oma-coding-agent 1.1.4

package/src/tools/todo.ts

@@ -0,0 +1,938 @@
+import type { AgentTool, AgentToolContext, AgentToolResult, AgentToolUpdateCallback } from "@oh-my-pi/pi-agent-core";
+import type { ToolExample } from "@oh-my-pi/pi-ai";
+import type { Component } from "@oh-my-pi/pi-tui";
+import { Text } from "@oh-my-pi/pi-tui";
+import { prompt } from "@oh-my-pi/pi-utils";
+import { type } from "arktype";
+import chalk from "chalk";
+import type { RenderResultOptions } from "../extensibility/custom-tools/types";
+import type { Theme } from "../modes/theme/theme";
+import todoDescription from "../prompts/tools/todo.md" with { type: "text" };
+import type { ToolSession } from "../sdk";
+import type { SessionEntry } from "../session/session-entries";
+import { framedBlock, renderStatusLine, renderTreeList } from "../tui";
+import { formatErrorDetail, PREVIEW_LIMITS } from "./render-utils";
+
+// =============================================================================
+// Types
+// =============================================================================
+
+export type TodoStatus = "pending" | "in_progress" | "completed" | "abandoned";
+
+export interface TodoItem {
+	content: string;
+	status: TodoStatus;
+}
+
+export interface TodoPhase {
+	name: string;
+	tasks: TodoItem[];
+}
+
+export interface TodoCompletionTransition {
+	phase: string;
+	content: string;
+}
+
+export interface TodoToolDetails {
+	phases: TodoPhase[];
+	storage: "session" | "memory";
+	completedTasks?: TodoCompletionTransition[];
+}
+
+// =============================================================================
+// Schema
+// =============================================================================
+
+const TodoOp = type('"init" | "start" | "done" | "rm" | "drop" | "append" | "view"').describe("operation to apply");
+
+const InitListEntry = type({
+	phase: type("string").describe("phase name"),
+	items: type("string").describe("task content").array().atLeastLength(1).describe("tasks for this phase"),
+});
+
+const TodoOpEntry = type({
+	op: TodoOp,
+	"list?": InitListEntry.array().describe("phased task list (init)"),
+	"task?": type("string").describe("task content"),
+	"phase?": type("string").describe("phase name"),
+	"items?": type("string").describe("task content").array().atLeastLength(1).describe("tasks to append"),
+});
+
+const todoSchema = type({
+	ops: TodoOpEntry.array().atLeastLength(1).describe("ordered todo operations"),
+}).describe("apply ordered todo operations");
+
+type TodoParams = TodoSchema;
+type TodoSchema = typeof todoSchema.infer;
+type TodoOpEntryValue = TodoParams["ops"][number];
+
+// =============================================================================
+// State helpers
+// =============================================================================
+
+function findTaskByContent(phases: TodoPhase[], content: string): { task: TodoItem; phase: TodoPhase } | undefined {
+	for (const phase of phases) {
+		const task = phase.tasks.find(t => t.content === content);
+		if (task) return { task, phase };
+	}
+	return undefined;
+}
+
+function findPhaseByName(phases: TodoPhase[], name: string): TodoPhase | undefined {
+	return phases.find(phase => phase.name === name);
+}
+
+function cloneTask(task: TodoItem): TodoItem {
+	return { content: task.content, status: task.status };
+}
+
+function clonePhases(phases: TodoPhase[]): TodoPhase[] {
+	return phases.map(phase => ({ name: phase.name, tasks: phase.tasks.map(cloneTask) }));
+}
+
+function todoTransitionKey(phase: string, content: string): string {
+	return `${phase}\u0000${content}`;
+}
+
+function getCompletionTransitions(previous: TodoPhase[], updated: TodoPhase[]): TodoCompletionTransition[] {
+	const previousStatuses = new Map<string, TodoStatus>();
+	for (const phase of previous) {
+		for (const task of phase.tasks) {
+			previousStatuses.set(todoTransitionKey(phase.name, task.content), task.status);
+		}
+	}
+
+	const transitions: TodoCompletionTransition[] = [];
+	for (const phase of updated) {
+		for (const task of phase.tasks) {
+			if (task.status !== "completed") continue;
+			const previousStatus = previousStatuses.get(todoTransitionKey(phase.name, task.content));
+			if (previousStatus && previousStatus !== "completed") {
+				transitions.push({ phase: phase.name, content: task.content });
+			}
+		}
+	}
+	return transitions;
+}
+
+function normalizeInProgressTask(phases: TodoPhase[]): void {
+	const orderedTasks = phases.flatMap(phase => phase.tasks);
+	if (orderedTasks.length === 0) return;
+
+	const inProgressTasks = orderedTasks.filter(task => task.status === "in_progress");
+	if (inProgressTasks.length > 1) {
+		for (const task of inProgressTasks.slice(1)) {
+			task.status = "pending";
+		}
+	}
+
+	if (inProgressTasks.length > 0) return;
+
+	const firstPendingTask = orderedTasks.find(task => task.status === "pending");
+	if (firstPendingTask) firstPendingTask.status = "in_progress";
+}
+
+export const USER_TODO_EDIT_CUSTOM_TYPE = "user_todo_edit";
+
+export function getLatestTodoPhasesFromEntries(entries: SessionEntry[]): TodoPhase[] {
+	for (let i = entries.length - 1; i >= 0; i--) {
+		const entry = entries[i];
+		if (entry.type === "custom" && entry.customType === USER_TODO_EDIT_CUSTOM_TYPE) {
+			const data = entry.data as { phases?: unknown } | undefined;
+			if (data && Array.isArray(data.phases)) {
+				return clonePhases(data.phases as TodoPhase[]);
+			}
+			continue;
+		}
+		if (entry.type !== "message") continue;
+		const message = entry.message as { role?: string; toolName?: string; details?: unknown; isError?: boolean };
+		if (message.role !== "toolResult" || message.toolName !== "todo" || message.isError) continue;
+
+		const details = message.details as { phases?: unknown } | undefined;
+		if (!details || !Array.isArray(details.phases)) continue;
+
+		return clonePhases(details.phases as TodoPhase[]);
+	}
+
+	return [];
+}
+
+/**
+ * Pick the actionable window of tasks to display in the sticky todo panel.
+ *
+ * Returns up to `maxVisible` open (pending / in_progress) tasks in their
+ * original phase order, plus the count of remaining open tasks not shown so
+ * the caller can render a `+N more` hint. When every task in `tasks` is
+ * closed (completed or abandoned), returns the trailing `maxVisible` tasks
+ * with `hiddenOpenCount = 0`, so the panel keeps useful context until the
+ * active-phase pointer advances on the next `todo`.
+ *
+ * Task identity and order are preserved — this is a slice, never a sort.
+ */
+export function selectStickyTodoWindow(
+	tasks: TodoItem[],
+	maxVisible = 5,
+): { visible: TodoItem[]; hiddenOpenCount: number } {
+	const openTasks = tasks.filter(t => t.status === "pending" || t.status === "in_progress");
+	if (openTasks.length > 0) {
+		const visible = openTasks.slice(0, maxVisible);
+		return { visible, hiddenOpenCount: openTasks.length - visible.length };
+	}
+	const start = Math.max(0, tasks.length - maxVisible);
+	return { visible: tasks.slice(start), hiddenOpenCount: 0 };
+}
+
+/** Minimum overlap (after normalization) required for a substring match.
+ * Picked at six chars to admit single-word identifiers like "review" /
+ * "Sonnet" without admitting tiny common substrings like "test" / "fix"
+ * that would collide across unrelated todos. */
+const TODO_DESCRIPTION_MIN_OVERLAP = 6;
+
+function normalizeForTodoMatch(value: string): string {
+	return value
+		.toLowerCase()
+		.replace(/[^\p{L}\p{N}]+/gu, " ")
+		.trim();
+}
+
+/**
+ * Report whether `content` likely names the same work as any entry in
+ * `descriptions`. Used by the sticky todo panel to light up a pending todo
+ * when an in-flight subagent is doing the work for it, without requiring
+ * the caller to flip the todo's status.
+ *
+ * Matching is normalize-then-equal first (lowercased; punctuation and
+ * whitespace runs both collapsed to a single space; trimmed), with a
+ * substring fallback in either direction so minor wording drift
+ * ("Sonnet #2: bug scan" vs "Sonnet #2") still links up. The substring
+ * fallback requires at least {@link TODO_DESCRIPTION_MIN_OVERLAP} chars on
+ * the contained side.
+ */
+export function todoMatchesAnyDescription(content: string, descriptions: readonly string[]): boolean {
+	const target = normalizeForTodoMatch(content);
+	if (!target) return false;
+	for (const desc of descriptions) {
+		const candidate = normalizeForTodoMatch(desc);
+		if (!candidate) continue;
+		if (target === candidate) return true;
+		if (target.length >= TODO_DESCRIPTION_MIN_OVERLAP && candidate.includes(target)) return true;
+		if (candidate.length >= TODO_DESCRIPTION_MIN_OVERLAP && target.includes(candidate)) return true;
+	}
+	return false;
+}
+
+function resolveTaskOrError(
+	phases: TodoPhase[],
+	content: string | undefined,
+	errors: string[],
+): { task: TodoItem; phase: TodoPhase } | undefined {
+	if (!content) {
+		errors.push("Missing task content");
+		return undefined;
+	}
+	const hit = findTaskByContent(phases, content);
+	if (!hit) {
+		if (/^task-\d+$/.test(content)) {
+			errors.push(
+				`Task "${content}" not found. Tasks are referenced by content, not by IDs — pass the task's full text from the previous result.`,
+			);
+		} else {
+			const totalTasks = phases.reduce((sum, phase) => sum + phase.tasks.length, 0);
+			const hint = totalTasks === 0 ? " (todo list is empty — was it replaced or not yet created?)" : "";
+			errors.push(`Task "${content}" not found${hint}`);
+		}
+	}
+	return hit;
+}
+
+function resolvePhaseOrError(phases: TodoPhase[], name: string | undefined, errors: string[]): TodoPhase | undefined {
+	if (!name) {
+		errors.push("Missing phase name");
+		return undefined;
+	}
+	const phase = findPhaseByName(phases, name);
+	if (!phase) errors.push(`Phase "${name}" not found`);
+	return phase;
+}
+
+function getTaskTargets(phases: TodoPhase[], entry: TodoOpEntryValue, errors: string[]): TodoItem[] {
+	if (entry.task) {
+		const hit = resolveTaskOrError(phases, entry.task, errors);
+		return hit ? [hit.task] : [];
+	}
+	if (entry.phase) {
+		const phase = resolvePhaseOrError(phases, entry.phase, errors);
+		return phase ? [...phase.tasks] : [];
+	}
+	return phases.flatMap(phase => phase.tasks);
+}
+
+/** Phase name for `init` given a flat `items` list with no explicit `phase`. */
+const DEFAULT_INIT_PHASE = "Tasks";
+
+function initPhases(entry: TodoOpEntryValue, errors: string[]): TodoPhase[] {
+	// Models routinely flatten the single-phase init into `{op:"init", items:[...]}`
+	// (optionally with a bare `phase`) instead of the canonical
+	// `list: [{phase, items}]`. Accept that shape by synthesizing a one-phase list
+	// so a common, recoverable mistake isn't a hard error.
+	const list =
+		entry.list ??
+		(entry.items && entry.items.length > 0
+			? [{ phase: entry.phase ?? DEFAULT_INIT_PHASE, items: entry.items }]
+			: undefined);
+	if (!list) {
+		errors.push("Missing list for init operation");
+		return [];
+	}
+	// Duplicate phase names / task contents would be permanently unaddressable
+	// (every targeting op resolves the first match), so reject them up front.
+	const seenPhases = new Set<string>();
+	const seenTasks = new Set<string>();
+	for (const listEntry of list) {
+		if (seenPhases.has(listEntry.phase)) {
+			errors.push(`Duplicate phase "${listEntry.phase}" in init list`);
+		}
+		seenPhases.add(listEntry.phase);
+		for (const content of listEntry.items) {
+			if (seenTasks.has(content)) {
+				errors.push(`Duplicate task "${content}" in init list`);
+			}
+			seenTasks.add(content);
+		}
+	}
+	return list.map(listEntry => ({
+		name: listEntry.phase,
+		tasks: listEntry.items.map<TodoItem>(content => ({ content, status: "pending" })),
+	}));
+}
+
+function appendItems(phases: TodoPhase[], entry: TodoOpEntryValue, errors: string[]): TodoPhase[] {
+	if (!entry.phase) {
+		errors.push("Missing phase name for append operation");
+		return phases;
+	}
+	if (!entry.items || entry.items.length === 0) {
+		errors.push("Missing items for append operation");
+		return phases;
+	}
+
+	// Validate the whole batch before mutating so a failing op reports every
+	// duplicate and leaves nothing half-applied.
+	const seen = new Set<string>();
+	let hasDuplicate = false;
+	for (const content of entry.items) {
+		if (seen.has(content) || findTaskByContent(phases, content)) {
+			errors.push(`Task "${content}" already exists`);
+			hasDuplicate = true;
+		}
+		seen.add(content);
+	}
+	if (hasDuplicate) return phases;
+
+	let phase = findPhaseByName(phases, entry.phase);
+	if (!phase) {
+		phase = { name: entry.phase, tasks: [] };
+		phases.push(phase);
+	}
+
+	for (const content of entry.items) {
+		phase.tasks.push({ content, status: "pending" });
+	}
+	return phases;
+}
+
+function removeTasks(phases: TodoPhase[], entry: TodoOpEntryValue, errors: string[]): TodoPhase[] {
+	if (entry.task) {
+		const hit = resolveTaskOrError(phases, entry.task, errors);
+		if (!hit) return phases;
+		hit.phase.tasks = hit.phase.tasks.filter(candidate => candidate !== hit.task);
+		return phases;
+	}
+	if (entry.phase) {
+		const phase = resolvePhaseOrError(phases, entry.phase, errors);
+		if (!phase) return phases;
+		phase.tasks = [];
+		return phases;
+	}
+	for (const phase of phases) {
+		phase.tasks = [];
+	}
+	return phases;
+}
+
+function applyEntry(phases: TodoPhase[], entry: TodoOpEntryValue, errors: string[]): TodoPhase[] {
+	switch (entry.op) {
+		case "init":
+			return initPhases(entry, errors);
+		case "start": {
+			const hit = resolveTaskOrError(phases, entry.task, errors);
+			if (!hit) return phases;
+			for (const phase of phases) {
+				for (const candidate of phase.tasks) {
+					if (candidate.status === "in_progress" && candidate !== hit.task) {
+						candidate.status = "pending";
+					}
+				}
+			}
+			hit.task.status = "in_progress";
+			return phases;
+		}
+		case "done": {
+			for (const task of getTaskTargets(phases, entry, errors)) {
+				task.status = "completed";
+			}
+			return phases;
+		}
+		case "drop": {
+			for (const task of getTaskTargets(phases, entry, errors)) {
+				task.status = "abandoned";
+			}
+			return phases;
+		}
+		case "rm":
+			return removeTasks(phases, entry, errors);
+		case "append":
+			return appendItems(phases, entry, errors);
+		case "view":
+			return phases;
+	}
+}
+
+function applyParams(phases: TodoPhase[], params: TodoParams): { phases: TodoPhase[]; errors: string[] } {
+	const errors: string[] = [];
+	let next = phases;
+	for (const entry of params.ops) {
+		next = applyEntry(next, entry, errors);
+	}
+	normalizeInProgressTask(next);
+	return { phases: next, errors };
+}
+
+/** Apply an array of `todo`-style ops to existing phases. Used by /todo slash command. */
+export function applyOpsToPhases(
+	currentPhases: TodoPhase[],
+	ops: TodoParams["ops"],
+): { phases: TodoPhase[]; errors: string[] } {
+	return applyParams(clonePhases(currentPhases), { ops });
+}
+
+// =============================================================================
+// Markdown round-trip
+// =============================================================================
+
+const STATUS_TO_MARKER: Record<TodoStatus, string> = {
+	pending: " ",
+	in_progress: "/",
+	completed: "x",
+	abandoned: "-",
+};
+
+/** Render todo phases as a Markdown checklist suitable for editing/copying. */
+export function phasesToMarkdown(phases: TodoPhase[]): string {
+	if (phases.length === 0) return "# Todos\n";
+	const out: string[] = [];
+	for (let i = 0; i < phases.length; i++) {
+		if (i > 0) out.push("");
+		out.push(`# ${phases[i].name}`);
+		for (const task of phases[i].tasks) {
+			out.push(`- [${STATUS_TO_MARKER[task.status]}] ${task.content}`);
+		}
+	}
+	return `${out.join("\n")}\n`;
+}
+
+const MARKER_TO_STATUS: Record<string, TodoStatus> = {
+	" ": "pending",
+	"": "pending",
+	x: "completed",
+	X: "completed",
+	"/": "in_progress",
+	">": "in_progress",
+	"-": "abandoned",
+	"~": "abandoned",
+};
+
+/** Parse a Markdown checklist back into todo phases. */
+export function markdownToPhases(md: string): { phases: TodoPhase[]; errors: string[] } {
+	const errors: string[] = [];
+	const phases: TodoPhase[] = [];
+	let currentPhase: TodoPhase | undefined;
+
+	const lines = md.split(/\r?\n/);
+	for (let lineNum = 0; lineNum < lines.length; lineNum++) {
+		const raw = lines[lineNum];
+
+		const trimmed = raw.trim();
+		if (!trimmed) continue;
+
+		const headingMatch = /^#{1,6}\s+(.+?)\s*$/.exec(trimmed);
+		if (headingMatch) {
+			currentPhase = { name: headingMatch[1].trim(), tasks: [] };
+			phases.push(currentPhase);
+			continue;
+		}
+
+		const taskMatch = /^[-*+]\s*\[(.?)\]\s+(.+?)\s*$/.exec(trimmed);
+		if (taskMatch) {
+			if (!currentPhase) {
+				currentPhase = { name: "Todos", tasks: [] };
+				phases.push(currentPhase);
+			}
+			const marker = taskMatch[1];
+			const status = MARKER_TO_STATUS[marker];
+			if (!status) {
+				errors.push(`Line ${lineNum + 1}: unknown status marker "[${marker}]" (use [ ], [x], [/], [-])`);
+				continue;
+			}
+			currentPhase.tasks.push({ content: taskMatch[2].trim(), status });
+			continue;
+		}
+
+		errors.push(`Line ${lineNum + 1}: unrecognized syntax "${trimmed}"`);
+	}
+
+	normalizeInProgressTask(phases);
+	return { phases, errors };
+}
+
+function formatSummary(phases: TodoPhase[], errors: string[], readOnly = false): string {
+	const tasks = phases.flatMap(phase => phase.tasks);
+	if (tasks.length === 0) {
+		if (errors.length > 0) return `Errors: ${errors.join("; ")}`;
+		return readOnly ? "Todo list is empty." : "Todo list cleared.";
+	}
+
+	const remainingByPhase = phases
+		.map(phase => ({
+			name: phase.name,
+			tasks: phase.tasks.filter(task => task.status === "pending" || task.status === "in_progress"),
+		}))
+		.filter(phase => phase.tasks.length > 0);
+	const remainingTasks = remainingByPhase.flatMap(phase => phase.tasks.map(task => ({ ...task, phase: phase.name })));
+
+	let currentIdx = phases.findIndex(phase =>
+		phase.tasks.some(task => task.status === "pending" || task.status === "in_progress"),
+	);
+	if (currentIdx === -1) currentIdx = phases.length - 1;
+	const current = phases[currentIdx];
+	const done = current.tasks.filter(task => task.status === "completed" || task.status === "abandoned").length;
+
+	const lines: string[] = [];
+	if (errors.length > 0) lines.push(`Errors: ${errors.join("; ")}`);
+	if (remainingTasks.length === 0) {
+		lines.push("Remaining items: none.");
+	} else {
+		lines.push(`Remaining items (${remainingTasks.length}):`);
+		for (const task of remainingTasks) {
+			lines.push(`  - ${task.content} [${task.status}] (${task.phase})`);
+		}
+	}
+	lines.push(
+		`Phase ${currentIdx + 1}/${phases.length} "${current.name}" — ${done}/${current.tasks.length} tasks complete`,
+	);
+	for (const phase of phases) {
+		lines.push(`  ${phase.name}:`);
+		for (const task of phase.tasks) {
+			const sym =
+				task.status === "completed"
+					? "✓"
+					: task.status === "in_progress"
+						? "→"
+						: task.status === "abandoned"
+							? "✗"
+							: "○";
+			lines.push(`    ${sym} ${task.content}`);
+		}
+	}
+	return lines.join("\n");
+}
+
+// =============================================================================
+// Tool Class
+// =============================================================================
+
+export class TodoTool implements AgentTool<typeof todoSchema, TodoToolDetails> {
+	readonly name = "todo";
+	readonly approval = "read" as const;
+	readonly label = "Todo";
+	readonly summary = "Write a structured todo list to track progress within a session";
+	readonly description: string;
+	readonly parameters = todoSchema;
+	readonly concurrency = "exclusive";
+	readonly strict = true;
+
+	readonly examples: readonly ToolExample<typeof todoSchema.infer>[] = [
+		{
+			caption: "Initial setup (multi-phase)",
+			call: {
+				ops: [
+					{
+						op: "init",
+						list: [
+							{ phase: "Foundation", items: ["Scaffold crate", "Wire workspace"] },
+							{ phase: "Auth", items: ["Port credential store", "Wire OAuth providers"] },
+							{ phase: "Verification", items: ["Run cargo test"] },
+						],
+					},
+				],
+			},
+		},
+		{
+			caption: "View current state (read-only)",
+			call: {
+				ops: [{ op: "view" }],
+			},
+		},
+		{
+			caption: "Initial setup (single phase)",
+			call: {
+				ops: [
+					{
+						op: "init",
+						list: [{ phase: "Implementation", items: ["Apply fix", "Run tests"] }],
+					},
+				],
+			},
+		},
+		{
+			caption: "Complete one task",
+			call: {
+				ops: [{ op: "done", task: "Wire workspace" }],
+			},
+		},
+		{
+			caption: "Complete a whole phase",
+			call: {
+				ops: [{ op: "done", phase: "Auth" }],
+			},
+		},
+		{
+			caption: "Remove all tasks",
+			call: {
+				ops: [{ op: "rm" }],
+			},
+		},
+		{
+			caption: "Drop one task",
+			call: {
+				ops: [{ op: "drop", task: "Run cargo test" }],
+			},
+		},
+		{
+			caption: "Append tasks to a phase",
+			call: {
+				ops: [{ op: "append", phase: "Auth", items: ["Handle retries", "Run tests"] }],
+			},
+		},
+	];
+	readonly loadMode = "discoverable";
+	constructor(private readonly session: ToolSession) {
+		this.description = prompt.render(todoDescription);
+	}
+
+	async execute(
+		_toolCallId: string,
+		params: TodoParams,
+		_signal?: AbortSignal,
+		_onUpdate?: AgentToolUpdateCallback<TodoToolDetails>,
+		_context?: AgentToolContext,
+	): Promise<AgentToolResult<TodoToolDetails>> {
+		const previousPhases = clonePhases(this.session.getTodoPhases?.() ?? []);
+		// Pure-view calls are reads: no normalization, no state write.
+		const readOnly = params.ops.every(entry => entry.op === "view");
+		const { phases: updated, errors } = readOnly
+			? { phases: previousPhases, errors: [] as string[] }
+			: applyParams(clonePhases(previousPhases), params);
+		// A batch with any error is discarded wholesale: persisting a
+		// half-applied batch makes the natural retry hit "already exists" for
+		// the ops that did land. State and rendered summary stay at previous.
+		const failed = errors.length > 0;
+		const effective = failed ? previousPhases : updated;
+		const completedTasks = readOnly || failed ? [] : getCompletionTransitions(previousPhases, updated);
+		if (!readOnly && !failed) this.session.setTodoPhases?.(updated);
+		const storage = this.session.getSessionFile() ? "session" : "memory";
+		const details: TodoToolDetails = { phases: effective, storage };
+		if (completedTasks.length > 0) details.completedTasks = completedTasks;
+
+		return {
+			content: [{ type: "text", text: formatSummary(effective, errors, readOnly) }],
+			details,
+			isError: errors.length > 0 ? true : undefined,
+		};
+	}
+}
+
+// =============================================================================
+// TUI Renderer
+// =============================================================================
+
+type TodoRenderArgs = {
+	ops?: Array<{
+		op?: string;
+		task?: string;
+		phase?: string;
+		items?: string[];
+	}>;
+};
+
+// =============================================================================
+// Phase numbering (display-only)
+// =============================================================================
+
+const ROMAN_PAIRS: Array<[number, string]> = [
+	[1000, "M"],
+	[900, "CM"],
+	[500, "D"],
+	[400, "CD"],
+	[100, "C"],
+	[90, "XC"],
+	[50, "L"],
+	[40, "XL"],
+	[10, "X"],
+	[9, "IX"],
+	[5, "V"],
+	[4, "IV"],
+	[1, "I"],
+];
+
+/** One-based ASCII roman numeral for display (I, II, III, IV, …). */
+export function phaseRomanNumeral(oneBasedIndex: number): string {
+	if (oneBasedIndex <= 0) return "";
+	let out = "";
+	let rem = oneBasedIndex;
+	for (const [value, sym] of ROMAN_PAIRS) {
+		while (rem >= value) {
+			out += sym;
+			rem -= value;
+		}
+	}
+	return out;
+}
+
+/** Display-only phase header: `I. Foundation`. State and prompts never see this. */
+export function formatPhaseDisplayName(name: string, oneBasedIndex: number): string {
+	return `${phaseRomanNumeral(oneBasedIndex)}. ${name}`;
+}
+
+export const TODO_STRIKE_HOLD_FRAMES = 2;
+export const TODO_STRIKE_REVEAL_FRAMES = 12;
+export const TODO_STRIKE_TOTAL_FRAMES = TODO_STRIKE_HOLD_FRAMES + TODO_STRIKE_REVEAL_FRAMES;
+const EMPTY_COMPLETION_KEYS = new Set<string>();
+const STRIKE_START = "\x1b[9m";
+const STRIKE_END = "\x1b[29m";
+
+function strikethroughText(text: string): string {
+	return `${STRIKE_START}${text}${STRIKE_END}`;
+}
+
+function partialStrikethrough(text: string, visibleChars: number): string {
+	if (visibleChars <= 0) return text;
+	const chars = [...text];
+	if (visibleChars >= chars.length) return strikethroughText(text);
+	return `${strikethroughText(chars.slice(0, visibleChars).join(""))}${chars.slice(visibleChars).join("")}`;
+}
+
+function strikeRevealCount(text: string, frame: number | undefined): number | undefined {
+	if (frame === undefined) return undefined;
+	if (frame <= TODO_STRIKE_HOLD_FRAMES) return 0;
+	const chars = [...text];
+	if (chars.length === 0) return undefined;
+	const revealFrame = Math.min(frame - TODO_STRIKE_HOLD_FRAMES, TODO_STRIKE_REVEAL_FRAMES);
+	return Math.ceil((chars.length * revealFrame) / TODO_STRIKE_REVEAL_FRAMES);
+}
+
+function formatTodoLine(
+	item: TodoItem,
+	uiTheme: Theme,
+	prefix: string,
+	completionKeys: Set<string>,
+	frame: number | undefined,
+): string {
+	const checkbox = uiTheme.checkbox;
+	switch (item.status) {
+		case "completed": {
+			const revealCount = completionKeys.has(item.content) ? strikeRevealCount(item.content, frame) : undefined;
+			const content =
+				revealCount === undefined
+					? strikethroughText(item.content)
+					: partialStrikethrough(item.content, revealCount);
+			return uiTheme.fg("success", `${prefix}${checkbox.checked} ${content}`);
+		}
+		case "in_progress":
+			return uiTheme.fg("accent", `${prefix}${checkbox.unchecked} ${item.content}`);
+		case "abandoned":
+			return uiTheme.fg("error", `${prefix}${checkbox.unchecked} ${strikethroughText(item.content)}`);
+		default:
+			return uiTheme.fg("dim", `${prefix}${checkbox.unchecked} ${item.content}`);
+	}
+}
+
+/**
+ * Phases the latest update touched, plus the active (in_progress) phase.
+ * Returns `null` when there is no usable signal, meaning "render every phase
+ * fully" — this preserves the legacy view and the manual-expand path.
+ */
+function computeTouchedPhases(
+	args: TodoRenderArgs | undefined,
+	phases: TodoPhase[],
+	completedTasks: TodoCompletionTransition[],
+): Set<string> | null {
+	const touched = new Set<string>();
+	// The phase holding the in_progress task is where attention sits after the
+	// auto-promotion that follows every completion.
+	for (const phase of phases) {
+		if (phase.tasks.some(task => task.status === "in_progress")) touched.add(phase.name);
+	}
+	// Phases with a task that just transitioned to completed in this update.
+	for (const transition of completedTasks) touched.add(transition.phase);
+	// Phases explicitly named by the ops that ran. `init` replaces the whole
+	// list, so the entire plan is fresh and every phase counts as touched.
+	const ops = Array.isArray(args?.ops) ? args.ops : [];
+	for (const op of ops) {
+		if (!op || typeof op !== "object") continue;
+		if (op.op === "init") {
+			for (const phase of phases) touched.add(phase.name);
+			break;
+		}
+		if (typeof op.phase === "string" && op.phase) {
+			const named = phases.find(phase => phase.name === op.phase);
+			if (named) touched.add(named.name);
+		}
+		if (typeof op.task === "string" && op.task) {
+			const located = findTaskByContent(phases, op.task);
+			if (located) touched.add(located.phase.name);
+		}
+	}
+	return touched.size > 0 ? touched : null;
+}
+
+/** One-line summary for a collapsed (untouched) phase: dim header + progress. */
+function formatPhaseSummary(phase: TodoPhase, oneBasedIndex: number, uiTheme: Theme): string {
+	const total = phase.tasks.length;
+	const done = phase.tasks.filter(task => task.status === "completed").length;
+	const name = uiTheme.fg("dim", chalk.bold(formatPhaseDisplayName(phase.name, oneBasedIndex)));
+	return `${name}${uiTheme.fg("dim", `  ${done}/${total}`)}`;
+}
+
+export const todoToolRenderer = {
+	renderCall(args: TodoRenderArgs, options: RenderResultOptions, uiTheme: Theme): Component {
+		// `args` here is the raw partially-parsed JSON from the streaming
+		// tool-call delta and may not satisfy `TodoRenderArgs` at runtime:
+		// `parseStreamingJson` can hand back `{ ops: "[" }` mid-delta, or
+		// entries that are `null` / strings before fields stream. Guard
+		// against non-array `ops` and non-object entries so a malformed
+		// delta never breaks the TUI render loop (#2005).
+		const opsList = Array.isArray(args?.ops) ? args.ops : [];
+		const ops =
+			opsList.length === 0
+				? ["update"]
+				: opsList.map(entry => {
+						const e = entry && typeof entry === "object" ? entry : ({} as NonNullable<typeof entry>);
+						const parts = [e.op ?? "update"];
+						if (e.task) parts.push(e.task);
+						if (e.phase) parts.push(e.phase);
+						if (Array.isArray(e.items) && e.items.length) {
+							parts.push(`${e.items.length} item${e.items.length === 1 ? "" : "s"}`);
+						}
+						return parts.join(" ");
+					});
+		// No body worth boxing while the call streams — a lone status line reads
+		// cleaner than an empty frame. The container renders it without chrome.
+		const header = renderStatusLine(
+			{ icon: "pending", spinnerFrame: options?.spinnerFrame, title: "Todo", meta: ops },
+			uiTheme,
+		);
+		return new Text(header, 0, 0);
+	},
+
+	renderResult(
+		result: { content: Array<{ type: string; text?: string }>; details?: TodoToolDetails; isError?: boolean },
+		options: RenderResultOptions,
+		uiTheme: Theme,
+		args?: TodoRenderArgs,
+	): Component {
+		if (result.isError) {
+			const errorText = result.content?.find(content => content.type === "text")?.text ?? "Todo operation failed";
+			const header = renderStatusLine({ icon: "error", title: "Todo" }, uiTheme);
+			return framedBlock(uiTheme, width => ({
+				header,
+				sections: [{ lines: formatErrorDetail(errorText, uiTheme).split("\n") }],
+				state: "error",
+				borderColor: "error",
+				width,
+			}));
+		}
+
+		const phases = (result.details?.phases ?? []).filter(phase => phase.tasks.length > 0);
+		const completedTasks = result.details?.completedTasks ?? [];
+		const completionKeysByPhase = new Map<string, Set<string>>();
+		for (const task of completedTasks) {
+			let keys = completionKeysByPhase.get(task.phase);
+			if (!keys) {
+				keys = new Set<string>();
+				completionKeysByPhase.set(task.phase, keys);
+			}
+			keys.add(task.content);
+		}
+		const allTasks = phases.flatMap(phase => phase.tasks);
+		const header = renderStatusLine(
+			{
+				iconOverride: uiTheme.styledSymbol("tool.todo", "accent"),
+				title: "Todo",
+				meta: [`${allTasks.length} tasks`],
+			},
+			uiTheme,
+		);
+		if (allTasks.length === 0) {
+			const fallback = result.content?.find(content => content.type === "text")?.text ?? "No todos";
+			return new Text(`${header}\n  ${uiTheme.fg("dim", fallback)}`, 0, 0);
+		}
+
+		return framedBlock(uiTheme, width => {
+			const { expanded, spinnerFrame } = options;
+			const multiPhase = phases.length > 1;
+			const indent = multiPhase ? "  " : "";
+			// Collapse phases this update didn't touch down to a one-line summary so
+			// a single task flip doesn't redraw every phase's full task list. The
+			// manual expand toggle (and the no-signal fallback) still shows all.
+			const touched = expanded || !multiPhase ? null : computeTouchedPhases(args, phases, completedTasks);
+			const bodyLines: string[] = [];
+			for (let p = 0; p < phases.length; p++) {
+				const phase = phases[p];
+				if (touched && !touched.has(phase.name)) {
+					bodyLines.push(formatPhaseSummary(phase, p + 1, uiTheme));
+					continue;
+				}
+				if (multiPhase) {
+					bodyLines.push(uiTheme.fg("accent", chalk.bold(formatPhaseDisplayName(phase.name, p + 1))));
+				}
+				const completionKeys = completionKeysByPhase.get(phase.name) ?? EMPTY_COMPLETION_KEYS;
+				const treeLines = renderTreeList(
+					{
+						items: phase.tasks,
+						expanded,
+						maxCollapsed: PREVIEW_LIMITS.COLLAPSED_ITEMS,
+						itemType: "todo",
+						truncateFrom: "start",
+						renderItem: todo => formatTodoLine(todo, uiTheme, "", completionKeys, spinnerFrame),
+					},
+					uiTheme,
+				);
+				for (const line of treeLines) {
+					bodyLines.push(`${indent}${line}`);
+				}
+			}
+			while (bodyLines.length > 0 && bodyLines[0].trim() === "") bodyLines.shift();
+			return {
+				header,
+				sections: bodyLines.length > 0 ? [{ lines: bodyLines }] : [],
+				state: options.isPartial ? "pending" : "success",
+				borderColor: "borderMuted",
+				applyBg: false,
+				width,
+			};
+		});
+	},
+	mergeCallAndResult: true,
+};