@bogyie/opencode-kiro-plugin 0.1.0 → 0.2.2

package/CHANGELOG.md

@@ -15,6 +15,7 @@ Initial OpenCode Kiro provider plugin implementation.
 - OpenAI-compatible request adapter for system prompts, chat history, multimodal data URL images/documents, tool specs, and tool results.
 - OpenAI-compatible generation options for temperature, token limit, and reasoning/thinking effort are preserved as best-effort model request fields in direct fetch mode.
 - CodeWhisperer metadata token usage is mapped to OpenAI-compatible non-streaming usage fields on a best-effort basis.
+- CodeWhisperer reasoning content is preserved separately from assistant text as OpenAI-compatible `reasoning_content`.
 - CodeWhisperer/Kiro transport with streaming text, tool-call deltas, configurable retry attempts, and request timeout.
 - CLI fallback using `kiro-cli chat --no-interactive`.
 - CLI fallback responses are wrapped as OpenAI-compatible SSE for streaming OpenCode calls and sanitized to remove terminal prompt/credits output.

package/README.md

@@ -10,16 +10,7 @@ After the package is published to npm, add it directly to your OpenCode config:
 
 ```jsonc
 {
-  "plugin": ["@bogyie/opencode-kiro-plugin"],
-  "provider": {
-    "kiro": {
-      "models": {
-        "sonnet": {
-          "name": "Sonnet alias"
-        }
-      }
-    }
-  }
+  "plugin": ["@bogyie/opencode-kiro-plugin"]
 }
 ```
 
@@ -34,16 +25,7 @@ Then add the plugin to your OpenCode config. See [examples/opencode.jsonc](examp
 
 ```jsonc
 {
-  "plugin": ["file:/absolute/path/to/opencode-kiro-plugin"],
-  "provider": {
-    "kiro": {
-      "models": {
-        "sonnet": {
-          "name": "Sonnet alias"
-        }
-      }
-    }
-  }
+  "plugin": ["file:/absolute/path/to/opencode-kiro-plugin"]
 }
 ```
 
@@ -92,7 +74,7 @@ Supported values:
 
 ## Model Churn Handling
 
-The resolver intentionally avoids a hard whitelist. Fallback presets are used for OpenCode UI metadata and cache bootstrap only.
+The resolver intentionally avoids a hard whitelist. By default, the plugin reads the current Kiro CLI model list with `kiro-cli chat --list-models --format json`. Bundled presets are only used as UI metadata and as a fallback when discovery is unavailable.
 
 Useful options:
 
@@ -104,13 +86,13 @@ Useful options:
       {
         "modelCacheTtlSeconds": 21600,
         "modelDiscovery": "auto",
-        "modelDiscoveryCommand": ["kiro-cli", "models", "--json"],
+        "modelDiscoveryCommand": ["kiro-cli", "chat", "--list-models", "--format", "json"],
         "modelAliases": {
-          "sonnet": "claude-sonnet-4.6",
-          "opus": "claude-opus-4.6"
+          "sonnet": "claude-sonnet-5",
+          "opus": "claude-opus-4.8"
         },
         "extraModels": {
-          "claude-opus-4-9": {
+          "claude-opus-4.9": {
             "name": "Claude Opus 4.9",
             "limit": { "context": 1000000, "output": 64000 },
             "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] },
@@ -138,9 +120,11 @@ Resolution order:
 6. Hidden/manual model mapping
 7. Optimistic pass-through unless disabled
 
-This keeps new Kiro model ids usable before the package is updated. Use `extraModels` when a new model should appear in OpenCode's model picker immediately. Set `disableModelPassThrough: true` only when you need strict model governance.
+This keeps new Kiro model ids visible before the package is updated. Use `extraModels` only when a new model should appear in OpenCode's model picker but is not yet listed by your installed Kiro CLI. Set `disableModelPassThrough: true` only when you need strict model governance.
+
+The plugin injects `provider.kiro` automatically. You only need to add `provider.kiro.models` yourself when overriding OpenCode model-picker metadata, such as a display name or context limit. Use plugin `modelAliases` for aliases that should resolve one requested model id to another.
 
-`modelDiscoveryCommand` is optional and intentionally not guessed by default because Kiro CLI model-list flags may vary by version. When configured, stdout can be a JSON array, `{ "models": [...] }`, `{ "data": [...] }`, or one model id per line.
+`modelDiscoveryCommand` defaults to `["kiro-cli", "chat", "--list-models", "--format", "json"]`. Set `modelDiscovery` to `"off"` to skip runtime discovery. Discovery stdout can be a JSON array, `{ "models": [...] }`, `{ "data": [...] }`, Kiro CLI list-models JSON, Kiro CLI plain list output, or one model id per line.
 
 ## Troubleshooting
 
@@ -155,6 +139,9 @@ Direct fetch mode uses AWS SDK standard retry behavior. Tune `maxAttempts` and `
 
 OpenAI-compatible `temperature`, `max_tokens`, `max_completion_tokens`, `reasoning_effort`, `reasoning.effort`, and `thinking.effort` are preserved for direct fetch mode through Kiro's `additionalModelRequestFields` path on a best-effort basis.
 
+When Kiro emits token metadata in direct fetch mode, non-streaming responses map it to OpenAI-compatible `usage` fields.
+When Kiro emits reasoning text, it is preserved separately from assistant text as `reasoning_content`.
+
 For local checks:
 
 ```sh
@@ -167,13 +154,18 @@ For real Kiro/OpenCode validation, use [docs/e2e-validation.md](docs/e2e-validat
 
 ## Release
 
-GitHub Release publishing drives npm publishing. To publish a new npm package:
+The npm package is published through npm Trusted Publishing with the GitHub Actions environment named `npm`. No `NPM_TOKEN` repository secret is required.
+
+Configure the npm package trusted publisher to use:
+
+- repository owner: `bogyie`
+- repository name: `opencode-kiro-plugin`
+- workflow filename: `npm-publish.yml`
+- environment name: `npm`
 
-1. Set the repository secret `NPM_TOKEN` to an npm automation token with publish access to `@bogyie/opencode-kiro-plugin`.
-2. Update `package.json` and `package-lock.json` to the release version.
-3. Create and publish a GitHub Release whose tag matches the package version, for example `v0.1.0`.
+To publish a new version, update `package.json` and `package-lock.json` to the release version and merge the change to `main`. `package.json.version` is the source of truth: when it changes on `main`, the `Publish npm package` workflow publishes that package version to npm and creates the matching GitHub Release tag, for example `v0.1.0`.
 
-The `Publish npm package` workflow runs tests, typecheck, package smoke validation, `npm pack --dry-run`, and then `npm publish --provenance --access public`.
+The workflow runs tests, typecheck, package smoke validation, `npm pack --dry-run`, and then `npm publish --access public`.
 
 ## License And References
 

package/dist/config.d.ts

@@ -22,6 +22,7 @@ export interface KiroPluginOptions {
 }
 export declare const DEFAULT_PROVIDER_ID = "kiro";
 export declare const DEFAULT_REGION = "us-east-1";
+export declare const DEFAULT_MODEL_DISCOVERY_COMMAND: readonly ["kiro-cli", "chat", "--list-models", "--format", "json"];
 export declare const DEFAULT_MODEL_CACHE_TTL_SECONDS: number;
 export declare const DEFAULT_MAX_ATTEMPTS = 3;
 export declare function loadOptions(raw?: unknown): KiroPluginOptions;

package/dist/config.js

@@ -1,5 +1,6 @@
 export const DEFAULT_PROVIDER_ID = "kiro";
 export const DEFAULT_REGION = "us-east-1";
+export const DEFAULT_MODEL_DISCOVERY_COMMAND = ["kiro-cli", "chat", "--list-models", "--format", "json"];
 export const DEFAULT_MODEL_CACHE_TTL_SECONDS = 6 * 60 * 60;
 export const DEFAULT_MAX_ATTEMPTS = 3;
 const BACKENDS = new Set(["auto", "fetch", "cli-chat", "acp"]);
@@ -53,13 +54,14 @@ export function loadOptions(raw = {}) {
     const profileArn = optionalString(input.profileArn);
     const userAgent = optionalString(input.userAgent);
     const agentMode = optionalString(input.agentMode);
+    const modelDiscoveryCommand = "modelDiscoveryCommand" in input ? stringArray(input.modelDiscoveryCommand) : [...DEFAULT_MODEL_DISCOVERY_COMMAND];
     return {
         providerID: typeof input.providerID === "string" && input.providerID ? input.providerID : DEFAULT_PROVIDER_ID,
         region: typeof input.region === "string" && input.region ? input.region : DEFAULT_REGION,
         ...(endpoint ? { endpoint } : {}),
         backend,
         modelDiscovery,
-        modelDiscoveryCommand: stringArray(input.modelDiscoveryCommand),
+        modelDiscoveryCommand,
         modelCacheTtlSeconds: positiveNumber(input.modelCacheTtlSeconds, DEFAULT_MODEL_CACHE_TTL_SECONDS),
         ...(requestTimeoutMs ? { requestTimeoutMs } : {}),
         maxAttempts: positiveInteger(input.maxAttempts, DEFAULT_MAX_ATTEMPTS),

package/dist/fetch-adapter.js

@@ -8,6 +8,8 @@ const unsupportedTransport = {
 };
 async function* streamGeneratedResponse(transport, request) {
     const response = await transport.generate(request);
+    if (response.reasoning)
+        yield { type: "reasoning", text: response.reasoning, modelId: response.modelId ?? request.modelId };
     if (response.text)
         yield { type: "text", text: response.text, modelId: response.modelId ?? request.modelId };
     for (const toolCall of response.toolCalls ?? [])

package/dist/index.d.ts

@@ -16,7 +16,7 @@ export { FALLBACK_MODELS } from "./models.js";
 export { toKiroGenerateRequest } from "./request-adapter.js";
 export type { KiroGenerateRequest, KiroModelOptions } from "./request-adapter.js";
 export { toOpenAIChatResponse, toOpenAIChatStreamResponse } from "./response-adapter.js";
-export type { KiroStreamEvent, KiroToolCallChunk } from "./response-adapter.js";
+export type { KiroReasoningChunk, KiroStreamEvent, KiroToolCallChunk } from "./response-adapter.js";
 declare const _default: {
     id: string;
     server: import("@opencode-ai/plugin").Plugin;

package/dist/kiro-transport.js

@@ -135,6 +135,7 @@ export async function collectAssistantText(stream) {
     if (!stream)
         return { text: "" };
     let text = "";
+    let reasoning = "";
     let modelId;
     let usage;
     for await (const event of stream) {
@@ -142,6 +143,9 @@ export async function collectAssistantText(stream) {
             text += event.assistantResponseEvent.content ?? "";
             modelId = event.assistantResponseEvent.modelId ?? modelId;
         }
+        if (event.reasoningContentEvent?.text) {
+            reasoning += event.reasoningContentEvent.text;
+        }
         usage = usageFromMetadataEvent(event) ?? usage;
         if (event.error) {
             throw new Error(event.error.message ?? "Kiro stream returned an error event");
@@ -149,6 +153,7 @@ export async function collectAssistantText(stream) {
     }
     return {
         text,
+        ...(reasoning ? { reasoning } : {}),
         ...(modelId ? { modelId } : {}),
         ...(usage ? { usage } : {}),
     };
@@ -165,6 +170,12 @@ export async function* streamAssistantText(stream) {
                 ...(event.assistantResponseEvent.modelId ? { modelId: event.assistantResponseEvent.modelId } : {}),
             };
         }
+        if (event.reasoningContentEvent?.text) {
+            yield {
+                type: "reasoning",
+                text: event.reasoningContentEvent.text,
+            };
+        }
         if (event.toolUseEvent?.toolUseId && event.toolUseEvent.name) {
             const id = event.toolUseEvent.toolUseId;
             const current = toolInputs.get(id) ?? { name: event.toolUseEvent.name, input: "" };
@@ -196,6 +207,7 @@ export async function* streamAssistantText(stream) {
 }
 async function collectChunks(chunks, fallbackModelId) {
     let text = "";
+    let reasoning = "";
     let modelId;
     let usage;
     const toolCalls = [];
@@ -205,6 +217,11 @@ async function collectChunks(chunks, fallbackModelId) {
             modelId = chunk.modelId ?? modelId;
             continue;
         }
+        if (chunk.type === "reasoning") {
+            reasoning += chunk.text;
+            modelId = chunk.modelId ?? modelId;
+            continue;
+        }
         text += chunk.text;
         modelId = chunk.modelId ?? modelId;
         if (chunk.usage)
@@ -212,6 +229,7 @@ async function collectChunks(chunks, fallbackModelId) {
     }
     return {
         text,
+        ...(reasoning ? { reasoning } : {}),
         modelId: modelId ?? fallbackModelId,
         ...(toolCalls.length > 0 ? { toolCalls } : {}),
         ...(usage ? { usage } : {}),

package/dist/model-discovery.js

@@ -6,6 +6,9 @@ function record(value) {
 function stringValue(value) {
     return typeof value === "string" && value.trim() ? value.trim() : undefined;
 }
+function positiveNumberValue(value) {
+    return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : undefined;
+}
 function fromItem(item) {
     if (typeof item === "string") {
         const id = normalizeModelName(item);
@@ -14,16 +17,25 @@ function fromItem(item) {
     const object = record(item);
     if (!object)
         return undefined;
-    const id = stringValue(object.id) ?? stringValue(object.modelId) ?? stringValue(object.model) ?? stringValue(object.name);
+    const id = stringValue(object.id) ??
+        stringValue(object.modelId) ??
+        stringValue(object.model_id) ??
+        stringValue(object.model) ??
+        stringValue(object.model_name) ??
+        stringValue(object.name);
     if (!id)
         return undefined;
     const normalized = normalizeModelName(id);
     if (!normalized)
         return undefined;
     const name = stringValue(object.name) ?? stringValue(object.displayName) ?? stringValue(object.label);
+    const contextLimit = positiveNumberValue(object.contextLimit) ?? positiveNumberValue(object.context_window_tokens);
+    const outputLimit = positiveNumberValue(object.outputLimit) ?? positiveNumberValue(object.output_limit_tokens);
     return {
         id: normalized,
         ...(name ? { name } : {}),
+        ...(contextLimit ? { contextLimit } : {}),
+        ...(outputLimit ? { outputLimit } : {}),
         raw: item,
     };
 }
@@ -34,13 +46,18 @@ function fromJson(value) {
     if (!object)
         return [];
     const items = object.models ?? object.data ?? object.items;
-    return Array.isArray(items) ? fromJson(items) : [];
+    if (Array.isArray(items))
+        return fromJson(items);
+    const configuredModel = object["chat.defaultModel"] ?? object.defaultModel ?? object.model;
+    return configuredModel ? fromJson([configuredModel]) : [];
 }
 function fromLines(raw) {
     return raw
         .split(/\r?\n/)
         .map((line) => line.trim())
-        .filter((line) => line && !line.includes(" ") && !line.includes("\t"))
+        .map((line) => line.replace(/^\*\s*/, ""))
+        .map((line) => /^\S+/.exec(line)?.[0] ?? "")
+        .filter((line) => line === "auto" || /[.-]/.test(line))
         .map(fromItem)
         .filter((item) => item !== undefined);
 }

package/dist/models.js

@@ -7,31 +7,61 @@ export const FALLBACK_MODELS = {
         modalities: TEXT_IMAGE_PDF,
         tool_call: true,
     },
+    "claude-fable-5": {
+        name: "Claude Fable 5",
+        limit: { context: 1_000_000, output: 64_000 },
+        modalities: TEXT_IMAGE_PDF,
+        tool_call: true,
+    },
+    "claude-sonnet-5": {
+        name: "Claude Sonnet 5",
+        limit: { context: 1_000_000, output: 64_000 },
+        modalities: TEXT_IMAGE_PDF,
+        tool_call: true,
+    },
     "claude-sonnet-4": {
-        name: "Claude Sonnet 4",
+        name: "Claude Sonnet 4.0",
         limit: { context: 200_000, output: 64_000 },
         modalities: TEXT_IMAGE_PDF,
         tool_call: true,
     },
-    "claude-sonnet-4-5": {
+    "claude-sonnet-4.5": {
         name: "Claude Sonnet 4.5",
         limit: { context: 200_000, output: 64_000 },
         modalities: TEXT_IMAGE_PDF,
         tool_call: true,
     },
-    "claude-sonnet-4-6": {
+    "claude-sonnet-4.6": {
         name: "Claude Sonnet 4.6",
         limit: { context: 1_000_000, output: 64_000 },
         modalities: TEXT_IMAGE_PDF,
         tool_call: true,
     },
-    "claude-opus-4-8": {
+    "claude-opus-4.8": {
         name: "Claude Opus 4.8",
+        limit: { context: 1_000_000, output: 128_000 },
+        modalities: TEXT_IMAGE_PDF,
+        tool_call: true,
+    },
+    "claude-opus-4.7": {
+        name: "Claude Opus 4.7",
+        limit: { context: 1_000_000, output: 64_000 },
+        modalities: TEXT_IMAGE_PDF,
+        tool_call: true,
+    },
+    "claude-opus-4.6": {
+        name: "Claude Opus 4.6",
         limit: { context: 1_000_000, output: 64_000 },
         modalities: TEXT_IMAGE_PDF,
         tool_call: true,
     },
-    "claude-haiku-4-5": {
+    "claude-opus-4.5": {
+        name: "Claude Opus 4.5",
+        limit: { context: 200_000, output: 64_000 },
+        modalities: TEXT_IMAGE_PDF,
+        tool_call: true,
+    },
+    "claude-haiku-4.5": {
         name: "Claude Haiku 4.5",
         limit: { context: 200_000, output: 64_000 },
         modalities: { input: ["text", "image"], output: ["text"] },

package/dist/plugin.js

@@ -21,6 +21,14 @@ function discoveredProviderModels(cache) {
         model.id,
         {
             name: model.name ?? model.id,
+            ...(model.contextLimit || model.outputLimit
+                ? {
+                    limit: {
+                        context: model.contextLimit ?? 200_000,
+                        output: model.outputLimit ?? 64_000,
+                    },
+                }
+                : {}),
         },
     ]));
 }

package/dist/response-adapter.d.ts

@@ -1,5 +1,6 @@
 export interface KiroGenerateResponse {
     readonly text: string;
+    readonly reasoning?: string;
     readonly modelId?: string;
     readonly toolCalls?: ReadonlyArray<KiroToolCallChunk>;
     readonly usage?: {
@@ -20,6 +21,11 @@ export interface KiroToolCallChunk {
     readonly arguments: string;
     readonly modelId?: string;
 }
-export type KiroStreamEvent = KiroStreamChunk | KiroToolCallChunk;
+export interface KiroReasoningChunk {
+    readonly type: "reasoning";
+    readonly text: string;
+    readonly modelId?: string;
+}
+export type KiroStreamEvent = KiroStreamChunk | KiroToolCallChunk | KiroReasoningChunk;
 export declare function toOpenAIChatResponse(response: KiroGenerateResponse, model: string): Response;
 export declare function toOpenAIChatStreamResponse(chunks: AsyncIterable<KiroStreamEvent>, model: string): Response;

package/dist/response-adapter.js

@@ -11,6 +11,7 @@ export function toOpenAIChatResponse(response, model) {
                 message: {
                     role: "assistant",
                     content: toolCalls.length > 0 ? response.text || null : response.text,
+                    ...(response.reasoning ? { reasoning_content: response.reasoning } : {}),
                     ...(toolCalls.length > 0
                         ? {
                             tool_calls: toolCalls.map((toolCall, index) => ({
@@ -70,6 +71,11 @@ export function toOpenAIChatStreamResponse(chunks, model) {
                             ],
                         };
                     }
+                    else if (chunk.type === "reasoning") {
+                        if (!chunk.text)
+                            continue;
+                        delta = { reasoning_content: chunk.text };
+                    }
                     else {
                         if (!chunk.text)
                             continue;

package/docs/e2e-validation.md

@@ -152,7 +152,7 @@ Before publishing, verify that new model ids can be used without a code release:
         "modelAliases": {
           "new-model": "new-model-id"
         },
-        "modelDiscoveryCommand": ["kiro-cli", "models", "--json"]
+        "modelDiscoveryCommand": ["kiro-cli", "chat", "--list-models", "--format", "json"]
       }
     ]
   ]
@@ -161,7 +161,7 @@ Before publishing, verify that new model ids can be used without a code release:
 
 Checks:
 
-- If your installed `kiro-cli` has a model-list command, set `modelDiscoveryCommand` to that command and confirm discovered models appear in provider metadata.
+- Confirm `kiro-cli chat --list-models --format json` succeeds and discovered models appear in provider metadata.
 - The model appears in OpenCode provider metadata.
 - The alias resolves to the configured id.
 - Unknown models still pass through when `disableModelPassThrough` is false.

package/examples/opencode.jsonc

@@ -13,15 +13,15 @@
         // "agentMode": "vibe",
         "modelCacheTtlSeconds": 21600,
         "modelDiscovery": "auto",
-        "modelDiscoveryCommand": [],
+        "modelDiscoveryCommand": ["kiro-cli", "chat", "--list-models", "--format", "json"],
         "maxAttempts": 3,
         "requestTimeoutMs": 120000,
         "modelAliases": {
-          "sonnet": "claude-sonnet-4.6",
-          "opus": "claude-opus-4.6"
+          "sonnet": "claude-sonnet-5",
+          "opus": "claude-opus-4.8"
         },
         "extraModels": {
-          "claude-opus-4-9": {
+          "claude-opus-4.9": {
             "name": "Claude Opus 4.9",
             "limit": { "context": 1000000, "output": 64000 },
             "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] },
@@ -34,14 +34,5 @@
         "trustAllTools": false
       }
     ]
-  ],
-  "provider": {
-    "kiro": {
-      "models": {
-        "sonnet": {
-          "name": "Claude Sonnet via Kiro"
-        }
-      }
-    }
-  }
+  ]
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bogyie/opencode-kiro-plugin",
-  "version": "0.1.0",
+  "version": "0.2.2",
   "description": "OpenCode plugin for using Kiro as a provider adapter",
   "type": "module",
   "main": "dist/index.js",