opencode-see-image 0.4.1 → 0.4.3

package/README.md

@@ -6,6 +6,14 @@ When a user attaches a screenshot to a text-only model, opencode rejects it with
 
 ## Install
 
+**Option A, one command (recommended):**
+```bash
+opencode plugin opencode-see-image --global
+```
+This installs the package and adds it to your config. Then restart opencode.
+
+**Option B, edit config manually:**
+
 Add the plugin to your opencode config:
 
 ```jsonc
@@ -17,20 +25,13 @@ Add the plugin to your opencode config:
 ```
 Then restart opencode.
 
+## Install via your agent
 
-## Install via your agent (copy-paste this)
-
-Paste this prompt to your opencode agent, and it'll install the plugin for you.
-
+Ask your agent:
 ```
-Install the opencode-see-image plugin so I can send you screenshots. Do this:
-
-1. Edit ~/.config/opencode/opencode.jsonc (create it if missing). Preserve any existing fields and the $schema. Add "opencode-see-image" to the "plugin" array. If "plugin" doesn't exist, add it as ["opencode-see-image"].
-2. Check that a vision-capable provider is connected by looking for ~/.local/share/opencode/auth.json with either an "opencode-go" entry (paid, fast) OR an "opencode" entry (free). If neither is present, tell me to run /connect and select either opencode-go or opencode (key from opencode.ai/auth).
-3. Tell me to quit and restart opencode for the plugin to load.
-
-After I restart and attach a screenshot, you should call the see_image tool to view it.
+install the opencode-see-image plugin
 ```
+It'll run `opencode plugin opencode-see-image --global` and tell you to restart.
 
 ## Prerequisites
 
@@ -71,7 +72,7 @@ plugin's system-prompt instructions tell the model to call see_image
 see_image tool:
   1. locates the file (macOS screenshot temp dirs, ~/Desktop, ~/Downloads, cwd)
   2. base64-encodes it
-  3. sends it to the vision model via the Anthropic Messages API
+  3. routes it to the vision model via opencode's SDK (or direct HTTP if SEE_IMAGE_API_KEY is set)
   4. returns the textual description
         │
         ▼
@@ -91,16 +92,16 @@ Your model calls this tool automatically when you attach a screenshot, you don't
 
 ## Configuration
 
-All settings are env-var overrides. Defaults work out-of-the-box for opencode-go + MiniMax M3.
+All settings are env-var overrides. The plugin uses opencode's SDK client by default (handles auth automatically). Set `SEE_IMAGE_API_KEY` to bypass the SDK and call an HTTP endpoint directly.
 
 | Env var | Default | Description |
 |---|---|---|
-| `SEE_IMAGE_MODEL` | `minimax-m3` | Vision model ID to call |
-| `SEE_IMAGE_PROVIDER` | `opencode-go` | Provider key in opencode's `auth.json` |
-| `SEE_IMAGE_ENDPOINT` | `https://opencode.ai/zen/go/v1/messages` | Anthropic-Messages-compatible endpoint |
-| `SEE_IMAGE_API_KEY` | _(reads auth.json)_ | Bypass auth.json with an explicit key |
-| `SEE_IMAGE_API_VERSION` | `2023-06-01` | `anthropic-version` header value |
-| `SEE_IMAGE_USER_AGENT` | _(Chrome UA)_ | Override the User-Agent header |
+| `SEE_IMAGE_MODEL` | `minimax-m3` | Vision model ID |
+| `SEE_IMAGE_PROVIDER` | `opencode-go` | Provider ID for SDK routing |
+| `SEE_IMAGE_API_KEY` | _(uses SDK)_ | Bypass SDK, call HTTP endpoint directly |
+| `SEE_IMAGE_ENDPOINT` | `https://opencode.ai/zen/go/v1/messages` | HTTP endpoint (only used if `SEE_IMAGE_API_KEY` is set) |
+| `SEE_IMAGE_API_VERSION` | `2023-06-01` | `anthropic-version` header (HTTP mode only) |
+| `SEE_IMAGE_USER_AGENT` | _(Chrome UA)_ | User-Agent header (HTTP mode only) |
 
 ### Using a different vision model
 
@@ -147,9 +148,14 @@ Then restart opencode. (No bun required, this uses opencode's own bun.)
 
 **Pin a version** in your config to opt out of auto-updates:
 ```jsonc
-"plugin": ["opencode-see-image@0.4.1"]
+"plugin": ["opencode-see-image@0.4.2"]
 ```
 
+## Limitations
+
+- **Clipboard pastes don't work** — when you paste an image from clipboard (Cmd+V), opencode processes it in-memory but discards it before writing to disk if the model doesn't support image input. The plugin can't access it. **Drag screenshots instead**, or save the clipboard image to a file first.
+- **macOS only** — file search locations target macOS screenshot temp dirs. Linux/Windows users need to pass absolute paths.
+
 ## File search locations
 
 When opencode rejects an image attachment, the model only receives a bare filename. `see_image` searches these locations in order:

package/index.ts

@@ -64,9 +64,19 @@ function resolveFilePath(name: string, cwd: string): string {
     } catch {}
   }
 
-  const searched = searchDirs.filter(Boolean).join(", ")
+  const nsirdCount = searchDirs.filter((d) =>
+    String(d).includes("NSIRD_screencaptureui"),
+  ).length
+  const summary = [
+    `${nsirdCount} macOS screenshot temp dirs`,
+    "~/Desktop",
+    "~/Downloads",
+    cwd,
+  ]
+    .filter(Boolean)
+    .join(", ")
   throw new Error(
-    `see_image: could not find "${name}". Searched: ${searched}. ` +
+    `see_image: could not find "${name}". Searched ${summary}. ` +
       `Pass an absolute filePath instead.`,
   )
 }
@@ -76,6 +86,7 @@ async function seeImageViaSDK(
   dataUrl: string,
   mediaType: string,
   prompt: string,
+  abort?: AbortSignal,
 ): Promise<{ text: string; model: string; provider: string }> {
   const envProvider = process.env.SEE_IMAGE_PROVIDER
   const envModel = process.env.SEE_IMAGE_MODEL
@@ -144,6 +155,7 @@ async function seeImageViaHTTP(
   b64: string,
   mediaType: string,
   prompt: string,
+  abort?: AbortSignal,
 ): Promise<{ text: string; model: string; provider: string }> {
   const key = process.env.SEE_IMAGE_API_KEY!
   const body = {
@@ -172,6 +184,7 @@ async function seeImageViaHTTP(
       "user-agent": USER_AGENT,
     },
     body: JSON.stringify(body),
+    signal: abort,
   })
 
   if (!res.ok) {
@@ -347,9 +360,9 @@ const SeeImagePlugin: Plugin = async (ctx) => {
       let result: { text: string; model: string; provider: string }
 
       if (process.env.SEE_IMAGE_API_KEY) {
-        result = await seeImageViaHTTP(b64, mediaType, prompt)
+        result = await seeImageViaHTTP(b64, mediaType, prompt, context.abort)
       } else {
-        result = await seeImageViaSDK(client, dataUrl, mediaType, prompt)
+        result = await seeImageViaSDK(client, dataUrl, mediaType, prompt, context.abort)
       }
 
       context.metadata({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-see-image",
-  "version": "0.4.1",
+  "version": "0.4.3",
   "description": "Give non-vision opencode models the ability to see images/screenshots by routing them to a vision-capable model (MiniMax M3 via opencode-go by default).",
   "type": "module",
   "main": "index.ts",