fix(provider): improve Kimi context, model names, and image attachments

.vscodeignore

@@ -2,6 +2,9 @@
 .vscode/**
 .vscode-test/**
 node_modules/**
+!node_modules/@silvia-odwyer/
+!node_modules/@silvia-odwyer/photon-node/
+!node_modules/@silvia-odwyer/photon-node/**
 src/**
 *.vsix
 tsconfig.json

CHANGELOG.md

@@ -2,6 +2,17 @@
 
 All notable changes to the **OpenCode Go BYOK Provider** extension are documented here.
 
+## [Unreleased]
+
+### Added
+
+- **`[Model Picker]` Optional provider prefixes (#92).** Added `opencodego.showProviderPrefix` (default `true`) to hide `OpenCode Go` / `OpenCode Zen` prefixes in narrow model pickers when desired. Changes refresh the registered model names immediately.
+- **`[Model Picker]` Kimi context-size selector (#87).** Kimi models with a context window larger than 256K now expose `256K` and the full window in the per-model configuration schema, with the smaller tier selected by default.
+
+### Fixed
+
+- **`[Vision]` Normalize image attachments before OpenCode Go requests (#94).** Image data URLs are resized and re-encoded using the same 2000x2000 / 5 MB base64 limits used by OpenCode's CLI, while preserving the original image when normalization is unavailable.
+
 ## [0.4.5] — 2026-08-03
 
 ### Fixed

README.md

@@ -326,6 +326,7 @@ To manage agent API keys separately or see agent vendors in the Manage panel, en
 | `opencodego.requestTimeoutSeconds` | `600` | Total request timeout |
 | `opencodego.streamIdleTimeoutSeconds` | `120` | Cancel if stream goes idle |
 | `opencodego.showUsageStatusBar` | `true` | Show usage summary in status bar |
+| `opencodego.showProviderPrefix` | `true` | Include `OpenCode Go` / `OpenCode Zen` in model names |
 | `opencodego.freeOnly` | `true` | Zen: free models only. `false` = include paid |
 | `opencodego.agentsWindow` | `true` | Expose agent-host model variants (`targetChatSessionType`) for the Agents window |
 | `opencodego.showAgentModelsInManagePanel` | `false` | Show agent vendors in Manage Language Models panel |

docs/issues/44-20260803-issue94-image-normalization.md

@@ -0,0 +1,18 @@
+**Status:** Implemented on `fix/open-issues-87-92-94`
+
+# Image Attachment Normalization (#94)
+
+OpenCode Go rejected some image attachments even though the same model and
+image worked in the OpenCode CLI. The extension sent the original bytes as an
+OpenAI `image_url` data URI, while the CLI normalizes images before sending.
+
+The request path now normalizes image data URLs with the same practical limits
+used by OpenCode: a maximum `2000x2000` image size and a `5 MB` base64 payload.
+It tries PNG first, then JPEG quality levels. If decoding or the optional
+normalizer fails, the original data URI is preserved and the final base64 guard
+decides whether it is still safe to send.
+
+Top-level images are normalized before the final `5 MB` base64 guard, so an
+image larger than the old `2 MB` raw-byte threshold can still be sent when it
+can be reduced successfully. The separate `1 MB` raw-byte guard for tool
+results remains in place to bound cumulative MCP screenshot history.

docs/issues/45-20260803-issue92-provider-prefix.md

@@ -0,0 +1,9 @@
+**Status:** Implemented on `fix/open-issues-87-92-94`
+
+# Optional Provider Prefixes (#92)
+
+Model names in the picker now respect the `opencodego.showProviderPrefix`
+setting. It defaults to `true`, preserving names such as
+`OpenCode Go / DeepSeek V4 Flash`; setting it to `false` displays only the
+formatted model name. Changing the setting refreshes Go, Zen, and agent-host
+model registrations without requiring a reload.

docs/issues/46-20260803-issue87-kimi-context-size.md

@@ -0,0 +1,9 @@
+**Status:** Implemented on `fix/open-issues-87-92-94`
+
+# Kimi Context Size Picker (#87)
+
+The model configuration schema now exposes `256K` and the full context window
+for Kimi models whose resolved metadata advertises more than `256K` (for
+example, Kimi K3). The `256K` choice is the default/cheaper tier. Explicit
+context tiers from `models.dev` still take precedence, and fixed-size Kimi
+models do not receive a redundant selector.

package.json

@@ -167,6 +167,11 @@
           "default": true,
           "description": "Show the latest OpenCode prompt/output/cache usage summary in the VS Code status bar after each response."
         },
+        "opencodego.showProviderPrefix": {
+          "type": "boolean",
+          "default": true,
+          "description": "Show the OpenCode Go or OpenCode Zen provider prefix in model names."
+        },
         "opencodego.freeOnly": {
           "type": "boolean",
           "default": true,
@@ -331,5 +336,8 @@
     "@types/vscode": "^1.125.0",
     "@vscode/vsce": "^3.9.2",
     "typescript": "^6.0.3"
+  },
+  "dependencies": {
+    "@silvia-odwyer/photon-node": "^0.3.4"
   }
 }