@aiworker/web 1.22.0 → 1.23.0

package/server/lib/small-model/DOCUMENTATION.md

@@ -91,6 +91,10 @@ other runtime API.
   a blocker instead of a raw 500 message.
 - `call.js` — wire formats and per-provider auth, replicating OpenCode's
   plugin auth loaders:
+  - OpenCode-hosted providers receive `x-opencode-session`. Session-backed
+    features reuse the real OpenCode session id, walkthrough retries reuse the
+    walkthrough cache key, and standalone one-shot actions receive a fresh
+    opaque id for that generation.
   - **GitHub Copilot**: fetches the requested model's authenticated `/models`
     metadata from `https://api.githubcopilot.com` (or
     `copilot-api.<enterprise>`) and honors its advertised endpoint, preferring
@@ -113,8 +117,9 @@ other runtime API.
   - Everything else: OpenAI-compatible `/chat/completions` against the
     provider's base URL, resolved from (1) `provider.<id>.options.baseURL`
     in the OpenCode config, (2) the hardcoded `https://api.openai.com/v1`
-     endpoint, (3) the endpoint OpenCode resolved at runtime, or (4) the
-    provider's `api` field from the models.dev catalog. The credential follows
+     endpoint, (3) the selected model's endpoint OpenCode resolved at runtime,
+     (4) the provider-level runtime endpoint, or (5) the provider's `api` field
+     from the models.dev catalog. The credential follows
     the same shape: config `options.apiKey`, then the runtime credential, then
     the auth.json entry. `provider.<id>.options.headers` is sent with the
     request and overrides the bearer default, so gateways that authenticate on

package/server/lib/small-model/call.js

@@ -1,6 +1,7 @@
 import fs from 'fs';
 import os from 'os';
 import path from 'path';
+import { randomUUID } from 'node:crypto';
 import { readAuthFile, writeAuthFile } from '../opencode/auth.js';
 import { readConfig, readConfigLayers, isPlainObject } from '../opencode/shared.js';
 import { getCatalogProvider } from './catalog.js';
@@ -17,7 +18,7 @@ const COPILOT_MODELS_TIMEOUT_MS = 5_000;
 // Qwen, …) spend part of this budget on reasoning before the actual answer.
 const DEFAULT_MAX_OUTPUT_TOKENS = 4_000;
 
-const USER_AGENT = 'opencode/1.0 openchamber';
+const USER_AGENT = 'opencode/1.0 aiworker';
 
 const mergeHeadersCaseInsensitive = (base, overrides) => {
   const merged = { ...base };
@@ -605,6 +606,8 @@ const readProviderConfig = (workingDirectory, providerID) => {
   }
 }
 
+const getRuntimeModel = (runtimeProvider, modelID) => runtimeProvider?.models?.get(modelID) ?? null;
+
 // ---------------------------------------------------------------------------
 // Dispatch
 // ---------------------------------------------------------------------------
@@ -645,10 +648,11 @@ export async function resolveProviderLogin({ auth, workingDirectory, providerID
     || null;
 }
 
-export async function callSmallModel({ auth, catalog, workingDirectory, providerID, modelID, prompt, system, maxOutputTokens, responseSchema, timeoutMs, signal }) {
+export async function callSmallModel({ auth, catalog, workingDirectory, sessionID, providerID, modelID, prompt, system, maxOutputTokens, responseSchema, timeoutMs, signal }) {
   const tokens = Number(maxOutputTokens) > 0 ? Number(maxOutputTokens) : DEFAULT_MAX_OUTPUT_TOKENS;
   const providerConfig = readProviderConfig(workingDirectory, providerID);
   const runtimeProvider = await getRuntimeProvider(providerID);
+  const runtimeModel = getRuntimeModel(runtimeProvider, modelID);
   // Match OpenCode's resolveSDK precedence: config `provider.<id>.options`
   // wins, then what OpenCode itself resolved at runtime (the only place a
   // plugin's credential exists), and the auth.json entry last.
@@ -758,9 +762,10 @@ export async function callSmallModel({ auth, catalog, workingDirectory, provider
   // base URL for that provider (openai itself included). When a custom provider
   // is not in the catalog (e.g. a user-configured OpenAI-compatible proxy),
   // fall back to its baseURL from the OpenCode provider config, then to the
-  // endpoint OpenCode resolved at runtime — which for a plugin provider is the
-  // only place it exists, and for several of them is a local proxy the plugin
-  // itself runs. The openai provider also respects
+  // selected model's endpoint OpenCode resolved at runtime, then to the
+  // provider-level runtime endpoint. For a plugin provider, the runtime listing
+  // is the only place those endpoints exist, and several are local proxies the
+  // plugin itself runs. The openai provider also respects
   // provider.openai.options.baseURL — OpenCode itself uses the same config for
   // all providers including openai.
   const provider = getCatalogProvider(catalog, providerID);
@@ -770,7 +775,8 @@ export async function callSmallModel({ auth, catalog, workingDirectory, provider
     ? providerConfigUrl
     : providerID === 'openai'
       ? defaultOpenaiUrl
-      : runtimeProvider?.baseURL
+      : runtimeModel?.api?.url
+        ?? runtimeProvider?.baseURL
         ?? (typeof provider?.api === 'string' && provider.api
           ? provider.api
           : null);
@@ -795,7 +801,12 @@ export async function callSmallModel({ auth, catalog, workingDirectory, provider
     baseURL,
     // Configured headers last: a gateway that authenticates on its own header
     // must be able to override the bearer default rather than sit beside it.
-    headers: mergeHeadersCaseInsensitive({ Authorization: `Bearer ${apiKey}` }, providerConfig?.headers),
+    headers: mergeHeadersCaseInsensitive(
+      mergeHeadersCaseInsensitive({ Authorization: `Bearer ${apiKey}` }, providerConfig?.headers),
+      providerID.startsWith('opencode')
+        ? { 'x-opencode-session': typeof sessionID === 'string' && sessionID.trim() ? sessionID.trim() : randomUUID() }
+        : null,
+    ),
     modelID,
     prompt,
     system,

package/server/lib/small-model/call.test.js

@@ -69,8 +69,8 @@ describe('callSmallModel — custom provider config', () => {
   afterEach(() => {
     globalThis.fetch = originalFetch;
     vi.restoreAllMocks();
-    delete process.env.OPENCHAMBER_TEST_PROVIDER_KEY;
-    delete process.env.OPENCHAMBER_TEST_GATEWAY_KEY;
+    delete process.env.AIWORKER_TEST_PROVIDER_KEY;
+    delete process.env.AIWORKER_TEST_GATEWAY_KEY;
   });
 
   describe('config-supplied credentials (no auth.json entry)', () => {
@@ -104,11 +104,11 @@ describe('callSmallModel — custom provider config', () => {
     });
 
     it('resolves an OpenCode environment variable before sending the API key', async () => {
-      process.env.OPENCHAMBER_TEST_PROVIDER_KEY = 'sk-env-key';
+      process.env.AIWORKER_TEST_PROVIDER_KEY = 'sk-env-key';
       readConfig.mockReturnValue({
         provider: {
           custom: {
-            options: { apiKey: '{env:OPENCHAMBER_TEST_PROVIDER_KEY}', baseURL: 'https://proxy.example.test/v1' },
+            options: { apiKey: '{env:AIWORKER_TEST_PROVIDER_KEY}', baseURL: 'https://proxy.example.test/v1' },
           },
         },
       });
@@ -127,7 +127,7 @@ describe('callSmallModel — custom provider config', () => {
     });
 
     it('sends configured provider headers alongside the bearer token', async () => {
-      process.env.OPENCHAMBER_TEST_GATEWAY_KEY = 'sub-key';
+      process.env.AIWORKER_TEST_GATEWAY_KEY = 'sub-key';
       readConfig.mockReturnValue({
         provider: {
           custom: {
@@ -135,7 +135,7 @@ describe('callSmallModel — custom provider config', () => {
               apiKey: 'sk-config',
               baseURL: 'https://proxy.example.test/v1',
               headers: {
-                'Ocp-Apim-Subscription-Key': '{env:OPENCHAMBER_TEST_GATEWAY_KEY}',
+                'Ocp-Apim-Subscription-Key': '{env:AIWORKER_TEST_GATEWAY_KEY}',
                 'x-tenant': 'team',
               },
             },
@@ -483,6 +483,29 @@ describe('callSmallModel — custom provider config', () => {
   });
 
   describe('catalog-based base URL (no config override)', () => {
+    it('identifies OpenCode Go requests with the owning conversation', async () => {
+      readConfig.mockReturnValue({});
+      fetchMock.mockResolvedValue(ok('ok'));
+
+      await callSmallModel({
+        auth: { 'opencode-go': { type: 'api', key: 'go-key' } },
+        catalog: {
+          'opencode-go': {
+            id: 'opencode-go',
+            api: 'https://opencode.ai/zen/go/v1',
+            models: { utility: { id: 'utility' } },
+          },
+        },
+        workingDirectory: '/proj',
+        sessionID: 'ses_conversation',
+        providerID: 'opencode-go',
+        modelID: 'utility',
+        prompt: 'hi',
+      });
+
+      expect(lastCall(fetchMock).init.headers['x-opencode-session']).toBe('ses_conversation');
+    });
+
     it('uses the catalog api field when no config baseURL is set', async () => {
       readConfig.mockReturnValue({});
       fetchMock.mockResolvedValue(ok('ok'));
@@ -542,6 +565,35 @@ describe('callSmallModel — custom provider config', () => {
       expect(init.headers.Authorization).toBe('Bearer plugin-key');
     });
 
+    it('uses the selected runtime model endpoint', async () => {
+      readConfig.mockReturnValue({});
+      getRuntimeProvider.mockResolvedValue({
+        id: 'runtime-provider',
+        apiKey: 'plugin-key',
+        baseURL: 'https://runtime-provider/v1beta',
+        models: new Map([
+          ['first-model', { api: { url: 'https://runtime-provider/v1beta', npm: '@ai-sdk/google' } }],
+          ['selected-model', { api: { url: 'https://runtime-provider/v1', npm: '@ai-sdk/openai' } }],
+        ]),
+        anonymousZen: false,
+      });
+      fetchMock.mockResolvedValue(ok('done'));
+
+      await callSmallModel({
+        auth: {},
+        catalog: {},
+        workingDirectory: '/proj',
+        providerID: 'runtime-provider',
+        modelID: 'selected-model',
+        prompt: 'hi',
+      });
+
+      const { url, init } = lastCall(fetchMock);
+      expect(url).toBe('https://runtime-provider/v1/chat/completions');
+      expect(url).not.toContain('/v1beta');
+      expect(JSON.parse(init.body).model).toBe('selected-model');
+    });
+
     it('keeps the ChatGPT-plan login on its own transport instead of the runtime key', async () => {
       readConfig.mockReturnValue({});
       // OpenCode reports an OAuth access token as `options.apiKey` for openai;

package/server/lib/small-model/index.js

@@ -6,6 +6,7 @@ import { readConfigLayers } from '../opencode/shared.js';
 import { getModelCatalog } from './catalog.js';
 import { resolveSmallModel, parseModelRef, isUsableAuthEntry, getAuthEntryForProvider } from './resolve.js';
 import { DEDICATED_WIRE_FORMAT_PROVIDERS, callSmallModel, resolveProviderLogin } from './call.js';
+import { readMergedSettingsSync } from '../opencode/settings-files.js';
 import { getRuntimeProviderSnapshot } from './runtime-providers.js';
 
 // Never a small model, whatever the transport looks like. A plugin can publish
@@ -16,26 +17,20 @@ import { getRuntimeProviderSnapshot } from './runtime-providers.js';
 // conditional on an endpoint existing.
 const CLAUDE_CODE_PROVIDER = 'claude-code';
 
-const OPENCHAMBER_SETTINGS_FILE = path.join(
-  process.env.OPENCHAMBER_DATA_DIR
-    ? path.resolve(process.env.OPENCHAMBER_DATA_DIR)
-    : path.join(os.homedir(), '.config', 'openchamber'),
+const AIWORKER_SETTINGS_FILE = path.join(
+  process.env.AIWORKER_DATA_DIR
+    ? path.resolve(process.env.AIWORKER_DATA_DIR)
+    : path.join(os.homedir(), '.config', 'aiworker'),
   'settings.json',
 );
 
-// OpenChamber's own settings: when the user unchecks "use default small model"
+// AiWorker's own settings: when the user unchecks "use default small model"
 // their explicit override outranks every other resolution step.
 const readSmallModelSettingsOverride = () => {
-  try {
-    const raw = fs.readFileSync(OPENCHAMBER_SETTINGS_FILE, 'utf8');
-    const settings = JSON.parse(raw);
-    if (!settings || typeof settings !== 'object') return null;
-    if (settings.smallModelUseDefault !== false) return null;
-    const override = typeof settings.smallModelOverride === 'string' ? settings.smallModelOverride.trim() : '';
-    return override || null;
-  } catch {
-    return null;
-  }
+  const settings = readMergedSettingsSync({ fs, path, settingsFilePath: AIWORKER_SETTINGS_FILE });
+  if (settings.smallModelUseDefault !== false) return null;
+  const override = typeof settings.smallModelOverride === 'string' ? settings.smallModelOverride.trim() : '';
+  return override || null;
 };
 
 // Rough safety clamp so a huge input never blows the model's context window.
@@ -102,7 +97,7 @@ const readConfiguredSmallModel = (workingDirectory) => {
  * Generates text with the user's small model, resolved and authenticated
  * entirely server-side from the OpenCode config and auth store.
  */
-export async function generateSmallModelText({ prompt, system, maxOutputTokens, model, directory, preferredProviderID, preferredModelID, restrictToPreferredProvider = false, responseSchema, timeoutMs, signal, onOverflow = 'truncate' }) {
+export async function generateSmallModelText({ prompt, system, maxOutputTokens, model, directory, sessionID, preferredProviderID, preferredModelID, restrictToPreferredProvider = false, responseSchema, timeoutMs, signal, onOverflow = 'truncate' }) {
   if (typeof prompt !== 'string' || !prompt.trim()) {
     throw Object.assign(new Error('prompt is required'), { statusCode: 400 });
   }
@@ -169,6 +164,7 @@ export async function generateSmallModelText({ prompt, system, maxOutputTokens,
     auth,
     catalog,
     workingDirectory: directory,
+    sessionID,
     providerID: resolved.providerID,
     modelID: resolved.modelID,
     prompt: clamped.prompt,

package/server/lib/small-model/index.test.js

@@ -7,7 +7,7 @@ import { afterAll, afterEach, beforeEach, describe, expect, it, vi } from 'vites
 // this the suite would resolve whatever small model the developer running it
 // happens to have configured.
 const TEMP_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), 'small-model-settings-'));
-process.env.OPENCHAMBER_DATA_DIR = TEMP_DATA_DIR;
+process.env.AIWORKER_DATA_DIR = TEMP_DATA_DIR;
 
 vi.mock('../opencode/auth.js', () => ({ readAuthFile: vi.fn() }));
 vi.mock('../opencode/shared.js', () => ({

package/server/lib/small-model/resolve.js

@@ -76,7 +76,7 @@ const pickWithinProvider = (providerID, auth, catalog, family) => {
 };
 
 export function resolveSmallModel({ auth, catalog, settingsSmallModel, configSmallModel, preferredProviderID, preferredModelID }) {
-  // OpenChamber's own setting (Settings → Sessions → Small Model override)
+  // AiWorker's own setting (Settings → Sessions → Small Model override)
   // outranks everything, including the OpenCode config.
   const fromSettings = parseModelRef(settingsSmallModel);
   if (fromSettings) {

package/server/lib/small-model/resolve.test.js

@@ -57,7 +57,7 @@ describe('isUsableAuthEntry', () => {
 });
 
 describe('resolveSmallModel', () => {
-  it('gives the OpenChamber settings override top priority', () => {
+  it('gives the AiWorker settings override top priority', () => {
     const result = resolveSmallModel({
       auth: { anthropic: { type: 'api', key: 'sk-x' } },
       catalog,

package/server/lib/small-model/routes.js

@@ -21,13 +21,14 @@ export function registerSmallModelRoutes(app, { getSmallModelService }) {
   app.post('/api/small-model/generate', async (req, res) => {
     try {
       const { generateSmallModelText } = await getSmallModelService();
-      const { prompt, system, maxOutputTokens, model, directory, preferredProviderID, preferredModelID, restrictToPreferredProvider } = req.body || {};
+      const { prompt, system, maxOutputTokens, model, directory, sessionID, preferredProviderID, preferredModelID, restrictToPreferredProvider } = req.body || {};
       const result = await generateSmallModelText({
         prompt,
         system,
         maxOutputTokens,
         model,
         directory,
+        sessionID,
         preferredProviderID,
         preferredModelID,
         restrictToPreferredProvider: restrictToPreferredProvider === true,

package/server/lib/small-model/runtime-providers.js

@@ -8,7 +8,8 @@
 //
 // `GET /provider` is where that state becomes visible. It reports, per
 // provider, the resolved `options.baseURL` and `options.apiKey`, and per model
-// the wire adapter (`api.npm`) and endpoint (`api.url`).
+// the wire adapter (`api.npm`) and endpoint (`api.url`). The provider-level
+// endpoint remains only as a fallback when the selected model has no endpoint.
 //
 // What it does NOT report is `options.fetch`. OpenCode strips functions from
 // the response, and a plugin is free to put its whole protocol in there:
@@ -61,8 +62,8 @@ export function resetOpenCodeRuntimeProviders() {
 /**
  * The boundary. Everything the `/provider` payload claims is checked here, so
  * the rest of this module and its callers work with settled values:
- * a credential we may use, an endpoint, and whether the provider is the
- * anonymous zen case.
+ * a credential we may use, provider and model endpoints, and whether the
+ * provider is the anonymous zen case.
  *
  * The credential deliberately prefers `options.apiKey` over the `key` field:
  * for a plugin provider the former is what its auth loader produced and what
@@ -82,13 +83,22 @@ function parseProviderListing(payload) {
     const id = text(record(raw).id);
     if (!id) continue;
     const options = record(record(raw).options);
-    const firstModel = record(Object.values(record(record(raw).models))[0]);
+    const models = new Map();
+    for (const [modelID, rawModel] of Object.entries(record(record(raw).models))) {
+      const model = record(rawModel);
+      const api = record(model.api);
+      const modelURL = endpoint(api.url);
+      const modelNpm = text(api.npm);
+      models.set(modelID, { api: { url: modelURL, npm: modelNpm } });
+    }
+    const firstModel = models.values().next().value;
     const declaredKey = text(options.apiKey);
     providers.set(id, {
       id,
       source: text(record(raw).source),
       apiKey: declaredKey === ZEN_ANONYMOUS_API_KEY ? null : (declaredKey ?? text(record(raw).key)),
-      baseURL: endpoint(options.baseURL) ?? endpoint(record(firstModel.api).url),
+      baseURL: endpoint(options.baseURL) ?? firstModel?.api?.url ?? null,
+      models,
       // True only for the zen-without-login case: a provider that is present
       // and usable through OpenCode, but that we must not call ourselves.
       anonymousZen: declaredKey === ZEN_ANONYMOUS_API_KEY,
@@ -152,5 +162,3 @@ export async function getRuntimeProvider(providerID) {
   const current = await getRuntimeProviderSnapshot();
   return current?.providers.get(providerID) ?? null;
 }
-
-

package/server/lib/small-model/runtime-providers.test.js

@@ -14,7 +14,10 @@ const providerPayload = (overrides = {}) => ({
       id: 'llmapi',
       source: 'config',
       options: { apiKey: 'plugin-key', baseURL: 'https://api.llmapi.ai/v1/' },
-      models: { 'claude-opus-4-8': { api: { id: 'claude-opus-4-8', url: '', npm: '@ai-sdk/anthropic' } } },
+      models: {
+        'claude-opus-4-8': { api: { id: 'claude-opus-4-8', url: '', npm: '@ai-sdk/anthropic' } },
+        'gpt-5.6-luna': { api: { id: 'gpt-5.6-luna', url: 'https://api.llmapi.ai/v1', npm: '@ai-sdk/openai' } },
+      },
     },
     {
       id: 'opencode',
@@ -59,6 +62,9 @@ describe('OpenCode runtime provider snapshot', () => {
     const provider = await getRuntimeProvider('llmapi');
 
     expect(provider).toMatchObject({ apiKey: 'plugin-key', baseURL: 'https://api.llmapi.ai/v1' });
+    expect(provider.models.get('gpt-5.6-luna')).toEqual({
+      api: { url: 'https://api.llmapi.ai/v1', npm: '@ai-sdk/openai' },
+    });
     expect(fetchMock.mock.calls[0][0]).toBe('http://127.0.0.1:4096/provider');
     expect(fetchMock.mock.calls[0][1].headers).toMatchObject({ Authorization: 'Basic test' });
   });

package/server/lib/system-prompt/runtime.js

@@ -1,5 +1,5 @@
-import { parse as parseJsonc } from 'jsonc-parser';
 import { pathToFileURL } from 'node:url';
+import { appendManagedPlugin } from '../opencode/managed-plugin-config.js';
 
 const PROVIDER_PROMPT_BOUNDARY = 'You are powered by the model named';
 const MINIMAL_IDENTITY = 'You are OpenCode, a coding agent.';
@@ -9,7 +9,7 @@ const PROVIDER_PROMPT_BOUNDARY = ${JSON.stringify(PROVIDER_PROMPT_BOUNDARY)}
 const MINIMAL_IDENTITY = ${JSON.stringify(MINIMAL_IDENTITY)}
 const optimizedSessions = new Map()
 
-export const OpenChamberSystemPromptPlugin = async () => ({
+export const AiWorkerSystemPromptPlugin = async () => ({
   "chat.message": async (input, output) => {
     if (!input.sessionID) return
     const agent = output?.message?.agent ?? input.agent
@@ -33,25 +33,6 @@ export const OpenChamberSystemPromptPlugin = async () => ({
 })
 `;
 
-const mergePluginConfig = (rawConfig, pluginUrl) => {
-  const errors = [];
-  const parsed = typeof rawConfig === 'string' && rawConfig.trim()
-    ? parseJsonc(rawConfig, errors, { allowTrailingComma: true })
-    : {};
-  if (errors.length > 0 || !parsed || typeof parsed !== 'object' || Array.isArray(parsed)) {
-    throw new Error('OPENCODE_CONFIG_CONTENT must contain a valid JSON object before OpenChamber can inject its system prompt optimizer');
-  }
-  if (parsed.plugin !== undefined && !Array.isArray(parsed.plugin)) {
-    throw new Error('OPENCODE_CONFIG_CONTENT plugin must be an array before OpenChamber can inject its system prompt optimizer');
-  }
-  const configured = Array.isArray(parsed.plugin) ? parsed.plugin : [];
-  parsed.plugin = [
-    ...configured.filter((value) => value !== pluginUrl && (!Array.isArray(value) || value[0] !== pluginUrl)),
-    pluginUrl,
-  ];
-  return JSON.stringify(parsed);
-};
-
 export const createSystemPromptRuntime = ({ fsPromises, path, dataDir }) => {
   const pluginDirectory = path.join(dataDir, 'system-prompt');
   const pluginPath = path.join(pluginDirectory, 'openchamber-system-prompt-plugin.js');
@@ -60,7 +41,7 @@ export const createSystemPromptRuntime = ({ fsPromises, path, dataDir }) => {
     await fsPromises.mkdir(pluginDirectory, { recursive: true });
     await fsPromises.writeFile(pluginPath, createPluginSource(), { mode: 0o600 });
     return {
-      OPENCODE_CONFIG_CONTENT: mergePluginConfig(rawConfig, pathToFileURL(pluginPath).href),
+      OPENCODE_CONFIG_CONTENT: appendManagedPlugin(rawConfig, pathToFileURL(pluginPath).href, 'system prompt optimizer'),
     };
   };
 

package/server/lib/system-prompt/runtime.test.js

@@ -25,7 +25,7 @@ describe('managed system prompt runtime', () => {
     expect(config.model).toBe('test/model');
     expect(config.plugin).toEqual(['file:///existing.js', pathToFileURL(pluginPath).href]);
     const pluginModule = await import(`${pathToFileURL(pluginPath).href}?test=${Date.now()}`);
-    const hooks = await pluginModule.OpenChamberSystemPromptPlugin();
+    const hooks = await pluginModule.AiWorkerSystemPromptPlugin();
     const output = {
       system: ['Behavioral prompt\nYou are powered by the model named GPT.\n<env>kept</env>'],
     };
@@ -46,7 +46,7 @@ describe('managed system prompt runtime', () => {
     await runtime.prepareManagedOpenCodeEnv('{}');
     const pluginPath = path.join(dataDir, 'system-prompt', 'openchamber-system-prompt-plugin.js');
     const pluginModule = await import(`${pathToFileURL(pluginPath).href}?test=${Date.now()}`);
-    const hooks = await pluginModule.OpenChamberSystemPromptPlugin();
+    const hooks = await pluginModule.AiWorkerSystemPromptPlugin();
     const output = { system: ['Unrecognized prompt'] };
     await hooks['chat.message']({ sessionID: 'session-1', agent: 'plan' });
     await hooks['experimental.chat.system.transform']({ sessionID: 'session-1' }, output);
@@ -60,7 +60,7 @@ describe('managed system prompt runtime', () => {
     await runtime.prepareManagedOpenCodeEnv('{}');
     const pluginPath = path.join(dataDir, 'system-prompt', 'openchamber-system-prompt-plugin.js');
     const pluginModule = await import(`${pathToFileURL(pluginPath).href}?test=${Date.now()}`);
-    const hooks = await pluginModule.OpenChamberSystemPromptPlugin();
+    const hooks = await pluginModule.AiWorkerSystemPromptPlugin();
     const output = {
       system: ['Custom agent prompt\nYou are powered by the model named GPT.\n<env>kept</env>'],
     };

package/server/lib/terminal/DOCUMENTATION.md

@@ -2,14 +2,14 @@
 
 ## Ownership
 
-`runtime.js` owns terminal identity, PTY processes, status, ordered output, bounded scrollback, WebSocket attachments, and lifecycle routes. `shells.js` discovers executable shell families and resolves the persisted shell ID without accepting command strings or arguments. Clients own tab arrangement and choose stable terminal IDs. Electron uses this same runtime in-process; VS Code returns an explicit unsupported error.
+`runtime.js` owns terminal identity, PTY processes, launch mode, session purpose, ordered output, bounded scrollback, WebSocket attachments, and lifecycle routes. `shells.js` discovers executable shell families, resolves the persisted shell ID, and builds the per-shell argv for interactive versus command launches. Clients own tab arrangement. Interactive terminals use stable IDs; project actions keep a stable UI tab and allocate a fresh terminal ID for each command execution. Electron uses this same runtime in-process; VS Code returns an explicit unsupported error.
 
 ## Protocol
 
 `/api/terminal/ws` is the only terminal data transport. It uses v3 binary JSON control frames and is opened through `openRuntimeWebSocket`, preserving direct, Electron proxy, URL-token authentication, and private-relay routing.
 
 - `attach` registers a connection for one terminal. One socket may attach to many terminals.
-- Every attach and reconnect begins with an authoritative `snapshot` containing bounded history and the current sequence.
+- Every attach and reconnect begins with an authoritative `snapshot` containing bounded history, the current sequence, and the PTY `cols`/`rows` the history was drawn for. The client replays the history at that size before fitting its viewport; replaying shell output at another width leaves stray fragments that the shell's own SIGWINCH redraw never clears.
 - A current socket that closes or errors before its initial `open` invalidates its URL-scoped auth token before retrying, so retries mint a fresh token instead of backing off against a rejected upgrade. Hidden or offline clients wait 60 seconds and wake promptly on visibility/online recovery.
 - `output`, `exit`, and `restarted` carry monotonically increasing per-terminal sequences. Output carries raw live bytes plus replay-safe bytes with terminal query exchanges removed.
 - Attach registers before capturing the snapshot, buffers concurrent events, drops events represented by the snapshot sequence, then enters live delivery.
@@ -19,21 +19,25 @@
 
 HTTP remains the authenticated command plane for create, resize, appearance updates, restart, close, and force-kill. There is no SSE output or HTTP input compatibility path.
 
-`GET /api/terminal/sessions` enumerates live sessions (optionally filtered by resolved `cwd`) so clients can adopt terminals their local tab projection does not know about — another device, a new browser tab, or cleared storage. `POST /api/terminal/touch` refreshes `lastActivity` for the listed session ids; open clients call it periodically so background tabs, which hold no WebSocket attachment, are not idle-reaped while a client still shows them.
+`GET /api/terminal/sessions` enumerates live sessions (optionally filtered by resolved `cwd`) so clients can adopt terminals their local tab projection does not know about — another device, a new browser tab, or cleared storage. Listings include the effective launch mode and the normalized purpose, but never the command text. `POST /api/terminal/touch` refreshes `lastActivity` for the listed session ids; open clients call it periodically so background tabs, which hold no WebSocket attachment, are not idle-reaped while a client still shows them.
 
 ## PTY Lifecycle
 
 - IDs are client-provided or generated with `randomUUID()`.
-- Concurrent creates for one ID are single-flight only when working directory and shell preference match. Existing IDs cannot be reused for another working directory.
+- Create defaults to interactive mode. Command mode requires a non-empty trimmed command no longer than the terminal input limit and launches the shell so the PTY exits when that command exits. Each session also carries a normalized purpose. Omitted purpose means `{ type: 'terminal' }`. Project actions use `{ type: 'project-action', actionId, executionId }`, and the server validates both IDs as non-empty bounded strings. Create responses and attach snapshots echo the effective mode and purpose.
+- Concurrent creates for one ID are single-flight only when working directory, shell preference, login mode, launch mode, and session purpose match. Command-mode creates must also match the command text, unless the purpose is a project action that is already running for the same resolved `(cwd, actionId)` pair. In that case the runtime returns the existing session and its existing execution identity, even when another client requested a different session ID. Existing IDs cannot be reused for another working directory or another purpose.
 - Dimensions are bounded to 1-1000 columns and 1-500 rows; input is capped at 64 KiB.
 - A client may create before its renderer has mounted. It derives an initial size from the container and font metrics (falling back to 80x24 when unavailable), then sends a resize once Ghostty reports its final dimensions. This allows shell startup and renderer initialization to overlap.
-- PTY children explicitly clear `NODE_CHANNEL_FD`; daemon IPC descriptors are host-private and invalid after PTY descriptor cleanup.
-- PTY children also strip AppImage `ARGV0` (and other host-private shell vars such as `ELECTRON_RUN_AS_NODE`, `BASH_ENV`, `ENV`, `BASH_XTRACEFD`). An exported `ARGV0` makes zsh rewrite argv[0] for every external command, which breaks Python venv detection and other argv[0]/$0 consumers while leaving `/proc/self/exe` correct. On Linux, PTY spawn is wrapped with `env -u ARGV0` because `bun-pty` merges the native OS environ and would otherwise reintroduce `ARGV0` after a JS-only delete.
-- `GET /api/terminal/shells` reports shell IDs available on the active server using the same augmented PATH provided to spawned PTYs, plus whether each executable has a supported login-mode argument. `auto` preserves environment/platform fallback order; an explicit unavailable shell fails creation instead of silently running a different shell. Login mode is opt-in and uses only built-in arguments for known shells. Preference changes affect new sessions and explicit restarts, not running PTYs.
-- PTY data and exit callbacks enter one FIFO queue. Stale callbacks from replaced processes are ignored.
+- PTY children never inherit `NODE_CHANNEL_FD`; daemon IPC descriptors are host-private and invalid after PTY descriptor cleanup, and an inherited value (even empty) makes Node CLIs such as opencode print `Failed to parse IPC channel number` on exit.
+- PTY children also strip AppImage `ARGV0` (and other host-private shell vars such as `ELECTRON_RUN_AS_NODE`, `BASH_ENV`, `ENV`, `BASH_XTRACEFD`). An exported `ARGV0` makes zsh rewrite argv[0] for every external command, which breaks Python venv detection and other argv[0]/$0 consumers while leaving `/proc/self/exe` correct. On POSIX, PTY spawn is wrapped with `env -u ARGV0 -u NODE_CHANNEL_FD` (`resolvePosixPtyLaunch`) because `bun-pty` merges the native OS environ and would otherwise reintroduce both after a JS-only delete.
+- `GET /api/terminal/shells` reports shell IDs available on the active server using the same augmented PATH provided to spawned PTYs, plus whether each executable has a supported login-mode argument. `auto` preserves environment/platform fallback order; an explicit unavailable shell fails creation instead of silently running a different shell. Login mode is opt-in and uses only built-in arguments for known shells. Interactive shells still launch as before. Command-mode launches reuse the same environment and login support, but switch argv by shell family: POSIX and Fish use interactive `-c`, Nushell uses `-c`, PowerShell uses `-Command`, and cmd uses `/d /s /c`. Preference changes affect new sessions and explicit restarts, not running PTYs.
+- PTY data and exit callbacks enter one FIFO queue. The runtime wires those listeners in the same synchronous turn that receives the PTY object. `node-pty` and `bun-pty` both expose the PTY before dispatching registered callbacks. If a backend emitted exit before listener registration, this layer could not recover it, so the wiring stays adjacent to PTY creation.
 - Scrollback is retained on the server and capped at 512 KiB with UTF-8-safe trimming. Device-status, device-attribute, cursor-position reply, and color-query exchanges are removed from replay history with incomplete control sequences carried across PTY chunks; live output remains byte-for-byte unchanged.
-- Exited sessions remain attachable until explicit close or idle cleanup.
-- Restarts are serialized per terminal. Each restart spawns and wires the replacement before terminating the old process, retaining the terminal ID.
+- Exited sessions remain attachable until explicit close, idle cleanup, or a successful replacement of the same project action. Creating a replacement retires only exited records for the same resolved directory and action, after the new PTY starts. Failed creation preserves the old record and output. These replaced records do not exhaust the terminal capacity limit.
+- Deduplicated create responses may describe another client's execution. Cancellation cleanup closes only the terminal ID allocated for the cancelled request; it never closes an adopted peer execution.
+- Create and restart validate the working directory with a real `stat` and answer HTTP 400 `Invalid working directory` when it is not a directory. When the path does not exist at all (`ENOENT`/`ENOTDIR`, a worktree deleted outside AiWorker) the body also carries `code: "TERMINAL_CWD_MISSING"`. The client shows the failure without moving the session. The runtime never substitutes a parent directory on its own.
+- Restarts are serialized per terminal. Each restart spawns and wires the replacement before terminating the old process, retaining the terminal ID. Command-mode sessions reject restart with HTTP 400 instead of silently turning into interactive shells with stale action metadata.
+- A delete that arrives while create is still pending leaves a cancellation tombstone. When the PTY arrives, the runtime terminates it immediately, never inserts the session into the live map, and returns a create error while the delete still succeeds.
 - Close uses SIGTERM with bounded SIGKILL escalation. Force-kill, idle cleanup, and runtime shutdown terminate process groups immediately where supported. Removal explicitly sends a fatal scoped closure and evicts client projections even when a PTY backend fails to emit `onExit`; attached terminals are not considered idle.
 
 ## Security And Relay