Flexion - Custom Implementations Tracker

LOCAL_AWS_SETUP.md

@@ -0,0 +1,179 @@
+# Local Build & AWS Bedrock Setup
+
+Instructions for cloning, building, and running the Flexion fork of opencode with AWS Bedrock.
+
+## Prerequisites
+
+- [Bun](https://bun.sh) v1.3+
+- [AWS CLI](https://aws.amazon.com/cli/) v2
+- Git + SSH key configured for GitHub (with access to the `flexion` org)
+
+## Clone & Build
+
+```bash
+git clone git@github.com:flexion/opencode.git
+cd opencode
+
+# Switch to the Flexion customizations branch
+git checkout flex
+
+# Install dependencies
+# Note: if your global ~/.npmrc redirects to a private registry (e.g. CMS Artifactory),
+# override it so public packages resolve correctly:
+BUN_CONFIG_REGISTRY=https://registry.npmjs.org bun install
+
+# Build for your current platform only
+BUN_CONFIG_REGISTRY=https://registry.npmjs.org bun run --cwd packages/opencode build --single --skip-embed-web-ui
+```
+
+The binary will be at:
+- macOS ARM64: `packages/opencode/dist/opencode-darwin-arm64/bin/opencode`
+- macOS x64:   `packages/opencode/dist/opencode-darwin-x64/bin/opencode`
+- Linux ARM64:  `packages/opencode/dist/opencode-linux-arm64/bin/opencode`
+- Linux x64:    `packages/opencode/dist/opencode-linux-x64/bin/opencode`
+
+Verify the build:
+
+```bash
+./packages/opencode/dist/opencode-darwin-arm64/bin/opencode --version
+```
+
+## AWS Bedrock Setup
+
+### 1. Configure AWS SSO profile
+
+Add to `~/.aws/config`:
+
+```ini
+[profile AdministratorAccess]
+sso_start_url = <your-sso-start-url>
+sso_region = <your-sso-region>
+sso_account_id = <your-account-id>
+sso_role_name = AdministratorAccess
+region = <your-preferred-region>
+```
+
+### 2. Configure opencode for Bedrock
+
+Create `~/.config/opencode/opencode.json`:
+
+```json
+{
+  "$schema": "https://opencode.ai/config.json",
+  "model": "amazon-bedrock/us.anthropic.claude-sonnet-4-6",
+  "enabled_providers": ["amazon-bedrock"],
+  "plugin": [],
+  "provider": {
+    "amazon-bedrock": {
+      "options": {
+        "region": "<your-preferred-region>"
+      },
+      "models": {
+        "writer.palmyra-x4": {
+          "id": "us.writer.palmyra-x4-v1:0",
+          "name": "Writer Palmyra X4",
+          "tool_call": false,
+          "limit": { "context": 128000, "output": 8192 }
+        },
+        "writer.palmyra-x5": {
+          "id": "us.writer.palmyra-x5-v1:0",
+          "name": "Writer Palmyra X5",
+          "tool_call": false,
+          "limit": { "context": 1000000, "output": 8192 }
+        },
+        "deepseek.r1": {
+          "id": "us.deepseek.r1-v1:0",
+          "name": "DeepSeek R1",
+          "tool_call": false,
+          "limit": { "context": 64000, "output": 32768 }
+        },
+        "mistral.pixtral-large-2502": {
+          "id": "us.mistral.pixtral-large-2502-v1:0",
+          "name": "Mistral Pixtral Large",
+          "tool_call": false,
+          "limit": { "context": 128000, "output": 8192 }
+        },
+        "meta.llama4-maverick-17b-instruct": {
+          "id": "us.meta.llama4-maverick-17b-instruct-v1:0",
+          "name": "Meta Llama 4 Maverick 17B",
+          "tool_call": false,
+          "limit": { "context": 1000000, "output": 8192 }
+        },
+        "meta.llama4-scout-17b-instruct": {
+          "id": "us.meta.llama4-scout-17b-instruct-v1:0",
+          "name": "Meta Llama 4 Scout 17B",
+          "tool_call": false,
+          "limit": { "context": 10000000, "output": 8192 }
+        },
+        "amazon.nova-2-lite": {
+          "id": "us.amazon.nova-2-lite-v1:0",
+          "name": "Amazon Nova 2 Lite",
+          "limit": { "context": 300000, "output": 5120 }
+        }
+      }
+    }
+  }
+}
+```
+
+> **Note on `tool_call: false`:** Models marked with `tool_call: false` do not support
+> tool use in streaming mode on Bedrock. This config prevents opencode from sending
+> tool definitions to those models. The `flex` branch includes a fix that makes
+> `tool_call: false` actually respected at runtime — see the tracking PR
+> [flexion/opencode#2](https://github.com/flexion/opencode/pull/2) for details.
+
+### 3. Shell alias
+
+Add to `~/.zshrc` or `~/.bashrc`:
+
+```bash
+opencode-work() {
+  local profile="AdministratorAccess"
+  echo "Logging in to AWS SSO ($profile)..."
+  aws sso login --profile "$profile" || return 1
+  eval "$(aws configure export-credentials --profile "$profile" --format env)"
+  /path/to/opencode/packages/opencode/dist/opencode-darwin-arm64/bin/opencode "$@"
+}
+```
+
+Replace `/path/to/opencode` with where you cloned the repo (e.g. `~/Code/personal/flexion-work-items/flexchat-stack/opencode`).
+
+### 4. Usage
+
+```bash
+# Login and launch
+opencode-work
+
+# Re-running after SSO session expires — just run again, it will re-authenticate
+opencode-work
+```
+
+## Keeping the Fork Up to Date
+
+When upstream releases a new version, sync `dev` and rebase `flex`:
+
+```bash
+git fetch upstream  # upstream = https://github.com/anomalyco/opencode.git
+git checkout dev
+git reset --hard upstream/dev
+git push origin dev --force
+
+git checkout flex
+git rebase dev
+# Resolve any conflicts, then:
+git push origin flex --force
+```
+
+See [flexion/opencode#2](https://github.com/flexion/opencode/pull/2) for the full list of Flexion customizations and conflict resolution notes.
+
+## What's Different in This Fork (`flex` branch)
+
+| Change | File(s) | Description |
+|--------|---------|-------------|
+| Hide skill prompt text from chat UI | `packages/opencode/src/session/prompt.ts` | Marks skill template as `synthetic` so the full prompt is sent to the model but hidden from the user |
+| Respect `tool_call: false` at runtime | `packages/opencode/src/session/llm.ts` | Gates tool resolution behind `capabilities.toolcall` — fixes failures on Bedrock models that don't support streaming + tool use |
+| Local build & AWS Bedrock setup docs | `LOCAL_AWS_SETUP.md` | This file |
+
+Full details and upstream tracking: [flexion/opencode#2](https://github.com/flexion/opencode/pull/2)
+
+Upstream issue: [anomalyco/opencode#19966](https://github.com/anomalyco/opencode/issues/19966)

packages/opencode/src/session/llm.ts

@@ -193,7 +193,8 @@ export namespace LLM {
           },
         )
 
-        const tools = resolveTools(input)
+        const canTool = input.model.capabilities.toolcall
+        const tools = canTool ? resolveTools(input) : {}
 
         // LiteLLM and some Anthropic proxies require the tools parameter to be present
         // when message history contains tool calls, even if no tools are being used.
@@ -211,6 +212,7 @@ export namespace LLM {
         // during compaction), inject a stub tool to satisfy the validation requirement.
         // The stub description explicitly tells the model not to call it.
         if (
+          canTool &&
           (isLiteLLMProxy || input.model.providerID.includes("github-copilot")) &&
           Object.keys(tools).length === 0 &&
           hasToolCalls(input.messages)
@@ -350,9 +352,13 @@ export namespace LLM {
           topP: params.topP,
           topK: params.topK,
           providerOptions: ProviderTransform.providerOptions(input.model, params.options),
-          activeTools: Object.keys(tools).filter((x) => x !== "invalid"),
-          tools,
-          toolChoice: input.toolChoice,
+          ...(canTool
+            ? {
+                activeTools: Object.keys(tools).filter((x) => x !== "invalid"),
+                tools,
+              }
+            : {}),
+          ...(canTool ? { toolChoice: input.toolChoice } : {}),
           maxOutputTokens: params.maxOutputTokens,
           abortSignal: input.abort,
           headers: {

packages/opencode/src/session/prompt.ts

@@ -124,7 +124,7 @@ export namespace SessionPrompt {
 
       const resolvePromptParts = Effect.fn("SessionPrompt.resolvePromptParts")(function* (template: string) {
         const ctx = yield* InstanceState.context
-        const parts: PromptInput["parts"] = [{ type: "text", text: template }]
+        const parts: PromptInput["parts"] = [{ type: "text", text: template, synthetic: true }]
         const files = ConfigMarkdown.files(template)
         const seen = new Set<string>()
         yield* Effect.forEach(
@@ -1637,7 +1637,15 @@ NOTE: At any point in time through this workflow you should feel free to ask the
                 prompt: templateParts.find((y) => y.type === "text")?.text ?? "",
               },
             ]
-          : [...templateParts, ...(input.parts ?? [])]
+          : [
+              ...templateParts,
+              {
+                type: "text" as const,
+                text: `Running skill: ${input.command}`,
+                ignored: true,
+              },
+              ...(input.parts ?? []),
+            ]
 
         const userAgent = isSubtask ? (input.agent ?? (yield* agents.defaultAgent())) : agentName
         const userModel = isSubtask

packages/opencode/test/session/llm.test.ts

@@ -561,6 +561,95 @@ describe("session.llm.stream", () => {
     })
   })
 
+  test("does not send tools when model toolcall is disabled", async () => {
+    const server = state.server
+    if (!server) {
+      throw new Error("Server not initialized")
+    }
+
+    const providerID = "alibaba"
+    const modelID = "qwen-plus"
+    const fixture = await loadFixture(providerID, modelID)
+    const model = fixture.model
+
+    const request = waitRequest(
+      "/chat/completions",
+      new Response(createChatStream("Hello"), {
+        status: 200,
+        headers: { "Content-Type": "text/event-stream" },
+      }),
+    )
+
+    await using tmp = await tmpdir({
+      init: async (dir) => {
+        await Bun.write(
+          path.join(dir, "opencode.json"),
+          JSON.stringify({
+            $schema: "https://opencode.ai/config.json",
+            enabled_providers: [providerID],
+            provider: {
+              [providerID]: {
+                options: {
+                  apiKey: "test-key",
+                  baseURL: `${server.url.origin}/v1`,
+                },
+              },
+            },
+          }),
+        )
+      },
+    })
+
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const resolved = await getModel(ProviderID.make(providerID), ModelID.make(model.id))
+        const sessionID = SessionID.make("session-test-no-tools")
+        const agent = {
+          name: "test",
+          mode: "primary",
+          options: {},
+          permission: [{ permission: "*", pattern: "*", action: "allow" }],
+        } satisfies Agent.Info
+
+        const user = {
+          id: MessageID.make("user-no-tools"),
+          sessionID,
+          role: "user",
+          time: { created: Date.now() },
+          agent: agent.name,
+          model: { providerID: ProviderID.make(providerID), modelID: resolved.id },
+        } satisfies MessageV2.User
+
+        await drain({
+          user,
+          sessionID,
+          model: {
+            ...resolved,
+            capabilities: {
+              ...resolved.capabilities,
+              toolcall: false,
+            },
+          },
+          agent,
+          system: ["You are a helpful assistant."],
+          messages: [{ role: "user", content: "Hello" }],
+          tools: {
+            question: tool({
+              description: "Ask a question",
+              inputSchema: z.object({}),
+              execute: async () => ({ output: "" }),
+            }),
+          },
+        })
+
+        const capture = await request
+        expect(capture.body.tools).toBeUndefined()
+        expect(capture.body.tool_choice).toBeUndefined()
+      },
+    })
+  })
+
   test("sends responses API payload for OpenAI models", async () => {
     const server = state.server
     if (!server) {