@outputai/cli 0.10.1-next.37650c5.0 → 0.10.1-next.52bedcf.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/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/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/waterfall.d.ts

@@ -8,7 +8,7 @@
  * Atlas's `StepGantt`; the bar geometry mirrors its leftPct/widthPct as integer
  * terminal columns.
  */
-import type { Span } from '#services/workflow_history/correlator.js';
+import type { Span, SpanStatus } from '#services/workflow_history/correlator.js';
 export interface WaterfallOptions {
     width: number;
     color: boolean;
@@ -22,6 +22,8 @@ export interface BarGeometry {
 }
 export declare const FULL_BLOCK = "\u2588";
 export declare const THIN_BLOCK = "\u258F";
+export declare const ANSI: Record<SpanStatus | 'dim' | 'reset', string>;
+export declare function makeTint(color: boolean): (text: string, code: string) => string;
 export declare function pickTickStep(totalMs: number): number;
 export declare function buildTicks(totalMs: number): number[];
 export declare function formatTickLabel(ms: number): string;

package/dist/utils/waterfall.js

@@ -12,7 +12,7 @@ const TICK_STEPS_MS = [
     60 * 60_000, 2 * 60 * 60_000, 6 * 60 * 60_000, 12 * 60 * 60_000, 24 * 60 * 60_000
 ];
 const TARGET_TICKS = 8;
-const ANSI = {
+export const ANSI = {
     completed: '',
     running: '',
     failed: '',
@@ -20,6 +20,12 @@ const ANSI = {
     dim: '',
     reset: ''
 };
+// Shared by renderWaterfall and monitor_log's live status lines so there's one
+// place that knows how to wrap text in an ANSI code (and reset it) -- or not,
+// when color is disabled.
+export function makeTint(color) {
+    return (text, code) => (color ? `${code}${text}${ANSI.reset}` : text);
+}
 function clamp(value, lo, hi) {
     return Math.min(Math.max(value, lo), hi);
 }
@@ -162,7 +168,7 @@ export default function renderWaterfall(spans, totalDurationMs, options) {
     if (spans.length === 0) {
         return [header, 'No steps found for this run.'].filter(Boolean).join('\n\n');
     }
-    const tint = (text, code) => (color ? `${code}${text}${ANSI.reset}` : text);
+    const tint = makeTint(color);
     const labelFor = (span) => labels?.get(span.id) ?? span.name;
     const durationFor = (span) => formatDurationLabel(Math.max(0, span.durationMs));
     const longestLabel = Math.max(...spans.map(s => labelFor(s).length));

package/dist/views/dev/hooks/use_run_detail.js

@@ -2,13 +2,16 @@ import { useEffect, useRef, useState } from 'react';
 import { readFile } from 'node:fs/promises';
 import { getWorkflowIdResult, getWorkflowIdRunsRidResult, getWorkflowIdTraceLog, getWorkflowIdRunsRidTraceLog } from '#api/generated/api.js';
 import { normalizeWorkflowStatus } from '#utils/normalize_workflow_status.js';
+import { TERMINAL_STATUSES } from '#utils/format_workflow_result.js';
+import { createBoundedCache } from '#views/dev/utils/bounded_cache.js';
 const EMPTY_DETAIL = {
     result: null,
     trace: null,
     steps: [],
     loading: false
 };
-const runDetailCache = new Map();
+const RUN_DETAIL_CACHE_MAX = 50;
+const runDetailCache = createBoundedCache(RUN_DETAIL_CACHE_MAX);
 const stepNameOf = (node) => {
     if (node.name) {
         return node.name;
@@ -99,13 +102,9 @@ const fetchResult = async (workflowId, runId) => {
         return null;
     }
 };
-/**
- * Statuses that mean the workflow has stopped advancing. The cache is
- * intentionally only populated for these — partial results from a still-
- * running workflow would otherwise stick and stall the UI when the run
- * eventually finishes.
- */
-const TERMINAL_STATUSES = new Set(['completed', 'failed', 'canceled', 'terminated', 'timed_out']);
+// The cache is intentionally only populated for terminal statuses — partial results
+// from a still-running workflow would otherwise stick and stall the UI when the run
+// eventually finishes. `TERMINAL_STATUSES` is shared with `workflow monitor`.
 export const isTerminalRunStatus = (status) => Boolean(status && TERMINAL_STATUSES.has(status));
 export const useRunDetail = (workflowId, runId, status) => {
     const [detail, setDetail] = useState(EMPTY_DETAIL);

package/dist/views/dev/hooks/use_step_graph.js

@@ -3,6 +3,7 @@ import { fetchWorkflowHistory } from '#services/workflow_history.js';
 import buildSpanLabels from '#utils/span_labels.js';
 import { isTerminalRunStatus } from '#views/dev/hooks/use_run_detail.js';
 import { usePoll, POLL_INTERVAL_MS } from '#views/dev/hooks/use_poll.js';
+import { createBoundedCache } from '#views/dev/utils/bounded_cache.js';
 const EMPTY_GRAPH = {
     spans: [],
     totalDurationMs: 0,
@@ -11,7 +12,8 @@ const EMPTY_GRAPH = {
     loading: false,
     error: null
 };
-const stepGraphCache = new Map();
+const STEP_GRAPH_CACHE_MAX = 50;
+const stepGraphCache = createBoundedCache(STEP_GRAPH_CACHE_MAX);
 /**
  * Fetches a run's correlated step spans for the dev TUI's waterfall overlay,
  * reusing the same `fetchWorkflowHistory` path as the `workflow history` CLI

package/dist/views/dev/utils/bounded_cache.d.ts

@@ -0,0 +1,14 @@
+export interface BoundedCache<K, V extends {}> {
+    get(key: K): V | undefined;
+    set(key: K, value: V): void;
+    has(key: K): boolean;
+    clear(): void;
+    size(): number;
+}
+/**
+ * A small LRU cache backed by an insertion-ordered `Map`, capped at `maxSize`
+ * entries. Reading a key refreshes its recency; once the cap is exceeded the
+ * least-recently-used entries are evicted. Drop-in compatible with the
+ * `Map.get` / `Map.set` calls it replaces.
+ */
+export declare const createBoundedCache: <K, V extends {}>(maxSize: number) => BoundedCache<K, V>;

package/dist/views/dev/utils/bounded_cache.js

@@ -0,0 +1,42 @@
+/**
+ * A small LRU cache backed by an insertion-ordered `Map`, capped at `maxSize`
+ * entries. Reading a key refreshes its recency; once the cap is exceeded the
+ * least-recently-used entries are evicted. Drop-in compatible with the
+ * `Map.get` / `Map.set` calls it replaces.
+ */
+export const createBoundedCache = (maxSize) => {
+    if (maxSize < 1) {
+        throw new Error('createBoundedCache: maxSize must be >= 1');
+    }
+    const entries = new Map();
+    return {
+        get(key) {
+            const value = entries.get(key);
+            if (value === undefined) {
+                return value;
+            }
+            entries.delete(key);
+            entries.set(key, value);
+            return value;
+        },
+        set(key, value) {
+            entries.delete(key);
+            entries.set(key, value);
+            if (entries.size > maxSize) {
+                const oldest = entries.keys().next();
+                if (!oldest.done) {
+                    entries.delete(oldest.value);
+                }
+            }
+        },
+        has(key) {
+            return entries.has(key);
+        },
+        clear() {
+            entries.clear();
+        },
+        size() {
+            return entries.size;
+        }
+    };
+};

package/dist/views/dev/utils/bounded_cache.spec.d.ts

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

package/dist/views/dev/utils/bounded_cache.spec.js

@@ -0,0 +1,53 @@
+import { describe, expect, it } from 'vitest';
+import { createBoundedCache } from './bounded_cache.js';
+describe('createBoundedCache', () => {
+    it('round-trips set and get', () => {
+        const cache = createBoundedCache(3);
+        cache.set('a', 1);
+        expect(cache.get('a')).toBe(1);
+        expect(cache.get('missing')).toBeUndefined();
+    });
+    it('evicts the oldest entry once maxSize is exceeded', () => {
+        const cache = createBoundedCache(2);
+        cache.set('a', 1);
+        cache.set('b', 2);
+        cache.set('c', 3);
+        expect(cache.has('a')).toBe(false);
+        expect(cache.get('b')).toBe(2);
+        expect(cache.get('c')).toBe(3);
+    });
+    it('refreshes recency on get so the touched entry survives eviction', () => {
+        const cache = createBoundedCache(2);
+        cache.set('a', 1);
+        cache.set('b', 2);
+        // Touch 'a' so 'b' becomes the least-recently-used entry.
+        expect(cache.get('a')).toBe(1);
+        cache.set('c', 3);
+        expect(cache.has('a')).toBe(true);
+        expect(cache.has('b')).toBe(false);
+        expect(cache.has('c')).toBe(true);
+    });
+    it('updates an existing key in place without growing', () => {
+        const cache = createBoundedCache(2);
+        cache.set('a', 1);
+        cache.set('a', 2);
+        expect(cache.get('a')).toBe(2);
+        expect(cache.size()).toBe(1);
+    });
+    it('supports has and clear', () => {
+        const cache = createBoundedCache(2);
+        cache.set('a', 1);
+        expect(cache.has('a')).toBe(true);
+        cache.clear();
+        expect(cache.has('a')).toBe(false);
+        expect(cache.size()).toBe(0);
+    });
+    it('never exceeds maxSize', () => {
+        const cache = createBoundedCache(3);
+        Array.from({ length: 100 }, (_, i) => i).forEach(i => {
+            cache.set(`key-${i}`, i);
+            expect(cache.size()).toBeLessThanOrEqual(3);
+        });
+        expect(cache.size()).toBe(3);
+    });
+});