dsh-image-edge-cap
限制发送至所选提供商或模型的图像最长边，并在 DeepSeek Harness 中提供独立的设置选项卡
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:mrbeandev/dsh-image-edge-cap#fa880a981ca1125d79bdc9d50cd4cf3579337e5cImage Edge Cap 以 dsh-image-edge-cap 发布，当前版本为 0.1.0。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
Caps the longest side of every image sent to the providers or models you choose in DeepSeek Harness (DSH), with its own Settings → Image size tab.
Images are shrunk with the aspect ratio kept (default limit 2000 px). Only the copy sent in the model request is resized: stored attachments and session logs never change. Providers and models you don't choose get exactly the same requests as before.
Contents: Supported DSH versions · Why this plugin exists · Install · Settings tab · How it works · Compatibility and risk · Development
Supported range: DSH 0.2.0-rc.1 up to (not including) 0.3.0. Check
yours with dsh --version.
| DSH version | npm tag (at the time of this release) | Status |
|---|---|---|
0.2.0-rc.2 | latest, next | ✅ Tested |
0.2.0-rc.1 | — | ✅ Tested |
0.2.1-alpha.1 | alpha | ⚠️ Integration tests pass; loads with an "untested" warning |
other releases from 0.2.0-rc.1 up to 0.3.0 (not included) | — | ⚠️ Untested: loads with a warning |
older than 0.2.0-rc.1 (0.1.x) | — | ❌ Not supported: they have no route-chosen request-image targets |
0.3.0 and newer (prereleases included) | — | ❌ Refused until tested (override with allowUntestedHarness: true) |
Node.js: ^22.19.0 or >=24.0.0.
The plugin has no runtime dependencies and does not need any other community plugin. A refused plugin logs a clear error at startup and does not load; the Settings tab then says the host plugin is not running. It never half-loads.
Anthropic's Messages API rejects a request that has more than 20 images when any image is over 2000 px on either side:
400 invalid_request_error: messages.N.content.1.image.source.base64.data:
At least one of the image dimensions exceed max allowed size for many-image requests: 2000 pixels
Every turn resends the whole history. Once a session passes 20 images and one
of them is a tall screenshot, every later turn fails, and /compact fails too
(it sends the same history). Full-page screenshots of mobile layouts are the
usual cause.
DSH's own requestImagePixelBudget limits total pixels (width × height),
not side length. Long thin images pass it unchanged:
| Image | Pixels | After a 1.15 MP budget | Anthropic many-image rule |
|---|---|---|---|
| 2460 × 392 (wide strip) | 0.96 MP | 2460 × 392 | ❌ over 2000 |
| 390 × 2475 (full-page mobile screenshot) | 0.97 MP | 390 × 2475 | ❌ over 2000 |
| 2718 × 1346 | 3.7 MP | 1524 × 755 | ✅ |
With this plugin set to 2000 px, those become 2000 × 319 and 315 × 2000, and the stuck session continues. The same limit is discussed in anthropics/claude-code#12351.
dsh plugin --profile web add dsh-image-edge-cap
Stop the running DSH Web process when your active work is finished, then start
it again (for example dsh web --no-open --port 3080) and refresh the page.
Open Settings → Image size and choose which providers or models to cap.
To install from a local checkout instead (keep the folder in place, the profile links to it):
dsh plugin --profile web add "/absolute/path/to/dsh-image-edge-cap"
To remove it:
dsh plugin --profile web remove dsh-image-edge-cap
Alternatively, set a provider to Off; its next request goes out unchanged.
The plugin adds its own Image size tab to Settings. It lists every registered model provider, and for each one you choose a mode:
| Mode | Effect |
|---|---|
| Off (default) | Nothing changes for this provider. |
| All models | Every model of this provider gets capped images. |
| Selected models | A checklist of the provider's models appears; only the checked models get capped images. Other models of the same provider are left alone. |
Then set Max edge (16–16384 px, default 2000) and press Apply.
Typical setup: a router provider that serves both Claude and GPT models →
Selected models, tick only the Claude models, max edge 2000. The other
models keep full-size images.
The rules are stored in this plugin's entry of the profile patch
(cordis.patch.yml), so you can also edit them by hand:
- id: dsh-image-edge-cap
name: dsh-image-edge-cap
config:
providers:
my-router:
mode: selected # off | all | selected
models: [cc/claude-a, cc/claude-b]
maxImageEdge: 2000
llm/stream listener reads the request's provider and model. If no
rule matches, it returns the stream untouched, so the request is
byte-identical to the one sent without the plugin.AsyncLocalStorage scope that holds the cap. The scope reaches the image
reads the adapter makes while the stream is consumed, even through other
llm/stream listeners (such as a rate limiter) and concurrent requests.readImageRequest on the attachments service. When the
pi-ai adapter asks for an image target inside a capped scope, the wrapper
lowers the target so its long edge is at most the cap. The short edge is
rounded the same way the store resizes, so the store's checks and cache
still match.Each capped image is logged once per process, for example:
dsh-image-edge-cap: capped a request image 390×2475 → 315×2000 (max edge 2000).
The plugin wraps one method of DSH's attachment store, and does it reversibly.
This is not an official extension API. It is installed when the attachments
service appears, removed when the plugin is disposed, and never installed
twice. The version gate refuses DSH releases outside the tested range, so a
changed internal API makes the plugin refuse to load instead of misbehaving.
readImageRequest. On DSH 0.2 that includes the pi-ai
adapter (custom OpenAI-compatible, Anthropic and other pi-ai providers),
which is what the tests cover, and the native DeepSeek adapter, so a rule on
a DeepSeek provider applies too. Providers you leave Off are untouched.Plugin config keys (on the dsh-image-edge-cap entry):
| Key | Meaning |
|---|---|
providers | The per-provider rules edited by the Settings tab. |
enabled: false | Load without capping anything. |
harnessEntry | Path of the running dsh CLI entry (detected automatically). |
allowUntestedHarness: true | Load on a DSH outside the supported range. Not recommended. |
The plugin has zero npm runtime or build dependencies. The host code is plain ESM, and a dependency-free builder writes the browser bundle in DSH's ModuleLoader format, using the shell's own React.
| File | Role |
|---|---|
index.mjs | Host entry: version gate, Config, llm/stream scope, store wrapper. |
cap.mjs | Pure rule matching and target math. |
client/index.mjs | Source of the Settings tab. |
lib/client.js | Generated by npm run build; never edit it by hand. |
scripts/check-release.mjs | Release gate (metadata, stale bundle, private data, packed files). |
scripts/e2e-gui.mjs | Headless-Chrome smoke test of the tab in a running DSH Web. |
npm run build # generate lib/client.js from client/index.mjs
npm test # unit tests: rules, scoping, wrapper, version gate, Settings tab
DSH_TEST_HARNESS_ENTRY="/absolute/path/to/dsh/lib/bin.js" npm run test:integration
npm run verify # build + tests + release gate
npm run pack:check # inspect the exact published file set
To run the integration suite against another DSH version, install that version
in a scratch folder and point DSH_TEST_HARNESS_ENTRY at it:
npm install --prefix /tmp/dsh-020rc2 @deepseek-ai/dsh@0.2.0-rc.2 --ignore-scripts
DSH_TEST_HARNESS_ENTRY=/tmp/dsh-020rc2/node_modules/@deepseek-ai/dsh/lib/bin.js npm run test:integration
The integration suite mounts the plugin in that DSH's own Cordis, next to the
real LocalAttachmentStore, the real LLM runtime and the real pi-ai adapter.
It saves real PNGs and sends real requests to a loopback mock server (fake
credentials, no paid API calls). Then it decodes the images in the HTTP body
and checks:
The UI tests use a fake React harness. Version 0.1.0 was also checked in the
real DSH Web app on 0.2.0-rc.2 and 0.2.0-rc.1, by installing the packed tarball
into an isolated DSH_HOME with a mock provider (see
LOCAL-SETUP.md). On each version:
MIT. Maintainer release steps are in RELEASE.md.