模式（只读查询 → `exit_plan_mode` 提交计划 → 审批），并核对业务答案；加 `--transport cli` 则改跑 cc 通道的检查（`scripts/claude-agent-check.mjs`）：在临时目录里让 Claude Code 自己读文件计算、改文件（授权批准）、删文件（授权拒绝）、计划审阅后实施、提问、看图和起标题，核对文件和答案，并统计网关的「客户端异常」和「反滥用」拒绝次数。`node scripts/claude-stability.mjs --start --hours 6 --key-label <备注名>` 开始有截止时间和请求上限的采样；定时执行同一脚本继续采样，用 `--status` 查看记录。业务场景版本和每轮构建分别记录；测试消息拒绝将对应检查记为失败，其余预定检查继续，真实凭据或额度错误仍会停止整个窗口。测试不改变已保存的凭据，也不启动 DSH。
 - 连续对话可用 `node scripts/claude-conversation-check.mjs --model claude-haiku-5-5 --rounds 3 --key-label <备注名> --output .artifacts/claude-conversation.json`：同一插件适配器处理独立会话，每组保留六轮对话和完整工具历史，检查追问中的数值一致性、最终 JSON 和异常字符；遇到失败停止该批次并保留记录。单请求默认 65 秒、输出上限 2048 token；诊断可通过 `--timeout-ms`（5000–300000）和 `--max-tokens`（1024–128000）显式调整。`--capture-wire` 记录这组模拟业务的完整请求、非认证头和中断前的流事件，用于精确重放；不记录 Key。报告包含场景版本、适配器生命周期、所用上限、首响应和完整响应耗时。这些选项不修改插件的运行参数。
 
 </details>
 
 ## 更新与卸载
 
-> **从 1.3.2 或更早版本升级：** 请在插件页填写 `dsh-packycode-api@1.4.7` 安装，不要点旧版设置页里的「更新插件」。旧按钮安装时会绕过 pnpm 的发布冷却期：装到的版本如果发布还不满 24 小时，这个 profile 在它满 24 小时之前就装不了其他插件。装好并重启后，按钮就换成了修好的版本，以后直接点即可。
+> **从 1.3.2 或更早版本升级：** 请在插件页填写 `dsh-packycode-api@1.4.9` 安装，不要点旧版设置页里的「更新插件」。旧按钮安装时会绕过 pnpm 的发布冷却期：装到的版本如果发布还不满 24 小时，这个 profile 在它满 24 小时之前就装不了其他插件。装好并重启后，按钮就换成了修好的版本，以后直接点即可。
 
-**更新：** 打开 **设置 → PackyCode**，有新版本时点「更新插件」。它会在当前 profile 目录里用 DSH 自带的 pnpm 安装精确版本，并读回 `package.json` 确认版本落地。安装遵守 pnpm 的发布冷却策略，不会绕过它：默认情况下 pnpm 会直接装上这个版本，并把它记进 profile 的 `minimumReleaseAgeExclude`；如果你给 pnpm 配了更严格的策略，按钮会提示什么时候可以更新。新版本号是从 npm 读到的，pnpm 却从 profile 配置的安装源下载：刚发布的版本，安装源（尤其是国内镜像）可能还没同步到，这时按钮会提示过几分钟再更新；连不上安装源时，会提示检查网络或代理。本地 `link:` 开发安装不会出现这个按钮。完成后完全退出并重新打开桌面应用。只刷新页面不会重载 Host。也可以在**插件**页重新安装，同样要填写带版本号的包名（例如 `dsh-packycode-api@1.4.7`），否则装上的可能是发布满 24 小时的旧版。
+**更新：** 打开 **设置 → PackyCode**，有新版本时点「更新插件」。它会在当前 profile 目录里用 DSH 自带的 pnpm 安装精确版本，并读回 `package.json` 确认版本落地。安装遵守 pnpm 的发布冷却策略，不会绕过它：默认情况下 pnpm 会直接装上这个版本，并把它记进 profile 的 `minimumReleaseAgeExclude`；如果你给 pnpm 配了更严格的策略，按钮会提示什么时候可以更新。新版本号是从 npm 读到的，pnpm 却从 profile 配置的安装源下载：刚发布的版本，安装源（尤其是国内镜像）可能还没同步到，这时按钮会提示过几分钟再更新；连不上安装源时，会提示检查网络或代理。本地 `link:` 开发安装不会出现这个按钮。完成后完全退出并重新打开桌面应用。只刷新页面不会重载 Host。也可以在**插件**页重新安装，同样要填写带版本号的包名（例如 `dsh-packycode-api@1.4.9`），否则装上的可能是发布满 24 小时的旧版。
 
 **从本地路径安装切回官方源时，必须先卸载。** 「更新」是 `pnpm add`，它只重写包目录里的文件，不会删除更早一次本地路径安装留在包目录内的 `node_modules`；那份旧依赖会继续挡住 DSH 自带的包，插件行加载失败且看不出原因。「卸载」执行的是 `pnpm remove`，会删掉整个包目录，重装才是干净的。用 `node scripts/install-check.mjs` 可以确认当前 profile 是否处于这个状态。
 
@@ -293,7 +300,7 @@ ls ~/.dsh/profiles/desktop/node_modules/dsh-packycode-api/node_modules
 
 有输出就是残留：从 npm 安装的包不会自带 `node_modules`，这一份通常来自更早一次「本地路径」安装（`file:` 或直接写路径，而不是 `link:`）。它位于 profile 目录内部，而 DSH 解析插件导入时优先使用 profile 内的副本，所以插件拿到的仍是那份旧依赖 —— 例如 `@deepseek-ai/schemastery` 3.18.2 没有 `Schema.volatile()`，模块在导入阶段就抛错。导入失败时 Loader 不会为这一行创建 fiber，客户端 bundle 也不会被组合，于是模型路由、RPC 通道和设置页全都不会出现。
 
-修复：到**插件**页**卸载**本插件，再用带版本号的包名重新**安装**（例如 `dsh-packycode-api@1.4.7`），然后完全退出并重新打开桌面应用。卸载执行的是 `pnpm remove`，会删掉整个包目录；直接重装或点「更新」只重写文件，不会清掉这份残留。也可以手动删掉上面那个 `node_modules` 再重启。
+修复：到**插件**页**卸载**本插件，再用带版本号的包名重新**安装**（例如 `dsh-packycode-api@1.4.9`），然后完全退出并重新打开桌面应用。卸载执行的是 `pnpm remove`，会删掉整个包目录；直接重装或点「更新」只重写文件，不会清掉这份残留。也可以手动删掉上面那个 `node_modules` 再重启。
 
 **首次安装后，设置页每张卡片都显示 `transport failure for /packycode-api/…: HTTP 405`？** 实际装上的是 1.1.6 或更早的版本，通常是因为安装时只填了包名：安装提示显示的是最新版，pnpm 实际装的却是发布满 24 小时的旧版（见「安装插件」）。这些旧版要等 DSH 重启、重建它的 `connection` 服务之后，才能注册设置通道。所以首次安装后 `/packycode-api` 通道还不存在，请求落到静态页面，返回 405；模型路由和设置页本身照常加载。1.1.7 起首次安装不需要重启。先确认装上的版本：
 

package/docs/research/image-api.md

@@ -1,4 +1,4 @@
-# GPT Image 2 / 2.5 / 2.6 Images API 核查
+# GPT Image 2 / 2.5 / 2.6 与 MiniMax H3 Image：Images API 核查
 
 核查日期：2026-10-15（Asia/Shanghai）。协议核查部分只查阅公开资料；末尾另记录用户添加独立图片 Key 并授权测试后的 PackyCode 网关结果。协议能力与实际网关结果分开记录，不记录任何完整凭据。
 
@@ -98,3 +98,63 @@ JSON 编辑的字段为 `images: [{ image_url: "…" }]` 或 `images: [{ file_id
 | `gpt-image-2.6-sunburst` | `max` | 通过 | 通过 | `max` | 95.9 / 96.6 秒 |
 
 共 15 次请求全部成功，均返回 1024×1024 PNG。抽查的生图为白底蓝色茶壶，编辑后茶壶变橙、构图和白底保留。`auto` 发出后网关回报 `low`。响应没有可识别的 `model` 字段，所以结论仍是这些型号在当前 Key 和网关下可用、回报的 quality 与请求一致，不独立证明网关内部的模型身份。本轮没有在 DSH 桌面对话里跑完整的模型 → 工具 → 聊天预览流程。PNG 保存在本地 `.artifacts/image-check-*/`，不进入 Git 或 npm 包。
+
+## MiniMax H3 Image 接入（2026-11-11）
+
+2026-11-11 的 `/api/pricing` 在 `image` 分组列出 `MiniMax-H3-Image` 和 `MiniMax-H3-Image-Highspeed`，端点类型为 `image-generation`（`/v1/images/generations`，附加路径 `/v1/images/edits`），计费方式 `image_billing: "token"`；两款都不在 `image-sale` 分组。站内公告（`/api/status` 的 `announcements`，同日发布）写两个端点两款都支持、编辑支持 1~16 张参考图、「两款模型参数与限制相同，Highspeed 版本出图更快」。PackyCode 的文档仓库还没有这两个型号的页面。
+
+### MiniMax 官方协议
+
+来源：MiniMax 的[生图指南](https://platform.minimax.io/docs/guides/image-generation)、[文生图参考](https://platform.minimax.io/docs/api-reference/image-generation-h3-generations)和[图片编辑参考](https://platform.minimax.io/docs/api-reference/image-generation-h3-edits)。H3 Image 的原生接口就是这两个 Images API 路径，字段与 GPT Image 不同：
+
+| 字段 | MiniMax H3 Image |
+| --- | --- |
+| `prompt` | 必填，最长 32000 字符 |
+| `n` | 1–10，默认 1 |
+| `size` | `auto`（默认）或 `WIDTHxHEIGHT`：面积 655,360–20,000,000 像素，宽高比 1:3 至 3:1，不缩放不取整，越界返回 400；没有 16 的倍数要求 |
+| `area` | `1K` / `2K` / `4K`，**默认 `4K`**；与 `WIDTHxHEIGHT` 的 `size` 同时发送返回 400 |
+| `aspect_ratio` | `auto` 或 `X:Y`；与 `WIDTHxHEIGHT` 的 `size` 同时发送返回 400 |
+| `output_format` | `png` / `jpeg` / `jpg` / `webp`，**默认 `jpeg`** |
+| `response_format` | `url`（默认）或 `b64_json`；`b64_json` 时 PNG 总面积上限 20,000,000 像素 |
+| `quality`、`background` | schema 里没有；指南把透明背景列为不支持 |
+| `mask` | 不支持，返回 400 |
+| 编辑参考图 | multipart 的 `image` 或 `image[]`，1–16 张，JPG / PNG / WEBP / HEIC / HEIF，单张不超过 64 MiB，边长 128–8192 |
+
+响应为 `data[].b64_json`（或 `url`）加每张图的实际 `size` 和 `revised_prompt`，顶层有 `output_format` 和 `usage`，没有 `quality` 字段。官方写服务端最多等待 600 秒，接口不幂等，重试会再次生成并计费。两款参数与限制相同；Highspeed 的输入图片和输出单价是标准版的两倍。
+
+### 插件的请求形状
+
+- 两款进入 `IMAGE_MODELS`，质量只有 `auto`，请求不带 `quality` 和 `background`。
+- 显式发送 `output_format: "png"` 和 `response_format: "b64_json"`：插件只保存 PNG 原图，不下载网关给的链接。
+- `size` 为 `auto` 时附带 `area: "2K"`（`MINIMAX_AUTO_AREA`），比例由模型按描述或参考图决定。不这样做，MiniMax 按默认 `4K` 出图，像素和费用是 2K 的四倍，并超过插件保存图片的上限（单边 3840、8,294,400 像素）。显式尺寸不带 `area`（两者同发返回 400）。
+- 显式尺寸沿用原有校验（两边为 16 的倍数、单边不超过 3840、宽高比不超过 3:1、655,360–8,294,400 像素），是 MiniMax 允许范围的子集；4K 档暂不提供。
+- 透明背景和 mask 在发出请求前拒绝，不丢弃后照常生图。工具调用为本次请求点名 MiniMax、而设置里保存的是另一型号的质量时，按 `auto` 处理；显式传入它没有的质量仍被拒绝。
+- 网关的路由拒绝（`error.code: "model_not_found"`，如 `分组 image-sale 下模型 MiniMax-H3-Image 无可用渠道`）改为提示图片 Key 所在分组不提供该型号，并明确要求不要换型号重试、让用户在设置里选型号或换 Key，不回显网关原文。
+- 超时仍是插件原有的 5 分钟，低于官方的 600 秒建议；实测 2K 生图和编辑都在 30 秒内。
+
+### 网关实测（2026-11-11）
+
+已构建的图片 client 调用 `https://cf.api.fan`，Key 只从 `packycode-api/image-keys` 读取（尾号 …ECMQ），提示和编辑指令同前（白底蓝色茶壶；编辑为改成橙色），每次 `n: 1`、PNG，无重试。
+
+这把 Key 当天先在 `image-sale` 分组：两款 MiniMax 型号都被网关在到达上游前拒绝（HTTP 503 `model_not_found`），没有产生图片。用户把它改到 `image` 分组后：
+
+| 请求型号 | 请求尺寸 | 生图 | 编辑 | 返回尺寸 | 生图 / 编辑耗时 |
+| --- | --- | --- | --- | --- | --- |
+| `MiniMax-H3-Image` | `1024x1024` | 通过 | 通过 | 1024×1024 | 11.4 / 16.3 秒 |
+| `MiniMax-H3-Image-Highspeed` | `1024x1024` | 通过 | 通过 | 1024×1024 | 8.1 / 9.8 秒 |
+| `MiniMax-H3-Image` | `auto`（附带 `area: "2K"`） | 通过 | 通过 | 2048×2048 | 14.6 / 25.6 秒 |
+
+6 次请求全部返回可解码的 PNG（`data[].b64_json`），尺寸与请求一致。抽查的生图为白底蓝色茶壶，编辑后茶壶为橙色。响应没有 `model` 和 `quality` 字段，带每张图的 `size`、`revised_prompt` 和 `usage`。
+
+另用一次直接请求确认网关转发 `area`：JSON 生图发 `size: "auto"` 加 `area: "1K"` 和一段宽幅风景描述，返回 1568×672 的 PNG（约 105 万像素，比例由模型选定），`usage.output_tokens` 为 4116。multipart 编辑同样转发 `area`（上表第三行的编辑返回 2048×2048）。1K 这张图的输出 token 与 2K 档的起点（约 4096）接近，所以 `auto` 选 2K。本轮没有测透明背景和 mask 的上游行为（插件在发出前拒绝）、4K 档、多张参考图和 5 分钟以上的请求。
+
+同日用同一把 Key 回归 `gpt-image-2.6-flare` 的 `low`：生图 34.1 秒、编辑 48.5 秒，都成功，响应回报 `model: gpt-image-2.6-flare` 和 `quality: low`；请求的是 `1024x1024`，两次返回的 PNG 都是 1254×1254（2026-11-05 同样的请求返回 1024×1024）。GPT Image 的请求字段本次没有改动。
+
+### 图片 Key 的分组（2026-11-11）
+
+图片 Key 所在分组决定它能用哪些型号：`gpt-image-*` 同时列在 `image` 和 `image-sale`，MiniMax 两款只列在 `image`。桌面对话里用 `image-sale` 的 Key 选 MiniMax 生图时，工具返回分组提示，对话模型（DeepSeek Flash）随即自行换了型号重试，这违背「不自动换型号」。因此：
+
+- 插件对图片 Key 复用对话 Key 的分组读取器（`createGroupReader`，`modelIds` 传图片型号）：向 `/v1/chat/completions` 发一个网关无法路由的模型名，从拒绝文本里读出分组；读 `/v1/models` 得到这把 Key 能用的模型；读公开的 `/api/pricing` 得到每个型号所在的分组。三个读取都不计费、有超时、拒绝重定向，返回浏览器的只有分组名、分组说明和每个型号的布尔值。
+- 实测（Key 在 `image` 分组）：拒绝文本为 `分组 image 下模型 dsh-packycode-group-probe 无可用渠道（distributor）`，`/v1/models` 列出 `MiniMax-H3-Image`、`MiniMax-H3-Image-Highspeed`、`gpt-image-2`、`gpt-image-2.5-flare`、`gpt-image-2.5-sunburst`、`gpt-image-2.6-flare`、`gpt-image-2.6-sunburst`，已构建的读取器 2.7 秒读完，五个型号都判为可用。
+- `/api/pricing` 的内容随请求出口变化：同一时刻从本机直连只返回 42 个模型、13 个分组，没有 Claude 和 `gpt-image-*`，也没有 `image-sale`；经代理返回 85 个模型、26 个分组。所以型号是否可用以「分组命中」或「`/v1/models` 列出」任一为准，只有两者都有反证时才判为不可用；定价里查不到型号时，设置卡不显示「需要分组」。
+- 设置卡在每个图片 Key 下显示分组；型号列表只列当前使用的 Key 可用的型号，下方说明被隐藏的型号；已保存的型号若不可用，保留在列表里并标注，另给警告，不自动改设置。分组没读到时不隐藏任何型号。图片请求只用当前使用的 Key，不会换到另一把 Key。