@outputai/cli 0.10.1-dev.b7b2fbe.0 → 0.10.1-next.14a191e.0

package/dist/services/workflow_history.spec.js

@@ -1,6 +1,6 @@
 import { describe, it, expect, vi, beforeEach } from 'vitest';
 import { getWorkflowIdHistory } from '#api/generated/api.js';
-import { fetchWorkflowHistory } from '#services/workflow_history.js';
+import { fetchWorkflowHistory, fetchWorkflowHistoryUpdates } from '#services/workflow_history.js';
 vi.mock('#api/generated/api.js', () => ({ getWorkflowIdHistory: vi.fn() }));
 // The real correlator runs (only the API client is mocked), so spans are derived
 // from the events below — letting us assert duration/offset behaviour end to end.
@@ -48,6 +48,39 @@ describe('fetchWorkflowHistory', () => {
         expect(result.totalDurationMs).toBe(300_000); // closeTime - startTime
         expect(result.spans).toHaveLength(1);
         expect(result.spans[0].durationMs).toBe(27_000);
+        // Even though the server's own nextPageToken for the final page is empty (nothing more
+        // buffered), the cursor keeps the *request* token that reached it — a genuinely resumable
+        // position for a follow-up fetchWorkflowHistoryUpdates call. See fetchPages.
+        expect(result.cursor.pageToken).toBe('token-2');
+    });
+    it('a follow-up fetchWorkflowHistoryUpdates call resumes from fetchWorkflowHistory\'s cursor instead of re-paging from page 1', async () => {
+        mockGet
+            .mockResolvedValueOnce(page({
+            workflow: { workflowId: 'wf-123', runId: 'run-456', status: 'running', startTime: at(0) },
+            runId: 'run-456',
+            events: [workflowStarted(0), scheduled('2', 'wf#step', 0)],
+            nextPageToken: 'token-2'
+        }))
+            .mockResolvedValueOnce(page({
+            workflow: null,
+            runId: 'run-456',
+            events: [started('3', '2', 0)],
+            nextPageToken: null
+        }));
+        const first = await fetchWorkflowHistory({ workflowId: 'wf-123' });
+        expect(mockGet).toHaveBeenCalledTimes(2);
+        // The next tick replays the last page (deduped) then finds nothing new and times out —
+        // one call, not a full re-walk from page 1 through both prior pages again.
+        mockGet.mockResolvedValueOnce(page({
+            workflow: null, runId: 'run-456', events: [], nextPageToken: 'token-2'
+        }));
+        const { result, cursor } = await fetchWorkflowHistoryUpdates({ workflowId: 'wf-123', runId: 'run-456', longPollTimeoutMs: 2500 }, first.cursor);
+        expect(mockGet).toHaveBeenCalledTimes(3);
+        expect(mockGet).toHaveBeenNthCalledWith(3, 'wf-123', {
+            runId: 'run-456', pageSize: 50, pageToken: 'token-2', includePayloads: false, longPollTimeoutMs: 2500
+        });
+        expect(result.events).toHaveLength(3);
+        expect(cursor.pageToken).toBe('token-2');
     });
     it('forwards an explicit runId and includePayloads, stopping after a single page', async () => {
         mockGet.mockResolvedValueOnce(page({
@@ -84,4 +117,153 @@ describe('fetchWorkflowHistory', () => {
         mockGet.mockResolvedValueOnce({ status: 200 });
         await expect(fetchWorkflowHistory({ workflowId: 'wf-x' })).rejects.toThrow(/invalid response/);
     });
+    it('extracts the chained run id from a WORKFLOW_EXECUTION_CONTINUED_AS_NEW event', async () => {
+        mockGet.mockResolvedValueOnce(page({
+            workflow: { workflowId: 'wf-4', runId: 'run-4', status: 'continued_as_new', startTime: at(0) },
+            runId: 'run-4',
+            events: [
+                workflowStarted(0),
+                {
+                    eventId: '9', eventTypeName: 'WORKFLOW_EXECUTION_CONTINUED_AS_NEW', eventTime: at(60),
+                    workflowExecutionContinuedAsNewEventAttributes: { newExecutionRunId: 'run-5' }
+                }
+            ],
+            nextPageToken: null
+        }));
+        const result = await fetchWorkflowHistory({ workflowId: 'wf-4' });
+        expect(result.continuedAsNewRunId).toBe('run-5');
+    });
+    it('returns null continuedAsNewRunId when there is no continue-as-new event', async () => {
+        mockGet.mockResolvedValueOnce(page({
+            workflow: { workflowId: 'wf-5', runId: 'run-5', status: 'completed', startTime: at(0), closeTime: at(1) },
+            runId: 'run-5', events: [workflowStarted(0)], nextPageToken: null
+        }));
+        const result = await fetchWorkflowHistory({ workflowId: 'wf-5' });
+        expect(result.continuedAsNewRunId).toBeNull();
+    });
+});
+describe('fetchWorkflowHistoryUpdates', () => {
+    it('stops as soon as the first hop finds new events, without draining further buffered pages', async () => {
+        // Two pages already exist beyond the resume point; the old behavior kept paging until a
+        // timeout, which could silently merge several transitions (even a terminal one) into a
+        // single delayed render. A poller needs each batch back as soon as it has something new.
+        mockGet.mockResolvedValueOnce(page({
+            workflow: { workflowId: 'wf-7', runId: 'run-7', status: 'running', startTime: at(0) },
+            runId: 'run-7',
+            events: [workflowStarted(0)],
+            nextPageToken: 'page-2'
+        }));
+        const { result, cursor } = await fetchWorkflowHistoryUpdates({ workflowId: 'wf-7', runId: 'run-7', longPollTimeoutMs: 2500 });
+        expect(mockGet).toHaveBeenCalledTimes(1);
+        expect(mockGet).toHaveBeenCalledWith('wf-7', {
+            runId: 'run-7', pageSize: 50, pageToken: undefined, includePayloads: false, longPollTimeoutMs: 2500
+        });
+        expect(result.events).toHaveLength(1);
+        expect(result.workflow?.status).toBe('running');
+        // Resumes from the still-unfetched page 2 next time, not from the start.
+        expect(cursor.pageToken).toBe('page-2');
+        expect(cursor.lastEventId).toBe(1);
+    });
+    it('drains buffered pages when the describe already reports the run closed, so the terminal events are not stranded', async () => {
+        // >1 page of events accumulated between polls, then the run completed: the fresh
+        // describe says 'completed' before the closing events have been paged in. Stopping
+        // after the first batch would make the poller act on the terminal status with the
+        // final steps' events unfetched.
+        mockGet
+            .mockResolvedValueOnce(page({
+            workflow: { workflowId: 'wf-8', runId: 'run-8', status: 'completed', startTime: at(0), closeTime: at(30) },
+            runId: 'run-8',
+            events: [workflowStarted(0), scheduled('2', 'wf#step', 0)],
+            nextPageToken: 'page-2'
+        }))
+            .mockResolvedValueOnce(page({
+            workflow: null,
+            runId: 'run-8',
+            events: [started('3', '2', 1), completed('4', '2', 30)],
+            nextPageToken: null
+        }));
+        const { result } = await fetchWorkflowHistoryUpdates({ workflowId: 'wf-8', runId: 'run-8' });
+        expect(mockGet).toHaveBeenCalledTimes(2);
+        expect(result.events).toHaveLength(4);
+        expect(result.spans).toHaveLength(1);
+        expect(result.spans[0].status).toBe('completed');
+    });
+    it('drains to the CONTINUED_AS_NEW event on a later page when the describe already reports continued_as_new', async () => {
+        mockGet
+            .mockResolvedValueOnce(page({
+            workflow: { workflowId: 'wf-9', runId: 'run-9', status: 'continued_as_new', startTime: at(0) },
+            runId: 'run-9',
+            events: [workflowStarted(0)],
+            nextPageToken: 'page-2'
+        }))
+            .mockResolvedValueOnce(page({
+            workflow: null,
+            runId: 'run-9',
+            events: [{
+                    eventId: '2', eventTypeName: 'WORKFLOW_EXECUTION_CONTINUED_AS_NEW', eventTime: at(60),
+                    workflowExecutionContinuedAsNewEventAttributes: { newExecutionRunId: 'run-10' }
+                }],
+            nextPageToken: null
+        }));
+        const { result } = await fetchWorkflowHistoryUpdates({ workflowId: 'wf-9', runId: 'run-9' });
+        expect(mockGet).toHaveBeenCalledTimes(2);
+        expect(result.continuedAsNewRunId).toBe('run-10');
+    });
+    it('forwards longPollTimeoutMs to the API on a resumed poll', async () => {
+        mockGet.mockResolvedValueOnce(page({
+            workflow: { workflowId: 'wf-7', runId: 'run-7', status: 'running', startTime: at(0) },
+            runId: 'run-7', events: [], nextPageToken: null
+        }));
+        await fetchWorkflowHistoryUpdates({ workflowId: 'wf-7', runId: 'run-7', longPollTimeoutMs: 2500 });
+        expect(mockGet).toHaveBeenCalledWith('wf-7', {
+            runId: 'run-7', pageSize: 50, pageToken: undefined, includePayloads: false, longPollTimeoutMs: 2500
+        });
+    });
+    it('sends no long-poll param when longPollTimeoutMs is not provided', async () => {
+        mockGet.mockResolvedValueOnce(page({
+            workflow: { workflowId: 'wf-7', runId: 'run-7', status: 'running', startTime: at(0) },
+            runId: 'run-7', events: [], nextPageToken: null
+        }));
+        await fetchWorkflowHistoryUpdates({ workflowId: 'wf-7', runId: 'run-7' });
+        expect(mockGet).toHaveBeenCalledWith('wf-7', expect.not.objectContaining({ longPollTimeoutMs: expect.anything() }));
+    });
+    it('times out with the sent token echoed back when there is genuinely nothing new', async () => {
+        mockGet.mockResolvedValueOnce(page({
+            workflow: { workflowId: 'wf-7', runId: 'run-7', status: 'running', startTime: at(0) },
+            runId: 'run-7',
+            events: [],
+            nextPageToken: null
+        }));
+        const { result, cursor } = await fetchWorkflowHistoryUpdates({ workflowId: 'wf-7', runId: 'run-7' });
+        expect(mockGet).toHaveBeenCalledTimes(1);
+        expect(result.events).toHaveLength(0);
+        expect(cursor.pageToken).toBeUndefined();
+    });
+    it('resumes from a prior cursor and picks up the workflow status finishing, not the status frozen on the cursor', async () => {
+        const cursor = {
+            pageToken: 'token-2',
+            lastEventId: 2,
+            meta: { workflowId: 'wf-6', runId: 'run-6', status: 'running', startTime: at(0) },
+            runId: 'run-6',
+            events: [workflowStarted(0), scheduled('2', 'wf#step', 0)]
+        };
+        mockGet.mockResolvedValueOnce(page({
+            // The server re-describes on every `wait` call specifically so a resumed poll can see
+            // status changes — this must win over the (now-stale) status carried on the cursor.
+            workflow: { workflowId: 'wf-6', runId: 'run-6', status: 'completed', startTime: at(0), closeTime: at(5) },
+            runId: 'run-6',
+            events: [started('3', '2', 1), completed('4', '2', 5)],
+            nextPageToken: null
+        }));
+        const { result, cursor: nextCursor } = await fetchWorkflowHistoryUpdates({ workflowId: 'wf-6', runId: 'run-6', longPollTimeoutMs: 2500 }, cursor);
+        expect(mockGet).toHaveBeenCalledTimes(1);
+        expect(mockGet).toHaveBeenCalledWith('wf-6', {
+            runId: 'run-6', pageSize: 50, pageToken: 'token-2', includePayloads: false, longPollTimeoutMs: 2500
+        });
+        expect(result.workflow?.status).toBe('completed');
+        // The events already carried on the cursor (2) plus the genuinely new ones (2).
+        expect(result.events).toHaveLength(4);
+        expect(nextCursor.pageToken).toBe('token-2');
+        expect(nextCursor.lastEventId).toBe(4);
+    });
 });

package/dist/templates/agent_instructions/CLAUDE.md.template

@@ -44,7 +44,7 @@ config/
 ## Key Conventions
 
 - **Workflows are deterministic**: No I/O, no `Date.now()`, no `Math.random()` in `workflow.ts`. All side effects go in steps or evaluators.
-- **HTTP clients**: Always use `httpClient` from `@outputai/http` -- never raw `fetch` or `axios`. This enables automatic tracing and cost tracking.
+- **HTTP clients**: Use `outputFetch` or `createKyClient` from `@outputai/http` instead of raw `fetch` or `axios`. Requests are automatically traced; use `addRequestCost` when cost tracking is needed.
 - **LLM calls**: Use `generateText` from `@outputai/llm` with `.prompt` files. Never call LLM APIs directly.
 
 ---

package/dist/templates/project/src/clients/jina.ts.template

@@ -1,4 +1,4 @@
-import { httpClient } from '@outputai/http';
+import { createKyClient } from '@outputai/http';
 
 export interface JinaReaderResponse {
   code: number;
@@ -12,13 +12,13 @@ export interface JinaReaderResponse {
   };
 }
 
-const jinaClient = httpClient( {
-  prefixUrl: 'https://r.jina.ai',
+const client = createKyClient( {
+  prefix: 'https://r.jina.ai',
   timeout: 30000
 } );
 
 export async function fetchBlogContent( url: string ): Promise<JinaReaderResponse> {
-  const response = await jinaClient.post( '', {
+  const response = await client.post( '', {
     json: { url },
     headers: {
       'Accept': 'application/json',

package/dist/utils/color.d.ts

@@ -0,0 +1,7 @@
+/**
+ * Standard TTY/env precedence for whether a command should colorize its
+ * output: the command's own --color/--no-color flag wins first, then
+ * NO_COLOR opts out, then FORCE_COLOR opts in even off a TTY, then finally
+ * fall back to whether stdout is an interactive terminal.
+ */
+export declare function shouldColorize(flag: boolean): boolean;

package/dist/utils/color.js

@@ -0,0 +1,12 @@
+/**
+ * Standard TTY/env precedence for whether a command should colorize its
+ * output: the command's own --color/--no-color flag wins first, then
+ * NO_COLOR opts out, then FORCE_COLOR opts in even off a TTY, then finally
+ * fall back to whether stdout is an interactive terminal.
+ */
+export function shouldColorize(flag) {
+    // Per the NO_COLOR convention (https://no-color.org/), presence disables color
+    // regardless of value — `NO_COLOR=` must opt out just like `NO_COLOR=1`.
+    return flag && process.env.NO_COLOR === undefined &&
+        (!!process.env.FORCE_COLOR || process.stdout.isTTY === true);
+}

package/dist/utils/color.spec.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/utils/color.spec.js

@@ -0,0 +1,43 @@
+import { describe, it, expect, beforeEach, afterEach } from 'vitest';
+import { shouldColorize } from '#utils/color.js';
+describe('shouldColorize', () => {
+    const originalEnv = { ...process.env };
+    const originalTTY = process.stdout.isTTY;
+    beforeEach(() => {
+        delete process.env.NO_COLOR;
+        delete process.env.FORCE_COLOR;
+        Object.defineProperty(process.stdout, 'isTTY', { value: true, configurable: true });
+    });
+    afterEach(() => {
+        process.env = { ...originalEnv };
+        Object.defineProperty(process.stdout, 'isTTY', { value: originalTTY, configurable: true });
+    });
+    it('returns false when the flag itself is false', () => {
+        expect(shouldColorize(false)).toBe(false);
+    });
+    it('returns true on a TTY with no overrides', () => {
+        expect(shouldColorize(true)).toBe(true);
+    });
+    it('disables color when NO_COLOR is set to a non-empty value', () => {
+        process.env.NO_COLOR = '1';
+        expect(shouldColorize(true)).toBe(false);
+    });
+    it('disables color when NO_COLOR is present but empty, per the NO_COLOR convention', () => {
+        process.env.NO_COLOR = '';
+        expect(shouldColorize(true)).toBe(false);
+    });
+    it('enables color off a TTY when FORCE_COLOR is set', () => {
+        Object.defineProperty(process.stdout, 'isTTY', { value: false, configurable: true });
+        process.env.FORCE_COLOR = '1';
+        expect(shouldColorize(true)).toBe(true);
+    });
+    it('disables color off a TTY with no FORCE_COLOR', () => {
+        Object.defineProperty(process.stdout, 'isTTY', { value: false, configurable: true });
+        expect(shouldColorize(true)).toBe(false);
+    });
+    it('NO_COLOR wins even when FORCE_COLOR is also set', () => {
+        process.env.FORCE_COLOR = '1';
+        process.env.NO_COLOR = '1';
+        expect(shouldColorize(true)).toBe(false);
+    });
+});

package/dist/utils/env_loader.js

@@ -5,7 +5,6 @@
  */
 import { existsSync } from 'node:fs';
 import { resolve } from 'node:path';
-import * as dotenv from 'dotenv';
 import debugFactory from 'debug';
 const debug = debugFactory('output-cli:env-loader');
 export function loadEnvironment() {
@@ -17,5 +16,10 @@ export function loadEnvironment() {
         return;
     }
     debug(`Loading env from: ${envPath}`);
-    dotenv.config({ path: envPath, quiet: true });
+    try {
+        process.loadEnvFile(envPath);
+    }
+    catch (err) {
+        debug(`Warning: Failed to load env file ${envPath}: ${err}`);
+    }
 }

package/dist/utils/env_loader.spec.js

@@ -1,43 +1,72 @@
-/**
- * Tests for the env loader utility
- */
-import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest';
-import { existsSync } from 'node:fs';
-import { resolve } from 'node:path';
-import * as dotenv from 'dotenv';
-vi.mock('node:fs');
-vi.mock('dotenv');
+import { afterAll, afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+import { mkdirSync, mkdtempSync, readdirSync, rmSync, writeFileSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { loadEnvironment } from './env_loader.js';
 describe('loadEnvironment', () => {
-    const originalEnv = { ...process.env };
-    const mockCwd = '/mock/project';
+    const mockCwd = mkdtempSync(join(tmpdir(), 'output-env-loader-'));
     beforeEach(() => {
-        vi.resetModules();
-        vi.clearAllMocks();
+        for (const name of readdirSync(mockCwd)) {
+            rmSync(join(mockCwd, name), { recursive: true, force: true });
+        }
         vi.spyOn(process, 'cwd').mockReturnValue(mockCwd);
-        vi.spyOn(console, 'log').mockImplementation(() => { });
-        vi.spyOn(console, 'warn').mockImplementation(() => { });
+        vi.stubEnv('OUTPUT_CLI_ENV', undefined);
+        vi.stubEnv('OUTPUT_API_URL', undefined);
+        vi.stubEnv('OUTPUT_API_TOKEN', undefined);
     });
     afterEach(() => {
-        process.env = { ...originalEnv };
         vi.restoreAllMocks();
+        vi.unstubAllEnvs();
     });
-    it('should load from OUTPUT_CLI_ENV when set and file exists', async () => {
-        process.env.OUTPUT_CLI_ENV = '.env.prod';
-        const expectedPath = resolve(mockCwd, '.env.prod');
-        vi.mocked(existsSync).mockReturnValue(true);
-        vi.mocked(dotenv.config).mockReturnValue({ parsed: { OUTPUT_API_URL: 'https://prod.api.com' } });
-        const { loadEnvironment } = await import('./env_loader.js');
+    afterAll(() => {
+        rmSync(mockCwd, { recursive: true, force: true });
+    });
+    it('loads variables from OUTPUT_CLI_ENV', () => {
+        writeFileSync(join(mockCwd, '.env'), [
+            'OUTPUT_API_URL=https://default.api.com',
+            'OUTPUT_API_TOKEN=default-token'
+        ].join('\n'));
+        writeFileSync(join(mockCwd, '.env.mock'), [
+            'OUTPUT_API_URL=https://mock.api.com',
+            'OUTPUT_API_TOKEN=mock-token'
+        ].join('\n'));
+        process.env.OUTPUT_CLI_ENV = '.env.mock';
+        loadEnvironment();
+        expect(process.env.OUTPUT_API_URL).toBe('https://mock.api.com');
+        expect(process.env.OUTPUT_API_TOKEN).toBe('mock-token');
+    });
+    it('loads variables from .env by default', () => {
+        writeFileSync(join(mockCwd, '.env'), [
+            'OUTPUT_API_URL=https://default.api.com',
+            'OUTPUT_API_TOKEN=default-token'
+        ].join('\n'));
+        writeFileSync(join(mockCwd, '.env.mock'), [
+            'OUTPUT_API_URL=https://mock.api.com',
+            'OUTPUT_API_TOKEN=mock-token'
+        ].join('\n'));
         loadEnvironment();
-        expect(dotenv.config).toHaveBeenCalledWith({ path: expectedPath, quiet: true });
-    });
-    it('should load .env by default and log', async () => {
-        delete process.env.OUTPUT_CLI_ENV;
-        const envPath = resolve(mockCwd, '.env');
-        vi.mocked(existsSync).mockImplementation(p => p === envPath);
-        vi.mocked(dotenv.config).mockReturnValue({ parsed: {} });
-        const { loadEnvironment } = await import('./env_loader.js');
+        expect(process.env.OUTPUT_API_URL).toBe('https://default.api.com');
+        expect(process.env.OUTPUT_API_TOKEN).toBe('default-token');
+    });
+    it('does nothing when the env file is missing', () => {
+        expect(() => loadEnvironment()).not.toThrow();
+        expect(process.env.OUTPUT_API_URL).toBeUndefined();
+        expect(process.env.OUTPUT_API_TOKEN).toBeUndefined();
+    });
+    it('does not throw when the env path is not a readable file', () => {
+        mkdirSync(join(mockCwd, 'not-a-file.env'));
+        process.env.OUTPUT_CLI_ENV = 'not-a-file.env';
+        expect(() => loadEnvironment()).not.toThrow();
+        expect(process.env.OUTPUT_API_URL).toBeUndefined();
+    });
+    it('does not overwrite already-set process.env values', () => {
+        vi.stubEnv('OUTPUT_API_URL', 'https://ambient.api.com');
+        writeFileSync(join(mockCwd, '.env'), [
+            'OUTPUT_API_URL=https://file.api.com',
+            'OUTPUT_API_TOKEN=file-token'
+        ].join('\n'));
         loadEnvironment();
-        expect(dotenv.config).toHaveBeenCalledTimes(1);
-        expect(dotenv.config).toHaveBeenCalledWith({ path: envPath, quiet: true });
+        expect(process.env.OUTPUT_API_URL).toBe('https://ambient.api.com');
+        expect(process.env.OUTPUT_API_TOKEN).toBe('file-token');
     });
 });

package/dist/utils/format_workflow_result.d.ts

@@ -1,5 +1,6 @@
 import type { WorkflowResultResponse, WorkflowResultResponseStatus } from '../api/generated/api.js';
 type WorkflowResult = Pick<WorkflowResultResponse, 'workflowId' | 'output' | 'status' | 'error'>;
 export declare const ERROR_STATUSES: ReadonlySet<WorkflowResultResponseStatus | undefined>;
+export declare const TERMINAL_STATUSES: ReadonlySet<string>;
 export declare function formatWorkflowResult(result: WorkflowResult): string;
 export {};

package/dist/utils/format_workflow_result.js

@@ -1,5 +1,9 @@
 import { normalizeWorkflowStatus } from './normalize_workflow_status.js';
 export const ERROR_STATUSES = new Set(['failed', 'canceled', 'terminated', 'timed_out']);
+// Every error status plus the one success status — derived so the two sets can't
+// silently drift apart as error statuses evolve. Shared by `workflow monitor` and
+// the dev TUI's `useRunDetail`/`useStepGraph` so both agree on what "done" means.
+export const TERMINAL_STATUSES = new Set(['completed', ...ERROR_STATUSES]);
 export function formatWorkflowResult(result) {
     const status = normalizeWorkflowStatus(result.status);
     const lines = [

package/dist/utils/monitor_log.d.ts

@@ -0,0 +1,20 @@
+/**
+ * Turns newly-correlated spans into append-only status lines for `workflow
+ * monitor`. Unlike the waterfall (which needs the full span set up front to
+ * lay out a time axis), a live monitor just reports each span's status
+ * transitions as they're observed on each poll.
+ */
+import type { Span, SpanStatus } from '#services/workflow_history/correlator.js';
+export interface SpanUpdate {
+    span: Span;
+    label: string;
+}
+/**
+ * Returns spans whose status changed since the last call. `seen` is mutated
+ * in place so callers can carry it across polls. Pending spans are skipped —
+ * nothing worth reporting until a step starts.
+ */
+export declare function diffSpanUpdates(spans: Span[], labels: Map<string, string>, seen: Map<string, SpanStatus>): SpanUpdate[];
+/** Formats the continue-as-new transition line, keeping its glyph in the formatting layer. */
+export declare function formatContinuedAsNew(runId: string): string;
+export declare function formatSpanUpdate(update: SpanUpdate, color: boolean): string;

package/dist/utils/monitor_log.js

@@ -0,0 +1,48 @@
+import { ANSI, formatDurationLabel, makeTint } from '#utils/waterfall.js';
+const GLYPH = {
+    pending: '·',
+    running: '●',
+    completed: '✓',
+    failed: '✗'
+};
+// Continue-as-new is a workflow-level transition, not a span status, so it gets its own glyph
+// here in the formatting layer rather than being concatenated into the message at the call site.
+const CONTINUED_AS_NEW_GLYPH = '↻';
+/**
+ * Returns spans whose status changed since the last call. `seen` is mutated
+ * in place so callers can carry it across polls. Pending spans are skipped —
+ * nothing worth reporting until a step starts.
+ */
+export function diffSpanUpdates(spans, labels, seen) {
+    const updates = [];
+    for (const span of spans) {
+        if (span.status === 'pending' || seen.get(span.id) === span.status) {
+            continue;
+        }
+        seen.set(span.id, span.status);
+        updates.push({ span, label: labels.get(span.id) ?? span.name });
+    }
+    return updates;
+}
+/** Formats the continue-as-new transition line, keeping its glyph in the formatting layer. */
+export function formatContinuedAsNew(runId) {
+    return `${CONTINUED_AS_NEW_GLYPH} continued as new run ${runId}`;
+}
+export function formatSpanUpdate(update, color) {
+    const { span, label } = update;
+    const glyph = GLYPH[span.status];
+    const tint = makeTint(color);
+    const tintStatus = (text) => tint(text, ANSI[span.status]);
+    switch (span.status) {
+        case 'running':
+            return `${tintStatus(glyph)} ${label} running…`;
+        case 'completed':
+            return `${tintStatus(glyph)} ${label}  ${formatDurationLabel(Math.max(0, span.durationMs))}`;
+        case 'failed': {
+            const reason = span.failureMessage ? `: ${span.failureMessage}` : '';
+            return `${tintStatus(glyph)} ${label} failed${reason}`;
+        }
+        default:
+            return `${glyph} ${label} ${span.status}`;
+    }
+}

package/dist/utils/monitor_log.spec.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/utils/monitor_log.spec.js

@@ -0,0 +1,71 @@
+import { describe, it, expect } from 'vitest';
+import { diffSpanUpdates, formatSpanUpdate } from '#utils/monitor_log.js';
+const span = (overrides) => ({
+    name: 'Step',
+    technicalName: 'wf#step',
+    description: null,
+    kind: 'activity',
+    attempt: 1,
+    startedAt: null,
+    scheduledAt: null,
+    completedAt: null,
+    startOffsetMs: 0,
+    endOffsetMs: 0,
+    durationMs: 0,
+    failureMessage: null,
+    ...overrides
+});
+describe('diffSpanUpdates', () => {
+    it('skips pending spans', () => {
+        const seen = new Map();
+        const updates = diffSpanUpdates([span({ id: '1', status: 'pending' })], new Map(), seen);
+        expect(updates).toHaveLength(0);
+        expect(seen.size).toBe(0);
+    });
+    it('reports a span the first time it is seen in a non-pending status', () => {
+        const seen = new Map();
+        const updates = diffSpanUpdates([span({ id: '1', status: 'running' })], new Map([['1', 'Fetch page']]), seen);
+        expect(updates).toHaveLength(1);
+        expect(updates[0].label).toBe('Fetch page');
+        expect(seen.get('1')).toBe('running');
+    });
+    it('does not re-report a span whose status is unchanged since the last call', () => {
+        const seen = new Map([['1', 'running']]);
+        const updates = diffSpanUpdates([span({ id: '1', status: 'running' })], new Map(), seen);
+        expect(updates).toHaveLength(0);
+    });
+    it('reports a span again once its status transitions (running -> completed)', () => {
+        const seen = new Map([['1', 'running']]);
+        const updates = diffSpanUpdates([span({ id: '1', status: 'completed' })], new Map(), seen);
+        expect(updates).toHaveLength(1);
+        expect(seen.get('1')).toBe('completed');
+    });
+    it('falls back to the span name when no label is provided', () => {
+        const seen = new Map();
+        const updates = diffSpanUpdates([span({ id: '1', status: 'running', name: 'Unlabeled' })], new Map(), seen);
+        expect(updates[0].label).toBe('Unlabeled');
+    });
+});
+describe('formatSpanUpdate', () => {
+    it('formats a running span', () => {
+        const line = formatSpanUpdate({ span: span({ id: '1', status: 'running' }), label: 'Fetch page' }, false);
+        expect(line).toBe('● Fetch page running…');
+    });
+    it('formats a completed span with its duration', () => {
+        const line = formatSpanUpdate({ span: span({ id: '1', status: 'completed', durationMs: 1234 }), label: 'Fetch page' }, false);
+        expect(line).toBe('✓ Fetch page  1s');
+    });
+    it('formats a failed span with its failure message', () => {
+        const line = formatSpanUpdate({ span: span({ id: '1', status: 'failed', failureMessage: 'boom' }), label: 'Fetch page' }, false);
+        expect(line).toBe('✗ Fetch page failed: boom');
+    });
+    it('formats a failed span without a failure message', () => {
+        const line = formatSpanUpdate({ span: span({ id: '1', status: 'failed' }), label: 'Fetch page' }, false);
+        expect(line).toBe('✗ Fetch page failed');
+    });
+    it('wraps the glyph in ANSI codes when color is enabled', () => {
+        const line = formatSpanUpdate({ span: span({ id: '1', status: 'running' }), label: 'Fetch page' }, true);
+        expect(line).toContain('●');
+        expect(line).not.toBe('● Fetch page running…'); // color codes present
+    });
+});

package/dist/utils/port_collision.d.ts

@@ -3,14 +3,21 @@
  * an actionable hint that names the conflicting port and the env var to
  * override.
  *
- * Docker compose surfaces port collisions through two common error shapes:
- *   - "Bind for 0.0.0.0:3001 failed: port is already allocated"
- *   - "failed to bind host port for 0.0.0.0:7233:.../tcp: address already in use"
+ * Docker wraps the same failure differently across versions and platforms —
+ * Docker 29 on macOS nests it three deep:
  *
- * We match both, extract the host port, then map it back to the env var that
- * sets it. The map prefers a runtime lookup of resolved ports (so a user who
- * already set OUTPUT_API_HOST_PORT=3050 sees that var named when 3050
- * collides) and falls back to a default-port table for the unresolved case.
+ *   Error response from daemon: failed to set up container networking: driver
+ *   failed programming external connectivity on endpoint out-api-1 (a1b2…):
+ *   Bind for 0.0.0.0:3001 failed: port is already allocated
+ *
+ * Matching whole message shapes means a new wrapper silently drops the hint, so
+ * we anchor on the terminal phrase instead and take the host port nearest to it.
+ * That survives wrappers we haven't seen.
+ *
+ * The port is then mapped back to the env var that sets it. The map prefers a
+ * runtime lookup of resolved ports (so a user who already set
+ * OUTPUT_API_HOST_PORT=3050 sees that var named when 3050 collides) and falls
+ * back to a default-port table for the unresolved case.
  */
 /**
  * Find the first host port mentioned in a docker compose bind failure.
@@ -24,6 +31,14 @@ export declare function extractCollidedPort(stderr: string): number | null;
  * that overrides it; otherwise it suggests freeing the port.
  */
 export declare function formatPortCollisionHint(stderr: string, resolvedPorts: Record<string, number>): string | null;
+/**
+ * Compose a docker-failure message from a caller-supplied core sentence and the
+ * process's recent output: an actionable port-collision hint (when one is
+ * detected) is prepended, and the raw recent output is appended. Shared by the
+ * foreground exit handler and the detached/reconcile path so both surface the
+ * same failure shape.
+ */
+export declare function formatComposeFailure(reason: string, output: string, resolvedPorts: Record<string, number>): string;
 /**
  * Build a hint from a known list of colliding ports. For a single collision
  * the output matches `formatPortCollisionHint` exactly so callers stay