Files
harness-tools/extensions/workflow/dashboard.test.ts
T

690 lines
19 KiB
TypeScript

import assert from "node:assert/strict";
import { createHash } from "node:crypto";
import { mkdtemp, rm, writeFile } from "node:fs/promises";
import { tmpdir } from "node:os";
import * as path from "node:path";
import test from "node:test";
import type {
ExtensionAPI,
ExtensionCommandContext,
} from "@earendil-works/pi-coding-agent";
import {
DEFAULT_WORKFLOW_CONFIG,
type WorkflowRun,
} from "./domain.ts";
import type { WorkflowCommand, WorkflowView } from "./engine.ts";
import { openWorkflowDashboard } from "./dashboard.ts";
function run(overrides: Partial<WorkflowRun> = {}): WorkflowRun {
return {
runId: "run-1",
revision: 12,
mode: "feature",
status: "waiting",
background: false,
request: "Build search page",
featureName: "search-page",
parentSessionIds: ["session-1"],
lease: {
processId: 100,
activeParentSessionId: "session-1",
heartbeatAt: 1,
},
workspace: {
state: "ready",
sourceRoot: "/repo",
root: "/repo/.worktrees/workflow-search-page-run-1",
branch: "feature/search-page",
baseHead: "abc123",
},
effectiveConfig: structuredClone(DEFAULT_WORKFLOW_CONFIG),
pinnedProfiles: [],
preflightEligibility: [],
facts: {
stage: "design",
taskKind: "feature",
requestTerms: ["search"],
candidatePaths: [],
languages: [],
tags: [],
riskTags: [],
},
stages: [
{ id: "discover", kind: "discover", status: "completed" },
{ id: "design", kind: "design", status: "waiting" },
{ id: "prototype", kind: "prototype", status: "pending" },
{ id: "implement", kind: "implement", status: "pending" },
{ id: "verify", kind: "verify", status: "pending" },
{ id: "review", kind: "review", status: "pending" },
],
gates: [
{
id: "gate-1",
kind: "plan-approval",
status: "open",
revision: 12,
openedAt: 1,
},
],
decisions: [],
assignments: [],
prototypeCandidates: [],
prototypeCleanupCompleted: false,
deliveries: [],
commandReceipts: [],
uiImpact: "minor",
uiImpactEvidence: ["One visible field changes"],
affectedUiSurfaces: ["Search page"],
mutatorProfileNames: [],
reviewCycles: 0,
verificationCycles: 0,
...overrides,
};
}
function prototypeCandidate() {
return {
id: "candidate-1",
revision: 2,
assignmentId: "prototype-assignment-1",
profileName: "frontend-prototyper",
status: "succeeded" as const,
path: "/repo/prototypes/candidate-1",
route: "/search?variant=1",
structuralRationale: "Compact drill-down flow",
responsiveChecks: ["mobile", "desktop"],
browserEvidence: ["validated in browser"],
launchInstructions: "Run the prototype server.",
visualDigest: "visual-digest",
artifactPath: "/state/artifacts/candidate-1.md",
retainedArtifacts: ["/state/artifacts/candidate-1.png"],
};
}
function view(workflow: WorkflowRun): WorkflowView {
return {
runId: workflow.runId,
featureName: workflow.featureName,
request: workflow.request,
revision: workflow.revision,
status: workflow.status,
background: workflow.background,
workspace: workflow.workspace,
openGate: workflow.gates.find((gate) => gate.status === "open"),
uiImpact: workflow.uiImpact,
uiImpactEvidence: workflow.uiImpactEvidence,
affectedUiSurfaces: workflow.affectedUiSurfaces,
planArtifact: workflow.planArtifact,
reportArtifact: workflow.reportArtifact,
implementationContext: workflow.implementationContext,
stages: workflow.stages,
assignments: workflow.assignments,
mutatorProfileNames: workflow.mutatorProfileNames,
reviewCycles: workflow.reviewCycles,
verificationCycles: workflow.verificationCycles,
};
}
function tuiContext(results: unknown[], sessionId = "session-1") {
const notifications: string[] = [];
const renders: string[][] = [];
let customCalls = 0;
const theme = {
fg: (_color: string, text: string) => text,
bg: (_color: string, text: string) => text,
bold: (text: string) => text,
};
const unsupported = () => {
throw new Error("built-in bounded dialog must not be called");
};
const context = {
hasUI: true,
mode: "tui",
sessionManager: {
getSessionId: () => sessionId,
},
ui: {
custom: async (
factory: (
tui: { requestRender(): void; terminal: { columns: number; rows: number } },
overlayTheme: typeof theme,
keybindings: object,
done: (value: unknown) => void,
) => { render(width: number): string[] },
) => {
customCalls++;
const component = factory(
{
requestRender() {},
terminal: { columns: 100, rows: 24 },
},
theme,
{},
() => {},
);
renders.push(component.render(100));
return results.shift();
},
select: unsupported,
confirm: unsupported,
input: unsupported,
editor: unsupported,
notify: (message: string) => notifications.push(message),
},
} as unknown as ExtensionCommandContext;
return { context, notifications, renders, customCalls: () => customCalls };
}
function interactiveTuiContext(inputs: string[][], sessionId = "session-1") {
let customCalls = 0;
const theme = {
fg: (_color: string, text: string) => text,
bg: (_color: string, text: string) => text,
bold: (text: string) => text,
};
const context = {
hasUI: true,
mode: "tui",
sessionManager: { getSessionId: () => sessionId },
ui: {
async custom(
factory: (
tui: { requestRender(): void; terminal: { columns: number; rows: number } },
overlayTheme: typeof theme,
keybindings: object,
done: (value: unknown) => void,
) => {
render(width: number): string[];
handleInput?(data: string): void;
},
) {
customCalls++;
let completed: unknown;
const component = factory(
{
requestRender() {},
terminal: { columns: 80, rows: 16 },
},
theme,
{},
(value) => { completed = value; },
);
component.render(56);
for (const input of inputs.shift() ?? []) component.handleInput?.(input);
return completed;
},
notify() {},
},
} as unknown as ExtensionCommandContext;
return { context, customCalls: () => customCalls };
}
const pi = {
getActiveTools: () => ["read"],
} as unknown as ExtensionAPI;
test("dashboard fresh-session action returns an immediate continuation result", async () => {
const workflow = run({
planArtifact: { path: "/state/plan.md", revision: 3, digest: "digest" },
gates: [
{
id: "context-gate",
kind: "implementation-context",
status: "open",
revision: 12,
openedAt: 1,
},
],
implementationContext: {
gateId: "context-gate",
runRevision: 12,
planRevision: 3,
preparationId: "prep-1",
activeParentSessionId: "session-1",
strategy: "fresh-session",
status: "handoff-pending",
method: "fresh-session",
},
});
const tui = tuiContext(["run-1", "continue"]);
const result = await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [workflow] } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
);
assert.deepEqual(result, { continueRunId: "run-1" });
assert.equal(tui.customCalls(), 2);
});
test("dashboard records Plan Approval using only custom TUI overlays", async () => {
const workflow = run({
planArtifact: {
path: "/state/artifacts/plan.md",
revision: 3,
digest: "digest",
},
});
const commands: WorkflowCommand[] = [];
const tui = tuiContext([
"run-1",
"decide",
"approve",
true,
undefined,
undefined,
]);
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [workflow] } as never,
engineForContext: async () =>
({
async execute(command: WorkflowCommand) {
commands.push(structuredClone(command));
return view(workflow);
},
}) as never,
},
tui.context,
);
const decision = commands[0];
assert.equal(decision?.action, "decide");
assert.deepEqual(
decision?.action === "decide" ? decision.decision : undefined,
{
kind: "approve-plan",
approved: true,
acceptedUiImpact: "minor",
},
);
assert.ok(tui.customCalls() >= 4);
});
test("prototype reader Escape returns to candidate actions without accepting", async () => {
const candidate = prototypeCandidate();
const workflow = run({
stages: run().stages.map((stage) =>
stage.kind === "prototype"
? { ...stage, status: "waiting" as const }
: stage,
),
gates: [
{
id: "prototype-gate",
kind: "prototype-acceptance",
status: "open",
revision: 12,
openedAt: 1,
},
],
prototypeCandidates: [candidate],
});
const commands: WorkflowCommand[] = [];
const tui = tuiContext([
"run-1",
"decide",
{ kind: "candidate", candidate },
"review",
undefined,
"back",
undefined,
undefined,
undefined,
]);
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [workflow] } as never,
engineForContext: async () =>
({
async execute(command: WorkflowCommand) {
commands.push(structuredClone(command));
return view(workflow);
},
}) as never,
},
tui.context,
);
assert.deepEqual(commands, []);
assert.ok(
tui.renders.filter((lines) => lines.join("\n").includes("Prototype 2"))
.length >= 2,
);
});
test("prototype acceptance requires the explicit candidate Accept action", async () => {
const candidate = prototypeCandidate();
const workflow = run({
gates: [
{
id: "prototype-gate",
kind: "prototype-acceptance",
status: "open",
revision: 12,
openedAt: 1,
},
],
prototypeCandidates: [candidate],
});
const commands: WorkflowCommand[] = [];
const tui = tuiContext([
"run-1",
"decide",
{ kind: "candidate", candidate },
"review",
undefined,
"accept",
true,
"",
undefined,
undefined,
]);
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [workflow] } as never,
engineForContext: async () =>
({
async execute(command: WorkflowCommand) {
commands.push(structuredClone(command));
return view(workflow);
},
}) as never,
},
tui.context,
);
const decision = commands[0];
assert.deepEqual(
decision?.action === "decide" ? decision.decision : undefined,
{
kind: "accept-prototype",
candidateId: candidate.id,
candidateRevision: candidate.revision,
browserReviewed: true,
},
);
});
test("Escape from run detail returns one level to the run list", async () => {
const tui = tuiContext(["run-1", undefined, undefined]);
const result = await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [run()] } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
);
assert.deepEqual(result, {});
assert.equal(tui.customCalls(), 3);
});
test("dashboard handleInput drives selection and Escape back-stack navigation", async () => {
const tui = interactiveTuiContext([["\r"], ["\x1b"], ["\x1b"]]);
const result = await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [run()] } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
);
assert.deepEqual(result, {});
assert.equal(tui.customCalls(), 3);
});
test("dashboard duplicate labels retain their canonical run selection IDs", async () => {
const workflows = [
run({ runId: "12345678-aaaa", featureName: "search-page" }),
run({ runId: "12345678-bbbb", featureName: "search-page" }),
];
const tui = tuiContext(["12345678-bbbb", undefined, undefined]);
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => workflows } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
);
const list = tui.renders[0]?.join("\n") ?? "";
assert.match(list, /search-page · 12345678-a/);
assert.match(list, /search-page · 12345678-b/);
assert.match(tui.renders[1]?.join("\n") ?? "", /search-page · 12345678-b/);
});
test("dashboard disambiguates duplicate labels across the full run list", async () => {
const workflows = Array.from({ length: 30 }, (_, index) =>
run({
runId:
index === 0
? "12345678-aaaa"
: index === 29
? "12345678-bbbb"
: `run-${index}`,
featureName:
index === 0 || index === 29 ? "shared-label" : `feature-${index}`,
}),
);
const tui = tuiContext([undefined]);
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => workflows } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
);
const rendered = tui.renders[0]?.join("\n") ?? "";
assert.match(rendered, /shared-label · 12345678-a/);
assert.match(rendered, /1-12 of 30/);
assert.doesNotMatch(rendered, /shared-label\s+WAITING/);
});
test("historical parent sessions receive inspection-only dashboard actions", async () => {
const workflow = run({
parentSessionIds: ["session-1", "session-2"],
planArtifact: { path: "/state/plan.md", revision: 3, digest: "digest" },
});
const tui = tuiContext(["run-1", undefined, undefined], "session-2");
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [workflow] } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
);
const actions = tui.renders[1]?.join("\n") ?? "";
assert.match(actions, /Review verified plan/);
assert.doesNotMatch(actions, /Record Gate Decision/);
assert.doesNotMatch(actions, /Continue context transition/);
});
test("dashboard verifies and opens retained report content", async () => {
const directory = await mkdtemp(path.join(tmpdir(), "workflow-dashboard-"));
try {
const reportPath = path.join(directory, "report.md");
const reportContents = "# Completed report\n\nVerified.";
await writeFile(reportPath, reportContents, "utf8");
const workflow = run({
status: "completed",
gates: [],
reportArtifact: {
path: reportPath,
digest: createHash("sha256").update(reportContents).digest("hex"),
revision: 12,
},
});
const tui = tuiContext([
"run-1",
"report",
undefined,
undefined,
undefined,
]);
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [workflow] } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
);
assert.ok(
tui.renders.some((lines) =>
lines.join("\n").includes("Completed report"),
),
);
assert.ok(tui.renders.some((lines) => lines.join("\n").includes("Verified.")));
} finally {
await rm(directory, { recursive: true, force: true });
}
});
test("retained artifact reader Escape returns through artifact and candidate selection", async () => {
const candidate = prototypeCandidate();
const workflow = run({ prototypeCandidates: [candidate] });
const tui = tuiContext([
"run-1",
"prototypes",
candidate,
candidate.artifactPath,
undefined,
undefined,
undefined,
undefined,
undefined,
]);
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [workflow] } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
);
assert.ok(
tui.renders.filter((lines) => lines.join("\n").includes("Retained artifact"))
.length >= 2,
);
assert.ok(
tui.renders.filter((lines) => lines.join("\n").includes("Prototype evidence"))
.length >= 2,
);
});
test("dashboard verifies and opens Assignment transcripts", async () => {
const directory = await mkdtemp(path.join(tmpdir(), "workflow-transcript-"));
try {
const transcriptPath = path.join(directory, "assignment.json");
const transcriptContents = '[{"role":"assistant","text":"verified"}]\n';
await writeFile(transcriptPath, transcriptContents, "utf8");
const workflow = run({
assignments: [
{
id: "assignment-1",
stageId: "review",
role: "review",
profileName: "independent-reviewer",
profileDigest: "digest",
selection: {
eligibleProfiles: [],
selectedProfileName: "independent-reviewer",
rationale: "Review",
},
status: "succeeded",
mutates: false,
ownedPaths: [],
transcriptArtifact: {
path: transcriptPath,
digest: createHash("sha256")
.update(transcriptContents)
.digest("hex"),
},
},
],
});
const assignment = workflow.assignments[0];
assert.ok(assignment);
const tui = tuiContext([
"run-1",
"transcripts",
assignment,
undefined,
undefined,
undefined,
undefined,
]);
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [workflow] } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
);
assert.ok(
tui.renders.some((lines) =>
lines.join("\n").includes('"text":"verified"'),
),
);
assert.ok(
tui.renders.filter((lines) =>
lines.join("\n").includes("Assignment transcripts"),
).length >= 2,
);
} finally {
await rm(directory, { recursive: true, force: true });
}
});
test("RPC and print dashboards stay linear and never call custom", async () => {
for (const mode of ["rpc", "print", "json"] as const) {
const notifications: string[] = [];
const ctx = {
hasUI: mode === "rpc",
mode,
sessionManager: { getSessionId: () => "session-1" },
ui: {
custom: async () => {
throw new Error("custom must not be called outside TUI mode");
},
notify: (message: string) => notifications.push(message),
},
} as unknown as ExtensionCommandContext;
assert.deepEqual(
await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [run()] } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
ctx,
),
{},
);
if (mode === "rpc") assert.match(notifications[0] ?? "", /search-page/);
}
});
test("dashboard reports the supplied run ID when direct inspection cannot resolve it", async () => {
const tui = tuiContext([]);
const result = await openWorkflowDashboard(
pi,
{
store: { listRuns: async () => [run()] } as never,
engineForContext: () => Promise.reject(new Error("unused")),
},
tui.context,
"missing-run-id",
);
assert.deepEqual(result, {});
assert.deepEqual(tui.notifications, [
"Workflow Run missing-run-id was not found for this session.",
]);
});