rovecode 0.3.3 → 0.4.0-beta.3

package/src/sextant/card-keys.ts

@@ -1,55 +0,0 @@
-/** Sextant card keys (port #43): the key route of the ONE modal card — ←→↑↓ (and Tab) move the
- *  selection, Enter resolves it (approval: once / always / deny; question: an option, the free-text
- *  row or skip), Esc dismisses (deny / null) and the free-text row takes typing and Backspace.
- *  Ported from the user's own sextant v0.4.0 prototype app.js:1503-1510 and :1434; split out of
- *  keys.ts for the line budget. Pure state mutation, no clock: the card leaves the state before
- *  its promise settles, so a re-entrant hook never sees a stale card. */
-
-import type { CardState, KeyEvent, SextantState } from "./types.ts";
-
-/** false = not a card key: it falls through to the prompt (typing a follow-up while a card waits) */
-export function onCardKey(s: SextantState, c: CardState, ev: KeyEvent): boolean {
-  const { name } = ev;
-  const n = c.kind === "approval" ? c.verdicts.length // allow · always · [all edits] · deny
-    : (c.prompt.options?.length ?? 0) + (c.prompt.allowFreeText !== false ? 1 : 0) + 1; // options · free text · skip
-  if (name === "left" || name === "up") return move(c, n, -1);
-  if (name === "right" || name === "down" || name === "tab") return move(c, n, 1);
-  if (name === "escape") { dismissCard(s); return true; }
-  if (c.kind === "approval") {
-    if (name !== "enter") return false;
-    s.card = null;
-    c.resolve(c.verdicts[c.selected] ?? "deny");
-    return true;
-  }
-  const opts = c.prompt.options?.length ?? 0, freeIdx = c.prompt.allowFreeText !== false ? opts : -1;
-  if (name === "enter") {
-    if (c.selected === n - 1) { dismissCard(s); return true; } // skip
-    if (c.selected !== freeIdx) { s.card = null; c.resolve({ kind: "option", index: c.selected }); return true; }
-    const t = c.freeText.trim();
-    if (t) { s.card = null; c.resolve({ kind: "text", text: t }); }
-    return true; // an empty free-text row is not an answer yet
-  }
-  if (c.selected !== freeIdx) return false;
-  if (name === "backspace") { // by code point: 😀 is one glyph, two UTF-16 units
-    c.freeText = [...c.freeText].slice(0, -1).join("");
-    return true;
-  }
-  const ch = name === "space" ? " " : ev.ch;
-  if (!ch || ev.alt) return false;
-  c.freeText += ch;
-  return true;
-}
-
-/** wrap the selection over the card's n rows */
-function move(c: CardState, n: number, d: -1 | 1): true {
-  const v = (c.selected + d + n) % n;
-  c.selected = v;
-  return true;
-}
-
-/** approval → deny, question → null; the card leaves the state before the promise settles */
-export function dismissCard(s: SextantState): void {
-  const c = s.card;
-  s.card = null;
-  if (c?.kind === "approval") c.resolve("deny"); else c?.resolve(null);
-}

package/src/sextant/context-source.ts

@@ -1,157 +0,0 @@
-/** The seam between the context overlay and core/context-report.ts.
- *
- *  draw-context.ts is pure over `ContextState` and never counts anything. This file is the only place that
- *  knows the report module exists: it lazily imports it (the boot rule — nothing in that dependency tree is
- *  loaded until someone opens /context), asks the LIVE runtime for the two rows a transcript cannot know,
- *  and maps the report onto the overlay's flat view.
- *
- *  The live runtime is the whole reason this overlay is worth having. `rovecode context` on a shell has to
- *  BUILD a runtime to learn the system prompt and the tool schemas, and in a fresh session those two are
- *  most of the window; here they are already in memory, so the number is complete without constructing
- *  anything. What is still missing is MCP tools — their schemas do not exist until a server is connected —
- *  and the panel says so on its own line rather than letting a complete-looking total imply otherwise.
- *
- *  Two numbers travel, never one. `raw` is what o200k counted; `estimated` is that corrected towards the
- *  model's own tokenizer. Presenting only the second would be presenting a corrected figure as a
- *  measurement, and someone comparing this panel against a provider's dashboard needs to know which is
- *  which. `scale.note` says where the factor came from, including when nobody has measured this model.
- *
- *  Nothing here throws. A runtime that cannot answer, a model the catalog does not know, an empty
- *  transcript — each is a state the overlay draws, because a panel that reports on the session must not be
- *  the thing that ends it. */
-
-import type { ContextState } from "./draw-context.ts";
-
-/** What the live runtime can tell us that the transcript cannot. Both optional: a session whose runtime
- *  is mid-rebuild still opens the panel, and `live: false` makes the total read as a floor. */
-export interface ContextFixed {
-  /** the system prompt a real turn would send — buildDef's, not the base alone */
-  system?: string;
-  /** rovecode's own tool schemas, serialized. MCP tools are deliberately absent. */
-  toolSchemas?: string;
-}
-
-/** the runtime shape this file needs, structurally — declared here so the overlay never imports the
- *  runtime's types and so a test can hand in three lines instead of a process */
-export interface LiveRuntime {
-  registry: { list(): { schema: unknown }[] };
-  // method syntax, and `unknown` for the prompt: the real AgentDefinition types `systemPrompt` as a string
-  // or a function over its own AgentVars, and pinning either shape here would make this file depend on the
-  // runtime's types — which is exactly what the overlay must not do. It is narrowed where it is read.
-  buildDef(ref: { provider: string; model: string; effort: "auto" }): { systemPrompt: unknown };
-  /** optional: the market overlay uses it after installing an MCP server (types.ts SextantAttach) */
-  reloadMcp?(): Promise<{ added: string[]; removed: string[]; failed: { name: string; error: string }[]; skipped: string[] }>;
-}
-
-/** Pull the two fixed rows out of a runtime that is ALREADY running.
- *
- *  Deliberately not `createRuntime`: the CLI builds one because it has none, and paid for that decision
- *  with a session directory it had to pass in to avoid creating a stray empty session. Inside the TUI the
- *  runtime is the one the next turn will use, so this is a read, not a construction — and it cannot leave
- *  anything behind.
- *
- *  `buildDef`, not `systemPrompt()`: the prompt a run actually sends is the base plus the model profile,
- *  the design section, and — for a model without native tool calling — the tool block. Asking for the base
- *  alone under-counts the largest fixed row by everything that makes it large. */
-export function fixedFrom(rt: LiveRuntime | null | undefined, ref: { provider: string; model: string }): ContextFixed {
-  if (!rt) return {};
-  try {
-    const schemas = rt.registry.list().map((t) => t.schema);
-    const def = rt.buildDef({ ...ref, effort: "auto" });
-    const sp = def.systemPrompt;
-    const system = typeof sp === "string" ? sp : typeof sp === "function" ? String((sp as (x: Record<string, unknown>) => unknown)({})) : undefined;
-    if (system === undefined) return {};
-    return { system, toolSchemas: JSON.stringify(schemas) };
-  } catch {
-    // a runtime that cannot describe itself is a panel that says so, not a crashed frame
-    return {};
-  }
-}
-
-/** the report module's shape, structurally: this file must compile and be testable without importing it */
-interface Slice { label: string; tokens: number; share: number; note?: string }
-interface Drift { estimated: number; reported: number; delta: number; fraction: number; beyondTolerance: boolean }
-interface Report {
-  model: { provider: string; model: string };
-  window?: number;
-  estimated: number;
-  corrected: number;
-  scale: { factor: number; measured: boolean; note: string };
-  slices: Slice[];
-  remaining?: number;
-  fraction?: number;
-  nearLimit?: boolean;
-  drift?: Drift;
-  totals: { inputTokens?: number; outputTokens?: number; cacheReadTokens?: number; cacheWriteTokens?: number };
-  costUsd?: number;
-  unpricedTurns: number;
-  images: number;
-}
-
-export interface ContextLoadInput {
-  messages: readonly unknown[];
-  current: { provider: string; model: string };
-  /** window + pricing, however the caller has them */
-  lookup: (r: { provider: string; model: string }) => unknown;
-  fixed: ContextFixed;
-  /** the tolerance drift is judged against; printed by the panel, never assumed by a reader */
-  tolerance?: number;
-  /** tests inject the report function; production lazily imports the real one */
-  report?: (input: unknown) => Report;
-}
-
-/** the fraction past which drift is called out rather than merely shown. Matches core/context-report.ts;
- *  it travels in the state so the panel can print it instead of asserting a number the reader cannot see. */
-const TOLERANCE = 0.05;
-
-/** Build the overlay's state. Never throws: every failure becomes something the panel can draw. */
-export async function loadContext(input: ContextLoadInput): Promise<Omit<ContextState, "scroll">> {
-  const tolerance = input.tolerance ?? TOLERANCE;
-  const model = `${input.current.provider}/${input.current.model}`;
-
-  let report: Report;
-  try {
-    const fn = input.report ?? (await import("../core/context-report.ts")).contextReport as unknown as (i: unknown) => Report;
-    report = fn({
-      messages: input.messages,
-      current: input.current,
-      lookup: input.lookup,
-      ...(input.fixed.system !== undefined ? { system: input.fixed.system } : {}),
-      ...(input.fixed.toolSchemas !== undefined ? { toolSchemas: input.fixed.toolSchemas } : {}),
-    });
-  } catch {
-    // the panel opens and says it has nothing, which is more useful than a frame that died counting
-    return {
-      model, raw: 0, estimated: 0, scale: { factor: 1, measured: false, note: "the context report could not be built" },
-      slices: [], images: 0, billed: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-      unpricedTurns: 0, live: false, tolerance,
-    };
-  }
-
-  const t = report.totals;
-  return {
-    model,
-    ...(report.window !== undefined ? { window: report.window } : {}),
-    // `estimated` on the panel is the CORRECTED number, because that is what the window and the bill are
-    // computed from; `raw` carries what o200k counted so the panel can show both
-    raw: report.estimated,
-    estimated: report.corrected,
-    scale: report.scale,
-    ...(report.remaining !== undefined ? { remaining: report.remaining } : {}),
-    ...(report.fraction !== undefined ? { fraction: report.fraction } : {}),
-    ...(report.nearLimit !== undefined ? { nearLimit: report.nearLimit } : {}),
-    slices: report.slices.map((s) => ({ label: s.label, tokens: s.tokens, share: s.share, ...(s.note ? { note: s.note } : {}) })),
-    images: report.images,
-    ...(report.drift ? { drift: { ...report.drift } } : {}),
-    billed: {
-      input: t.inputTokens ?? 0, output: t.outputTokens ?? 0,
-      cacheRead: t.cacheReadTokens ?? 0, cacheWrite: t.cacheWriteTokens ?? 0,
-    },
-    ...(report.costUsd !== undefined ? { costUsd: report.costUsd } : {}),
-    unpricedTurns: report.unpricedTurns,
-    // "live" is about the two rows a transcript cannot know. The system prompt is the one that matters:
-    // tool schemas can legitimately be an empty list, so their absence proves nothing.
-    live: input.fixed.system !== undefined,
-    tolerance,
-  };
-}

package/src/sextant/crew-cards.ts

@@ -1,350 +0,0 @@
-/** The CREW CARD (Berkay's pick, 2026-09-07): the plan panel's crew section as titled cards — one card per
- *  WORKFLOW rather than one row per task, with a chip per agent showing which CLI opened.
- *
- *      crew  3 working
- *      test-project-scan            ›
- *      3 agents · 01s
- *      ▪▪▫  codex claude codex
- *
- *  WHAT A WORKFLOW IS HERE. rovecode has no workflow concept and this file does not invent one: `s.crew` is a flat
- *  TaskInfo[], and the only grouping that exists in the data is "tasks started under the same run" — TaskManager
- *  stamps `batch` / `batchLabel` from bindRun (core/tasks.ts, nimbus-5f's slice), which every surface already calls
- *  once per run. No heuristic: no time buckets, no label-prefix matching. A task with no batch (started outside a
- *  bound run) gets its own single-task card titled by its own label, which is exactly what the flat list showed.
- *
- *  THE CHIP RULE — a chip must never claim a CLI ran when it did not:
- *   - queued is a HOLLOW chip and a dim name; only running/done chips are filled. "codex is queued" and "codex is
- *     working" cannot look the same, because the difference is whether a process exists.
- *   - only an EXTERNAL lane (TaskInfo.kind "external") shows a CLI name, from its `lane` adapter id. One of our own
- *     subagents shows its AGENT name; if such a task were ever named after an adapter (the ids are reserved in
- *     tasks.ts, so this is a defence rather than a case) it renders as `agent:<name>` and never as a bare CLI name.
- *
- *  WHAT GOES FIRST WHEN IT DOES NOT FIT (the panel is 32 cells; stated rather than discovered by clipping):
- *   1. the blank row between cards — spacing carries no information;
- *   2. the names row collapses: repeats become counts (`codex ×2`), then a `+N` tail. Whole names are dropped, never
- *      clipped mid-word — half a CLI's name is a different CLI's name;
- *   3. the title is truncated with an ellipsis, at a word boundary when there is one;
- *   4. the chips cap at what fits and the rest become `+N`;
- *   5. whole cards go, newest kept, with a `+N more` footer.
- *  NEVER dropped: the agent count and the clock. A card that cannot show those two is not drawn at all — a title
- *  alone would say a workflow exists without saying whether anything is running.
- *
- *  PURE: `now` is a parameter (the clock and the card's age), no timers, no process access, no state writes. The
- *  painter at the bottom is the only part that touches a screen. */
-
-import type { TaskInfo } from "../core/tasks.ts";
-import { isAdapterId } from "../lanes/types.ts";
-import { laneElapsed } from "./draw-agents.ts";
-import { fmtK } from "./model.ts";
-import { st } from "./theme.ts";
-import { ATTR, type ScreenLike, type Seg, type Theme } from "./types.ts";
-
-/** TaskInfo as the card reads it. `batch` / `batchLabel` / `progress` / `files` are all on TaskInfo now (core/tasks.ts);
- *  the alias stays as this module's own name for what it consumes. */
-export type CrewTask = TaskInfo;
-
-export type CardTone = "fg" | "fg2" | "muted" | "dim" | "ok" | "err" | "accent" | "warn";
-export interface CardSpan { text: string; tone: CardTone; bold?: boolean }
-/** one painted row; `right` is anchored to the card's right edge (the `›` affordance) */
-export interface CardLine { spans: CardSpan[]; right?: CardSpan }
-
-export interface CrewGroup {
-  /** the batch id, or `task:<id>` for a task that has none */
-  key: string;
-  label: string;
-  tasks: CrewTask[];
-  /** min(startedAt ?? createdAt) over the group */
-  startedAt: number;
-  /** max(finishedAt) when EVERY task has finished; undefined while any is live (the clock still runs) */
-  finishedAt?: number;
-  working: number;
-}
-
-const oneLine = (s: string): string => s.replace(/\s+/g, " ").trim();
-const cps = (s: string): string[] => [...s];
-const width = (s: string): number => cps(s).length;
-
-// ------------------------------------------------------------------ grouping
-
-/** Group by `batch`, first-seen order kept (the crew list is already in start order). A task without a batch is its
- *  own group. The label is the batch label when the surface set one, else the first task's label — never invented. */
-export function groupCrew(crew: readonly CrewTask[]): CrewGroup[] {
-  const out: CrewGroup[] = [];
-  const byKey = new Map<string, CrewGroup>();
-  for (const t of crew) {
-    const key = t.batch ?? `task:${t.id}`;
-    let g = byKey.get(key);
-    if (g === undefined) {
-      g = { key, label: oneLine(t.batchLabel ?? t.label) || "(no label)", tasks: [], startedAt: t.startedAt ?? t.createdAt, working: 0 };
-      byKey.set(key, g);
-      out.push(g);
-    }
-    g.tasks.push(t);
-    g.startedAt = Math.min(g.startedAt, t.startedAt ?? t.createdAt);
-  }
-  for (const g of out) {
-    g.working = g.tasks.filter((t) => t.status === "queued" || t.status === "running").length;
-    const ends = g.tasks.map((t) => t.finishedAt);
-    if (ends.every((e): e is number => e !== undefined)) g.finishedAt = Math.max(...ends);
-  }
-  return out;
-}
-
-/** the card's age: the WHOLE workflow's span, live until every task in it has finished */
-export function groupElapsed(g: CrewGroup, now: number): number {
-  if (g.finishedAt !== undefined) return Math.max(0, g.finishedAt - g.startedAt);
-  return Math.max(0, now - g.startedAt, ...g.tasks.map((t) => laneElapsed(t, now)));
-}
-
-/** `01s` · `04m` · `1h02m` — the compact age from the card Berkay picked; mm:ss (laneClock) is the per-task form */
-export function compactAge(ms: number): string {
-  const s = Math.max(0, Math.floor(ms / 1000));
-  if (s < 60) return `${String(s).padStart(2, "0")}s`;
-  const m = Math.floor(s / 60);
-  if (m < 60) return `${String(m).padStart(2, "0")}m`;
-  return `${Math.floor(m / 60)}h${String(m % 60).padStart(2, "0")}m`;
-}
-
-// ------------------------------------------------------------------ chips
-
-/** ▫ queued (nothing has run) · ▪ running · ◆ done · × failed · · cancelled */
-export function chipGlyph(status: TaskInfo["status"]): string {
-  switch (status) {
-    case "queued": return "▫";
-    case "running": return "▪";
-    case "done": return "◆";
-    case "failed": return "×";
-    default: return "·";
-  }
-}
-export function chipTone(status: TaskInfo["status"]): CardTone {
-  switch (status) {
-    case "queued": return "dim";
-    case "running": return "accent";
-    case "done": return "ok";
-    case "failed": return "err";
-    default: return "muted";
-  }
-}
-
-/** the name under a chip: a CLI id ONLY for an external lane; our own agents show their agent name. An external
- *  lane with no adapter id recorded says `cli?` rather than borrowing a name we do not have. */
-export function chipName(t: CrewTask): string {
-  if (t.kind === "external") return t.lane ?? "cli?";
-  return isAdapterId(t.agent) ? `agent:${t.agent}` : t.agent;
-}
-
-/** the chip row's glyphs, one per task in order; `+N` when more tasks than fit in `w` cells */
-export function chipRow(tasks: readonly CrewTask[], w: number): CardSpan[] {
-  if (w <= 0 || tasks.length === 0) return [];
-  const spans: CardSpan[] = [];
-  let used = 0;
-  for (let i = 0; i < tasks.length; i++) {
-    const rest = tasks.length - i;
-    const tail = rest > 1 ? `+${rest}` : "";
-    if (used + 1 + width(tail) > w && i > 0) { spans.push({ text: `+${rest}`, tone: "dim" }); return spans; }
-    if (used + 1 > w) return spans;
-    const t = tasks[i]!;
-    spans.push({ text: chipGlyph(t.status), tone: chipTone(t.status), bold: t.status === "running" });
-    used += 1;
-  }
-  return spans;
-}
-
-/** The names, collapsed to fit: `codex claude codex` → `codex ×2 claude` when repeats cost less, then a `+N` tail.
- *  Whole names only — a clipped name is a lie about which CLI ran. Empty when nothing fits. */
-export function nameRow(tasks: readonly CrewTask[], w: number): CardSpan[] {
-  if (w <= 0 || tasks.length === 0) return [];
-  const runs: { name: string; count: number; tone: CardTone }[] = [];
-  for (const t of tasks) {
-    const name = chipName(t), tone = chipTone(t.status);
-    const last = runs[runs.length - 1];
-    if (last !== undefined && last.name === name && last.tone === tone) last.count++;
-    else runs.push({ name, count: 1, tone });
-  }
-  const label = (r: { name: string; count: number }): string => (r.count > 1 ? `${r.name} ×${r.count}` : r.name);
-  const spans: CardSpan[] = [];
-  let used = 0;
-  for (let i = 0; i < runs.length; i++) {
-    const r = runs[i]!;
-    const text = label(r);
-    // the tail counts the AGENTS still unnamed, not the runs: "+2" when three agents are hidden would be a wrong number
-    const left = runs.slice(i).reduce((n, x) => n + x.count, 0);
-    const tail = ` +${left}`;
-    const sep = spans.length > 0 ? 1 : 0;
-    const more = i < runs.length - 1 ? width(` +${left - r.count}`) : 0;
-    if (used + sep + width(text) + more > w) {
-      if (spans.length > 0 && used + width(tail) <= w) spans.push({ text: tail, tone: "dim" });
-      return spans;
-    }
-    if (sep) { spans.push({ text: " ", tone: "dim" }); used += 1; }
-    spans.push({ text, tone: r.tone });
-    used += width(text);
-  }
-  return spans;
-}
-
-// ------------------------------------------------------------------ card rows
-
-/** ellipsis truncation at a word boundary when there is one in the last third of the space */
-export function fitTitle(title: string, w: number): string {
-  const c = cps(oneLine(title));
-  if (w <= 0) return "";
-  if (c.length <= w) return c.join("");
-  if (w === 1) return "…";
-  const head = c.slice(0, w - 1).join("");
-  const cut = head.lastIndexOf(" ");
-  return (cut >= Math.floor((w - 1) * 0.67) ? head.slice(0, cut) : head) + "…";
-}
-
-/** the card: title (+ `›`), `N agents · age`, then the chips and the names on one row. The meta row is the
- *  honesty core — it is never dropped, so a card is only produced when `w` can hold it. */
-export function cardLines(g: CrewGroup, w: number, now: number): CardLine[] {
-  const count = `${g.tasks.length} agent${g.tasks.length === 1 ? "" : "s"}`;
-  const meta = `${count} · ${compactAge(groupElapsed(g, now))}`;
-  if (w < width(meta)) return [];
-  const lines: CardLine[] = [];
-  const arrow = w >= width(meta) + 2 ? { text: "›", tone: "dim" as CardTone } : undefined;
-  lines.push({ spans: [{ text: fitTitle(g.label, w - (arrow ? 2 : 0)), tone: g.working > 0 ? "fg" : "fg2", bold: g.working > 0 }], ...(arrow ? { right: arrow } : {}) });
-  lines.push({ spans: [{ text: meta, tone: "dim" }] });
-  const chips = chipRow(g.tasks, Math.max(0, Math.min(g.tasks.length + 2, Math.floor(w / 2))));
-  const chipW = chips.reduce((n, s) => n + width(s.text), 0);
-  const names = nameRow(g.tasks, w - chipW - 2);
-  if (chips.length > 0) lines.push({ spans: names.length > 0 ? [...chips, { text: "  ", tone: "dim" }, ...names] : chips });
-  const progress = progressRow(g, w);
-  if (progress.length > 0) lines.push({ spans: progress });
-  return lines;
-}
-
-/** THE PROGRESS ROW — `wrote 4 files · 12 calls · 4.2k`, summed over a workflow's tasks, and it is the second thing
- *  dropped at a narrow width (after the blank separator): the chips and the CLI names are what the card exists for.
- *  Four rules, all of them about not overstating what a CLI did (lanes/progress.ts, nimbus-24's measurements):
- *   - a RUNNING lane with nothing confirmed shows no file segment at all. `filesWritten` is what the CLI has
- *     confirmed so far, and one of the four (agy) confirms nothing until it ends — it sits at zero for minutes and
- *     then jumps to the measured count. "wrote 0 files" for a lane that is writing would be a wrong statement, and
- *     treating that zero as a stall would be wrong about exactly one CLI.
- *   - a FINISHED task's list is the worktree diff, so there zero is a measurement and is worth printing.
- *   - a task whose patch never reached the tree did not WRITE anything the person has: its count is reported as
- *     `discarded` / `so far`, never as `wrote`. The signal is `progress.applied` (tri-state, written by the code
- *     that does the applying): absent = nothing offered to the tree yet, true = merged back, false = the writes
- *     stayed in a worktree that was deleted. NOT `patchLines`, and not the status either: a lane that finishes
- *     cleanly whose patch git refuses — a concurrent task touched the same lines — still reports ok AND gets a
- *     patchLines, so both of those would print `wrote 4 files` about a tree that has none of them (nimbus-24's
- *     lanes-job tests cover the case). One of OUR isolated children has no progress object, and orchestrator.ts
- *     has the same hole — so the signal there is `files`, which tasks.ts sets only for a patch that applied
- *     (5f gated it on progress.applied, else the runner's own "patch-apply-failed" marker). `patchLines` is a
- *     SIZE and never a verdict; the one thing it still settles is the empty patch, where 0 lines means there was
- *     nothing to apply and nothing that could fail.
- *   - `usage` ABSENT is not zero. A lane we could not read reports nothing; a lane that burned nothing reports
- *     zeros. Absent contributes no token segment; explicit zeros show `0`.
- *  `calls` counts every tool the CLI reported, reads included, so the word stays `calls`. Only lanes report them,
- *  so a mixed workflow's row says what its lanes did without implying the native agents did nothing. */
-/** Did this task's work reach the person's tree? `true` landed · `false` written and thrown away · `undefined`
- *  nothing offered yet. A lane answers from `progress.applied` (the code that applies, not the code beside it); one
- *  of our own children answers from `files`, which exists only for a patch that applied — except for the empty
- *  patch, where `patchLines === 0` means there was nothing to apply, and that counts as landed. */
-export function landedOf(t: CrewTask): boolean | undefined {
-  if (t.progress !== undefined && "applied" in t.progress) return t.progress.applied;
-  if (t.status === "running" || t.status === "queued") return undefined;
-  if (t.files !== undefined) return true;
-  return t.status === "done" && t.patchLines === 0;
-}
-
-export function progressRow(g: CrewGroup, w: number): CardSpan[] {
-  let calls = 0, applied = 0, pending = 0, discarded = 0, tokens = 0, sawUsage = false;
-  for (const t of g.tasks) {
-    calls += t.progress?.toolCalls ?? 0;
-    const n = t.files?.length ?? t.progress?.filesWrittenTotal ?? 0;
-    const landed = landedOf(t);
-    if (landed === true) applied += n;
-    else if (landed === undefined) pending += n;
-    else discarded += n;
-    const u = t.progress?.usage ?? t.usage;
-    if (u !== undefined) { sawUsage = true; tokens += u.input + u.output; }
-  }
-  const parts: { text: string; tone: CardTone }[] = [];
-  if (applied > 0) parts.push({ text: `wrote ${applied} file${applied === 1 ? "" : "s"}`, tone: "ok" });
-  if (pending > 0) parts.push({ text: `${pending} file${pending === 1 ? "" : "s"} so far`, tone: "dim" });
-  if (discarded > 0) parts.push({ text: `${discarded} discarded`, tone: "muted" });
-  // every task settled and nothing landed: that zero IS the measurement, and saying so beats an empty row
-  const merged = g.tasks.some((t) => landedOf(t) === true);
-  if (parts.length === 0 && g.working === 0 && merged) parts.push({ text: "wrote 0 files", tone: "dim" });
-  if (calls > 0) parts.push({ text: `${calls} call${calls === 1 ? "" : "s"}`, tone: "dim" });
-  if (sawUsage) parts.push({ text: fmtK(tokens), tone: "dim" });
-  if (parts.length === 0) return [];
-  const spans: CardSpan[] = [];
-  let used = 0;
-  for (const part of parts) {
-    const sep = spans.length > 0 ? 3 : 0; // " · "
-    if (used + sep + width(part.text) > w) break;
-    if (sep) { spans.push({ text: " · ", tone: "dim" }); used += 3; }
-    spans.push(part);
-    used += width(part.text);
-  }
-  return spans;
-}
-
-export interface CrewCardBlock {
-  lines: CardLine[];
-  /** cards that did not fit — the `+N more` footer is already the last line when this is > 0 */
-  hidden: number;
-}
-
-/** The card order: WORKING workflows first, newest of them at the top, then the settled ones newest first. The
- *  section's header says "2 working"; a panel that then shows two finished cards and hides the running one is
- *  answering a different question from the one it just asked. Within each half the newest card is first because
- *  the crew list is start-ordered, so its tail is the live work. */
-export function crewCardOrder(groups: readonly CrewGroup[]): CrewGroup[] {
-  const newestFirst = [...groups].reverse();
-  return [...newestFirst.filter((g) => g.working > 0), ...newestFirst.filter((g) => g.working === 0)];
-}
-
-/** Pack the cards into `rows` rows in crewCardOrder. Blank separators are dropped before any card is. */
-export function crewCardBlock(groups: readonly CrewGroup[], w: number, rows: number, now: number): CrewCardBlock {
-  if (rows <= 0 || w <= 0 || groups.length === 0) return { lines: [], hidden: groups.length };
-  const order = crewCardOrder(groups);
-  const cards = order.map((g) => cardLines(g, w, now)).filter((c) => c.length > 0);
-  const pack = (gap: boolean): CrewCardBlock & { shown: number } => {
-    const lines: CardLine[] = [];
-    let used = 0, shown = 0;
-    for (const card of cards) {
-      const need = card.length + (gap && shown > 0 ? 1 : 0);
-      const footer = cards.length - shown - 1 > 0 ? 1 : 0;
-      if (used + need + footer > rows) break;
-      if (gap && shown > 0) { lines.push({ spans: [] }); used++; }
-      lines.push(...card);
-      used += card.length;
-      shown++;
-    }
-    const hidden = groups.length - shown;
-    if (hidden > 0 && shown > 0 && used < rows) lines.push({ spans: [{ text: `+${hidden} more`, tone: "dim" }] });
-    return { lines: shown > 0 ? lines : [], hidden: shown > 0 ? hidden : groups.length, shown };
-  };
-  // separators are spacing; a card is information. Take the spaced packing only when it costs no card.
-  const spaced = pack(true), dense = pack(false);
-  return dense.shown > spaced.shown ? { lines: dense.lines, hidden: dense.hidden } : { lines: spaced.lines, hidden: spaced.hidden };
-}
-
-/** rows a full render of these groups would take (cards + separators), for a panel deciding its budget */
-export function crewCardHeight(groups: readonly CrewGroup[], w: number, now: number): number {
-  const cards = groups.map((g) => cardLines(g, w, now)).filter((c) => c.length > 0);
-  return cards.reduce((n, c) => n + c.length, 0) + Math.max(0, cards.length - 1);
-}
-
-// ------------------------------------------------------------------ painter
-
-const TONE: Record<CardTone, keyof Theme> = { fg: "fg", fg2: "fg2", muted: "muted", dim: "dim", ok: "ok", err: "err", accent: "accent", warn: "warn" };
-
-/** paint `lines` at (x, y) in `w` cells — the only impure part, and it writes nothing but those rows */
-export function paintCrewCards(scr: ScreenLike, x: number, y: number, w: number, lines: readonly CardLine[], theme: Theme): void {
-  lines.forEach((line, i) => {
-    const row = y + i;
-    if (line.right !== undefined) {
-      const r = line.right;
-      scr.put(x + w - width(r.text), row, r.text, st(theme[TONE[r.tone]] as number), width(r.text));
-    }
-    if (line.spans.length === 0) return;
-    const segs: Seg[] = line.spans.map((s) => [s.text, st(theme[TONE[s.tone]] as number, -1, s.bold ? ATTR.BOLD : 0)]);
-    scr.text(x, row, segs, w - (line.right ? width(line.right.text) + 1 : 0));
-  });
-}