@huseynsnmz/pi-board 0.15.3 → 0.17.0

package/docs/superpowers/plans/2026-10-23-issue-225-monotonic-foreground-state.md

@@ -1,450 +0,0 @@
-# Monotonic Foreground State Rebuild Implementation Plan
-
-> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
-
-**Goal:** Stop a stale `needs_input` projection from resurrecting the "Needs answer" group after a foreground question is answered.
-
-**Architecture:** A new module-level, pure projection cache stores the freshest status fields with a strictly increasing per-view stamp. `writeForegroundState` stamps and remembers; `syncRowEvent` adopts the snapshot when it is strictly newer than the disk rebuild (`statusFromRow` now carries `updatedAt`).
-
-**Tech Stack:** Node 22, `node:test` + `assert/strict`, plain ESM `.mjs`, no build step.
-
-**Spec:** `docs/superpowers/specs/2026-10-22-issue-225-monotonic-foreground-state-design.md`
-
-## Global Constraints
-
-- Tabs for indentation; English comments and commit messages.
-- Conventional commit subjects; the PR number is appended at merge time.
-- Run `npm run verify` before claiming completion (`npm test` alone skips the perf gate and coverage thresholds).
-- Do not change the reducer, `foreground-preview-cache.mjs`, the coordinator, or manual-verdict writes.
-- Cache keys are view ids; a view id is never reused (ids are generated), so leftover entries are only a memory concern.
-
-## Review Focus
-
-- **Same-millisecond bursts**: start/end of a question inside one millisecond must still order correctly (`nextStamp`, not `Date.now`).
-- **Manual verdicts**: a `completed` / `holding` row (`autoState === null`) must never be overwritten by a cached projection.
-- **Clone isolation**: mutating `status.pendingQuestions` after `remember`/`adopt` must not change the cache.
-- **Other views**: stamps and entries are per view id; one view's events must not order another's.
-- **Non-question fields untouched**: `review` / `diagnostics` / `autoState` / `lastVisitedAt` are not copied by adoption.
-
----
-
-### Task 1: `foreground-projection-cache` module (A1–A3, A6-unit)
-
-**Files:**
-- Create: `src/core/foreground-projection-cache.mjs`
-- Test: `test/foreground-projection-cache.test.mjs` (new file)
-
-**Interfaces:**
-- Produces: `createForegroundProjectionCache()` → `{ nextStamp(viewId), remember(viewId, status, stamp), adopt(viewId, status), forget(viewId), clear(), size() }` and the module-level singleton `foregroundProjectionCache`.
-
-- [ ] **Step 1: Write the failing tests**
-
-Create `test/foreground-projection-cache.test.mjs`:
-
-```js
-import assert from "node:assert/strict";
-import { test } from "node:test";
-import { createForegroundProjectionCache } from "../src/core/foreground-projection-cache.mjs";
-
-function status(overrides = {}) {
-	return {
-		semanticState: "working",
-		processState: "alive",
-		summary: "Working",
-		currentTool: null,
-		question: null,
-		pendingQuestions: [],
-		error: null,
-		lastActivityAt: 100,
-		latestAssistantPreview: "",
-		lastAgentActivityAt: null,
-		updatedAt: 100,
-		...overrides,
-	};
-}
-
-test("adopt copies the snapshot only when the cached stamp is strictly newer", () => {
-	const cache = createForegroundProjectionCache();
-	cache.remember("v1", status({ semanticState: "working", summary: "Done" }), 200);
-	const stale = status({ semanticState: "needs_input", question: "Q?", pendingQuestions: [{ toolCallId: "q1", question: "Q?" }] });
-	assert.equal(cache.adopt("v1", stale), true);
-	assert.equal(stale.semanticState, "working");
-	assert.equal(stale.summary, "Done");
-	assert.equal(stale.question, null);
-	assert.deepEqual(stale.pendingQuestions, []);
-	assert.equal(stale.updatedAt, 200);
-
-	const equal = status({ semanticState: "needs_input" });
-	equal.updatedAt = 200;
-	assert.equal(cache.adopt("v1", equal), false);
-	assert.equal(equal.semanticState, "needs_input");
-
-	const newer = status({ semanticState: "completed" });
-	newer.updatedAt = 300;
-	assert.equal(cache.adopt("v1", newer), false);
-	assert.equal(newer.semanticState, "completed");
-});
-
-test("non-whitelisted fields are never copied", () => {
-	const cache = createForegroundProjectionCache();
-	cache.remember("v1", status({ summary: "cached" }), 200);
-	const live = status({ summary: "disk", updatedAt: 100 });
-	live.review = { ready: true };
-	live.diagnostics = { code: "keep" };
-	live.autoState = { state: "keep" };
-	cache.adopt("v1", live);
-	assert.equal(live.summary, "cached");
-	assert.deepEqual(live.review, { ready: true });
-	assert.deepEqual(live.diagnostics, { code: "keep" });
-	assert.deepEqual(live.autoState, { state: "keep" });
-});
-
-test("remember and adopt clone nested values in both directions", () => {
-	const cache = createForegroundProjectionCache();
-	const source = status({ currentTool: { name: "bash", path: null }, pendingQuestions: [{ toolCallId: "q1", question: "Q?" }] });
-	cache.remember("v1", source, 200);
-	source.pendingQuestions[0].question = "mutated";
-	source.currentTool.name = "mutated";
-
-	const target = status({ updatedAt: 100 });
-	assert.equal(cache.adopt("v1", target), true);
-	assert.equal(target.pendingQuestions[0].question, "Q?");
-	assert.equal(target.currentTool.name, "bash");
-	target.pendingQuestions[0].question = "mutated again";
-	const again = status({ updatedAt: 100 });
-	cache.adopt("v1", again);
-	assert.equal(again.pendingQuestions[0].question, "Q?");
-});
-
-test("nextStamp is strictly increasing per view even when the clock is frozen", () => {
-	const cache = createForegroundProjectionCache();
-	const realNow = Date.now;
-	Date.now = () => 1_700_000_000_000;
-	try {
-		const a1 = cache.nextStamp("v1");
-		const a2 = cache.nextStamp("v1");
-		const b1 = cache.nextStamp("v2");
-		assert.ok(a2 > a1);
-		assert.equal(a1, 1_700_000_000_000);
-		assert.equal(b1, 1_700_000_000_000);
-		assert.ok(a2 === b1 || a2 > b1);
-	} finally {
-		Date.now = realNow;
-	}
-});
-
-test("forget drops the entry", () => {
-	const cache = createForegroundProjectionCache();
-	cache.remember("v1", status(), 200);
-	assert.equal(cache.size(), 1);
-	assert.equal(cache.forget("v1"), true);
-	assert.equal(cache.size(), 0);
-	const stale = status({ updatedAt: 100 });
-	assert.equal(cache.adopt("v1", stale), false);
-});
-```
-
-- [ ] **Step 2: Run the tests to verify they fail**
-
-Run: `node --test test/foreground-projection-cache.test.mjs`
-Expected: FAIL — `Cannot find module '../src/core/foreground-projection-cache.mjs'`.
-
-- [ ] **Step 3: Implement the module**
-
-Create `src/core/foreground-projection-cache.mjs`:
-
-```js
-/**
- * Process-local read-your-writes cache for foreground status projections
- * (issue #225).
- *
- * `syncRowEvent` rebuilds its status from `state.json` on every event, while
- * `writeForegroundState` is fire-and-forget: two events can have two
- * `sync_foreground` commands in flight, and an older projection can materialize
- * last. Because the rebuilt status carries no freshness signal, the next event
- * reduces onto the stale state and writes it back — a cleared question comes
- * back as `needs_input` until a terminal event lands.
- *
- * This cache stores the freshest whitelisted status fields with a strictly
- * increasing per-view stamp (`nextStamp`), so a disk rebuild that is older than
- * the last write is corrected before the new event is reduced.
- *
- * Module-level by necessity: `serviceFor()` creates a new service instance per
- * call (src/index.ts), so a per-instance cache would be discarded between
- * events.
- */
-
-/** Fields a cached snapshot may restore on a stale rebuild. */
-const CACHED_FIELDS = [
-	"semanticState",
-	"processState",
-	"summary",
-	"currentTool",
-	"question",
-	"pendingQuestions",
-	"error",
-	"lastActivityAt",
-	"latestAssistantPreview",
-	"lastAgentActivityAt",
-];
-
-/** @param {any} value */
-function cloneValue(value) {
-	if (value === null || typeof value !== "object") return value;
-	return JSON.parse(JSON.stringify(value));
-}
-
-/** @param {any} status */
-function snapshot(status) {
-	/** @type {Record<string, unknown>} */
-	const fields = {};
-	for (const key of CACHED_FIELDS) fields[key] = cloneValue(status?.[key]);
-	return fields;
-}
-
-export function createForegroundProjectionCache() {
-	/** @type {Map<string, { stamp: number, fields: Record<string, unknown> }>} */
-	const known = new Map();
-	/** @type {Map<string, number>} */
-	const stamps = new Map();
-
-	function nextStamp(viewId) {
-		const stamp = Math.max(Date.now(), (stamps.get(viewId) ?? 0) + 1);
-		stamps.set(viewId, stamp);
-		return stamp;
-	}
-
-	function remember(viewId, status, stamp) {
-		if (!viewId || !status || typeof stamp !== "number" || !Number.isFinite(stamp)) return;
-		known.set(viewId, { stamp, fields: snapshot(status) });
-	}
-
-	function adopt(viewId, status) {
-		if (!viewId || !status) return false;
-		const entry = known.get(viewId);
-		if (!entry) return false;
-		const at = typeof status.updatedAt === "number" ? status.updatedAt : 0;
-		if (entry.stamp <= at) return false;
-		for (const key of CACHED_FIELDS) status[key] = cloneValue(entry.fields[key]);
-		status.updatedAt = entry.stamp;
-		return true;
-	}
-
-	function forget(viewId) {
-		stamps.delete(viewId);
-		return known.delete(viewId);
-	}
-
-	function clear() {
-		known.clear();
-		stamps.clear();
-	}
-
-	function size() {
-		return known.size;
-	}
-
-	return { nextStamp, remember, adopt, forget, clear, size };
-}
-
-/** Shared cache used by the runtime service (module-level: see module doc). */
-export const foregroundProjectionCache = createForegroundProjectionCache();
-```
-
-- [ ] **Step 4: Run the tests to verify they pass**
-
-Run: `node --test test/foreground-projection-cache.test.mjs`
-Expected: PASS (5 tests).
-
-- [ ] **Step 5: Commit**
-
-```bash
-git add src/core/foreground-projection-cache.mjs test/foreground-projection-cache.test.mjs
-git commit -m "feat: foreground projection cache with monotonic per-view stamps (#225)"
-```
-
----
-
-### Task 2: Wire the cache into the service (A4–A7, A8)
-
-**Files:**
-- Modify: `src/runtime/service.mjs` (`statusFromRow` :814, `writeForegroundState` :854, `syncRowEvent` :1540, eviction `:775` and `:2131`)
-- Modify: `src/core/types.mjs` (`RunStatus` typedef)
-- Test: `test/service.test.mjs` (append after the existing foreground-sync tests)
-
-**Interfaces:**
-- Consumes: `foregroundProjectionCache` from Task 1 (`nextStamp` / `remember` / `adopt` / `forget`).
-- Produces: no public API change; `RunStatus` gains optional `updatedAt`.
-
-- [ ] **Step 1: Write the failing integration tests**
-
-Append to `test/service.test.mjs`:
-
-```js
-test("a stale needs_input snapshot cannot resurrect a cleared question (issue #225)", async () => {
-	const root = freshRoot();
-	const prevCoordinator = setEnv("AGENT_BOARD_COORDINATOR", "off");
-	const realNow = Date.now;
-	Date.now = () => 1_700_000_000_000;
-	try {
-		createView(root, { id: "v1", name: "a", cwd: "/r" });
-		const svc = service(root);
-		assert.equal(await svc.syncHostedEvent("v1", {
-			type: "tool_execution_start",
-			toolCallId: "q1",
-			toolName: "ask_questions",
-			args: { questions: [{ question: "Choose a mode?" }] },
-		}), true);
-		const stale = readState(root, "v1");
-		assert.equal(stale.semanticState, "needs_input");
-
-		assert.equal(await svc.syncHostedEvent("v1", { type: "tool_execution_end", toolCallId: "q1", toolName: "ask_questions", isError: false }), true);
-		// Simulate the older projection materializing after the newer one.
-		writeState(root, stale);
-
-		assert.equal(await svc.syncHostedEvent("v1", { type: "tool_execution_start", toolCallId: "t1", toolName: "bash", args: { command: "ls" } }), true);
-		const next = readState(root, "v1");
-		assert.equal(next.semanticState, "working");
-		assert.equal(next.needsInput, false);
-		assert.deepEqual(next.pendingQuestions, []);
-	} finally {
-		Date.now = realNow;
-		setEnv("AGENT_BOARD_COORDINATOR", prevCoordinator);
-		rmSync(root, { recursive: true, force: true });
-	}
-});
-
-test("a manual verdict survives a foreground tool event (issue #225)", async () => {
-	const root = freshRoot();
-	try {
-		createView(root, { id: "v1", name: "a", cwd: "/r" });
-		const done = readState(root, "v1");
-		done.semanticState = "completed";
-		done.processState = "exited";
-		done.autoState = null;
-		done.summary = "Done";
-		done.updatedAt = Date.now() - 1000;
-		writeState(root, done);
-
-		const svc = service(root, { sendStateCommand: async () => ({ status: "rejected", reason: "manual_fence" }) });
-		await svc.syncHostedEvent("v1", { type: "tool_execution_start", toolCallId: "t1", toolName: "bash", args: { command: "ls" } });
-		const next = readState(root, "v1");
-		assert.equal(next.semanticState, "completed");
-		assert.equal(next.autoState, null);
-	} finally {
-		rmSync(root, { recursive: true, force: true });
-	}
-});
-```
-
-- [ ] **Step 2: Run them to verify the first fails**
-
-Run: `node --test --test-name-pattern="issue #225" test/service.test.mjs`
-Expected: the stale-snapshot test FAILS with `actual: 'needs_input'`, `expected: 'working'` (pre-fix the disk rebuild wins); the manual-verdict test passes (it is a regression guard).
-
-- [ ] **Step 3: Add `updatedAt` to `RunStatus`**
-
-In `src/core/types.mjs`, inside the `RunStatus` typedef, after `lastEventAt`:
-
-```js
- * @property {number} [updatedAt] In-memory foreground rebuild freshness (issue #225);
- *   not persisted on run status files.
-```
-
-- [ ] **Step 4: Wire `statusFromRow`**
-
-In `statusFromRow`, after `lastActivityAt: s.lastActivityAt,` add:
-
-```js
-			updatedAt: typeof s.updatedAt === "number" ? s.updatedAt : 0,
-```
-
-- [ ] **Step 5: Stamp and remember in `writeForegroundState`**
-
-Replace the first two statements of `writeForegroundState`:
-
-```js
-	async function writeForegroundState(row, status, opts = {}) {
-		const projected = projectViewState(status, Date.now(), readState(root, row.meta.id) ?? row.state ?? null);
-```
-
-with:
-
-```js
-	async function writeForegroundState(row, status, opts = {}) {
-		// Issue #225: a strictly increasing stamp makes the background rebuild able
-		// to tell a stale disk snapshot from this write, even within one millisecond.
-		const stamp = foregroundProjectionCache.nextStamp(row.meta.id);
-		const projected = projectViewState(status, stamp, readState(root, row.meta.id) ?? row.state ?? null);
-		foregroundProjectionCache.remember(row.meta.id, status, stamp);
-```
-
-- [ ] **Step 6: Adopt in `syncRowEvent`**
-
-After `foregroundPreviewCache.backfill(row.meta.id, status);` add:
-
-```js
-		// Issue #225: correct a stale disk rebuild before the new event is reduced.
-		foregroundProjectionCache.adopt(row.meta.id, status);
-```
-
-- [ ] **Step 7: Evict alongside the preview cache**
-
-At `archiveView` (next to `foregroundPreviewCache.forget(viewId);`) and at `archiveByState` (next to `foregroundPreviewCache.forget(row.meta.id);`) add the matching call:
-
-```js
-		foregroundProjectionCache.forget(viewId);
-```
-
-```js
-				foregroundProjectionCache.forget(row.meta.id);
-```
-
-- [ ] **Step 8: Add the import**
-
-At the top of `src/runtime/service.mjs`, next to the preview-cache import:
-
-```js
-import { foregroundProjectionCache } from "../core/foreground-projection-cache.mjs";
-```
-
-- [ ] **Step 9: Run the tests**
-
-Run: `node --test --test-name-pattern="issue #225" test/service.test.mjs`
-Expected: PASS (both).
-
-Then run: `npm test`
-Expected: PASS — all existing foreground-sync, preview-cache, archive, and reducer tests stay green. (A7)
-
-- [ ] **Step 10: Run the full gate**
-
-Run: `npm run verify`
-Expected: PASS — typecheck, perf gate, tests, coverage thresholds (85/80/70), pack dry-run. (A8)
-
-- [ ] **Step 11: Commit**
-
-```bash
-git add src/runtime/service.mjs src/core/types.mjs test/service.test.mjs
-git commit -m "fix: keep foreground rows monotonic across out-of-order state writes (#225)"
-```
-
----
-
-### Task 3: Post-merge live verification (manual, not a code task)
-
-**Files:** none.
-
-- [ ] **Step 1: Update the board clone the running process loads**
-
-```bash
-cd ~/.pi/agent/git/github.com/zhuxixi/pi-agent-board
-git checkout -- package-lock.json
-git pull
-git log -1 --oneline
-```
-
-- [ ] **Step 2: Restart Pi / the board service**, open the dashboard.
-
-- [ ] **Step 3: Trigger a question** (`ask_user_question` or `question`), answer it, and keep watching.
-
-Expected: the row returns to **RUNNING** within about a second, not at turn end. (U1)