@aragon-agent/cli 0.6.22 → 0.6.24

package/README.md

@@ -406,7 +406,7 @@ signal ends the run (`130` SIGINT, `143` SIGTERM, `129` SIGHUP).
 | `/save [file]` | Save the session to JSON |
 | `/resume [file]` | Load a saved session |
 | `/plan [on\|off\|status]` | Toggle plan mode - the keyboard-free equivalent of `Shift+Tab` |
-| `/team [on\|off\|max <n>\|overseer on\|off]` | Team subagents: status and the live roster, the on/off switch, the fan-out width, the dispatch supervisor |
+| `/team [on\|off\|max <n>\|overseer on\|off\|model [main\|fast\|<profile-name>]]` | Team subagents: status and the live roster, the on/off switch, the fan-out width, the dispatch supervisor |
 | `/todo [on\|off\|panel on\|off\|follow <mode>\|clear\|continue]` | Todo planning: status, the on/off switch, the rail, follow-through, and picking an unfinished plan back up |
 | `/retry [show\|on\|off\|max <n>]` | API retry: the effective ladder, the on/off switch, the retry count. Takes effect **in the running session**, children included |
 | `/fast [on\|off\|model <id>\|provider <id>\|same\|review <n\|off>]` | Fast model tier: status and this session's totals, the on/off switch, which model, the review cadence |
@@ -792,6 +792,11 @@ questions.
 - **`ask_user`** renders a keyboard-driven wizard: 1–5 questions, 2–4 options
   each, exactly one marked `RECOMMENDED`, plus an `Other…` free-text option.
   `↑`/`↓` choose, `Enter` confirms and advances, `←` goes back, `Esc` cancels.
+  **`Space` selects**: on a multi-choice question it toggles the highlighted
+  option; on a single-choice question it marks the highlighted option without
+  advancing, and `Enter` then submits the marked item — while a plain `Enter`
+  without `Space` still takes the highlighted option, recommended preselect
+  and all. `Space` does nothing on the `Other…` row (`Enter` opens its editor).
   Pressing `Enter` through the whole wizard accepts every recommendation, which
   is a deliberate "use your judgement" path rather than an accident. A last
   screen lets you review before submitting.
@@ -1118,6 +1123,17 @@ parallel, then read one combined report of what they each found. It decides for
 itself whether to delegate; a session that never calls `task` is the session
 that existed before this feature.
 
+**Which model a subagent runs on is your setting, not the model's.** The
+default (`team.model: 'main'`) keeps main-agent parity: every child runs the
+lead's own connection. Switch it to the **fast tier** or a **saved profile**
+from the settings screen's *Subagent model* row or with
+`/team model main|fast|<profile-name>` — useful when a fan-out of read-only
+scouts does not need the main model. The choice is re-resolved on every
+dispatch, and an unusable one (tier off, profile deleted, key missing) falls
+back to the main model with a disclosed line in the dispatch report instead of
+failing the dispatch. The model itself still cannot downgrade its own
+delegation: a stale `model` key in a `task` call is ignored, exactly as before.
+
 ```
 > read the auth middleware, map the route table, and check the migration
 
@@ -2025,6 +2041,8 @@ env / `.env` → CLI flags**.
 | `team.maxTurnsPerSubagent` | `0` | Runaway-loop cap, clamped to `[0, 100]`. `0` (the default) = no cap, matching the lead. A subagent stopped at a positive cap is reported as `stopped: turn cap reached`, with whatever it had produced. |
 | `team.overseer` | `true` | The dispatch supervisor: one per dispatch on the fast model, ticking on a geometric cadence over every live subagent and deciding wait / nudge / replace / abandon. With the fast tier missing a tick degrades to an announced unassisted wait — never a kill; with this `false`, children run under the legacy hard ceilings exactly. `/team overseer on\|off` flips it live, and it is read per dispatch. |
 | `team.overseerIntervalMs` | `0` | Supervisor cadence base, clamped to `[0, 3600000]`. `0` (the default) = the structural 300 s; a positive `team.subagentTimeoutMs` is honoured first when this is `0` (compat). Later checks grow ×1.6, clamped to `[60 s, 900 s]`; any nudge/replace/abandon resets that child's cadence to the base. Only meaningful while `team.overseer` is true. |
+| `team.model` | `"main"` | Which model every subagent runs on: `'main'` (the default - the lead's own connection), `'fast'` (the fast tier, re-resolved per dispatch), or `'profile'` (the entry named by `team.modelProfileId`). A **user-owned setting** — the model itself still cannot downgrade its own delegation, and an unusable choice falls back to the main model with a disclosed line in the dispatch report. Also settable via `/team model main\|fast\|<profile-name>` and the settings screen's **Subagent model** row. |
+| `team.modelProfileId` | `null` | The profile entry `team.model: 'profile'` binds to. Reference existence is not validated here — a dangling id degrades to the main model at resolution time, so deleting a profile never breaks a dispatch. |
 | `todo.enabled` | `true` | Master switch for todo planning. `false` means the `todo_write` tool is never registered — `/todo on` in such a session saves the setting for next launch and says so, rather than advertising a tool that is not there. |
 | `todo.panel` | `true` | Render the right-hand rail. Independent of `todo.enabled` on purpose: a screen-reader user wants the planning discipline without the column, and the system prompt varies one sentence accordingly. |
 | `todo.followThrough` | `"notify"` | What happens when a run ends with steps left. `notify` says so and stops; `auto` continues the plan after a 3-second grace window, bounded by two structural limits (one fruitless attempt, 25 continuations per plan) that are deliberately **not** settings; `off` is silent. An unrecognized value falls back to `notify`. |
@@ -2521,7 +2539,7 @@ VS Code 用户设置添加 `"terminal.integrated.copyOnSelection": false`，快
 
 在 `/settings` 中使用 **Main profile** 和 **Fast profile**，可以分别为主任务与快速模型选择已保存的方案。两种角色共享配置库，但各自选择互不影响，也可以选择同一个方案。**Manage profiles** 提供新建、编辑、复制和删除；名称留空时采用模型名。同名方案通过地址主机和唯一 ID 区分。超过八条时按 `/` 搜索名称、模型或服务商。
 
-上下方向键或 Tab 移动，Enter 选择或编辑，根设置页 Ctrl+S 保存。普通文本字段的 Enter 仍保存设置。选择和子表单的修改先进入草稿；Esc 返回或确认丢弃。被主模型或快速模型引用的方案必须先解除绑定才能删除。复制包含本轮已经编辑的凭据，并产生新的 ID。
+上下方向键或 Tab 移动，Enter 选择或编辑，根设置页 Ctrl+S 保存。普通文本字段的 Enter 仍保存设置。选择和子表单的修改先进入草稿；Esc 返回或确认丢弃。列表页（角色选择页与管理页）聚焦在方案条目上时按 `e` 直接编辑、按 `d` 直接进入删除确认。删除被引用的方案（主模型 / 快速模型 / 子代理）走“解除绑定并删除”单次流程：确认页会列出引用方，确认后绑定同步清空；未被引用的方案仍显示普通的 Delete。复制包含本轮已经编辑的凭据，并产生新的 ID。
 
 方案保存服务商、模型、地址和凭据，不改变思考级别、token 上限、快速评审频率或预算。选择快速模型方案不会自动启用快速层。若本次启动时快速层未注册，首次启用会提示“已保存，待重启”；缺少模型或凭据时也需要先补齐。
 

package/dist/agent/controller.d.ts

@@ -20,6 +20,7 @@ import { type ConfirmRequest } from '../tools/index.js';
 import type { ToolPermission } from '../exec/permission.js';
 import { type ToolOutputListener } from '../tools/tool-output-store.js';
 import type { FilePatch } from '../tools/patch.js';
+import type { SubagentModelResolution } from '../team/model.js';
 import type { TeamEvent, TeamSnapshot } from '../team/types.js';
 import { ProcSupervisor } from '../proc/supervisor.js';
 import type { ProcEventListener, ServiceSnapshot } from '../proc/types.js';
@@ -478,6 +479,16 @@ export declare class AgentController {
      * value can never reach the runtime.
      */
     setTeamConfig(patch: Partial<TeamConfig>): TeamConfig;
+    /**
+     * The dispatch-time subagent model resolution (subagent-model §3.4).
+     *
+     * One authority for the `task` report's disclosure line and its price
+     * table: both read through here, so the header can never name one model
+     * while pricing another. Re-resolved on every call for the reason
+     * `resolveFastTier` records - `/fast off`, a settings save and `/team model`
+     * all rewrite the inputs mid-session.
+     */
+    subagentModelResolution(): SubagentModelResolution;
     /** Refused mid-dispatch by `/team`; half a dispatch under two ceilings is not
      *  a result anything can afterwards explain (the reasoning `/reload` uses). */
     isTeamBusy(): boolean;

package/dist/agent/controller.js

@@ -16,6 +16,7 @@ import { createChildReviewer } from '../team/child-reviewer.js';
 import { TeamHumanQueue } from '../team/human-queue.js';
 import { createFastOverseerProvider } from '../team/overseer.js';
 import { createTaskTool } from '../team/task-tool.js';
+import { resolveSubagentModel } from '../team/model.js';
 import { buildTeamBlock } from '../team/prompt.js';
 import { buildFastBlock } from '../fast/prompt.js';
 import { ProcSupervisor } from '../proc/supervisor.js';
@@ -516,9 +517,22 @@ export class AgentController {
                         maxSubagents: () => this.config.team.maxSubagents,
                         hasApiKey: () => this.hasApiKey(),
                         activeProvider: () => this.config.provider,
-                        // Every child runs the lead's own model, so the lead's cost table
-                        // is the one honest price for the dispatch total.
-                        modelCost: () => this.getModelInfo().cost,
+                        // Priced at the model children ACTUALLY run (subagent-model §3.4):
+                        // under the default `main` mode this is exactly the lead's own
+                        // table (I-2 byte parity); a fast or profile model the static
+                        // tables do not price yields undefined, so the report says
+                        // `pricing unknown` rather than fabricating $0.00 (C-11 / RV-4).
+                        modelCost: () => {
+                            const resolution = this.subagentModelResolution();
+                            if (resolution.effective === 'main')
+                                return this.getModelInfo().cost;
+                            return this.isPricedModel(resolution.ref)
+                                ? this.getModelInfoFor(resolution.ref).cost
+                                : undefined;
+                        },
+                        // The same resolution, re-read per dispatch, feeds the report's
+                        // disclosure line - one read per consumer, one authority both.
+                        subagentModel: () => this.subagentModelResolution(),
                         // The FAST table for the supervisor's own spend line
                         // (subagent-overseer-v2 D-7 / R-P1-2). Reuses the tier
                         // resolution and the unknown-pricing predicate the fast
@@ -575,6 +589,10 @@ export class AgentController {
             isWindowKnown: () => this.getModelInfo().contextWindowSource !== 'fallback',
             // LIVE, so a settings-screen edit moves the denominator without a relaunch.
             getWindowOverride: () => this.config.contextWindow,
+            // Beside the value, LIVE too: `/context` must name the surface that
+            // supplied the denominator (profile-context-window AC-9, one of four
+            // meter sites that must stay in lockstep).
+            getWindowOverrideOrigin: () => this.config.contextWindowOrigin,
         });
         // --- Context compaction, part 1: BEFORE the Agent ----------------------
         //
@@ -634,7 +652,15 @@ export class AgentController {
                     // `dispatchTimeoutMs: 0` (the default) reaches the executor as an
                     // override of `0` = NO TIMER AT ALL: the dispatch lasts as long as its
                     // slowest child, and only an Esc ends it early (main-agent parity).
-                    ...(this.teamRegistered ? { task: config.team.dispatchTimeoutMs } : {}),
+                    // D-4, THE SECOND ARMING SITE: the runtime arms its own dispatch
+                    // timer under the same gate, because the executor timeout being
+                    // cooperative means this override must never be the ONLY ceiling.
+                    // With the supervisor gate open the ceiling disarms HERE too, so
+                    // team.overseer can never leave two kill paths alive behind one
+                    // config. Read once at construction, exactly as before.
+                    ...(this.teamRegistered
+                        ? { task: config.team.overseer ? 0 : config.team.dispatchTimeoutMs }
+                        : {}),
                 },
             },
         });
@@ -1299,6 +1325,18 @@ export class AgentController {
         this.rebuildSystemPrompt();
         return team;
     }
+    /**
+     * The dispatch-time subagent model resolution (subagent-model §3.4).
+     *
+     * One authority for the `task` report's disclosure line and its price
+     * table: both read through here, so the header can never name one model
+     * while pricing another. Re-resolved on every call for the reason
+     * `resolveFastTier` records - `/fast off`, a settings save and `/team model`
+     * all rewrite the inputs mid-session.
+     */
+    subagentModelResolution() {
+        return resolveSubagentModel(this.config, (id, role) => this.hasApiKey(id, role));
+    }
     /** Refused mid-dispatch by `/team`; half a dispatch under two ceilings is not
      *  a result anything can afterwards explain (the reasoning `/reload` uses). */
     isTeamBusy() {

package/dist/agent/model-profile-settings.js

@@ -68,7 +68,11 @@ function liveCandidate(live, candidate, draft) {
         normalizedPatch[key] = candidate[key];
     }
     const next = { ...live, ...normalizedPatch };
-    for (const section of ['apiKeys', 'fast', 'compaction', 'log']) {
+    // 'team' joined the loop with the subagent-model picker (D-5): its rows
+    // write `draft.patch.team`, and `mergePersistedConfig` clamps the candidate
+    // through `clampTeamConfig` on the same per-key merge `/team` uses, so the
+    // live view and the file can never disagree about which model children run.
+    for (const section of ['apiKeys', 'fast', 'compaction', 'log', 'team']) {
         const normalized = { ...live[section] };
         for (const key of Object.keys(patch[section] ?? {})) {
             normalized[key] = candidate[section][key];
@@ -84,10 +88,33 @@ function liveCandidate(live, candidate, draft) {
         next.maxTokens = candidate.maxTokens ?? undefined;
     if ('baseUrl' in patch)
         next.baseUrl = candidate.baseUrl ?? undefined;
-    const projected = projectModelProfiles(candidate).file;
+    const projection = projectModelProfiles(candidate);
+    // `projected` keeps its name and every existing use (`projected.provider`
+    // etc.); the FULL projection is retained because `mainOwnsContextWindow`
+    // lives on it, not on `.file` (profile-context-window §6.4).
+    const projected = projection.file;
     if (draft.profiles !== undefined || candidate.modelProfiles !== undefined) {
         next.modelProfiles = structuredClone(projected.modelProfiles);
     }
+    // THE WINDOW VALUE AND ITS ORIGIN RECOMPUTE IN THE SAME BATCH (P0-1): the
+    // meter's `getWindowOverride()` reads `this.config.contextWindow`, so a save
+    // that only recomputed the origin would leave a freshly edited profile
+    // window invisible until the next launch. env is a STARTUP-layer fact and no
+    // save may displace it (AC-4 / AC-14c); an explicit root-row edit takes the
+    // candidate value; a bound Main profile owns the denominator (value or null,
+    // tradeoff B); unbinding falls back to the legacy top-level key.
+    next.contextWindow = live.contextWindowOrigin === 'env'
+        ? live.contextWindow
+        : 'contextWindow' in patch ? candidate.contextWindow
+            : projection.mainOwnsContextWindow ? projected.contextWindow ?? null
+                : candidate.contextWindow;
+    // ORIGIN OBEYS THE IFF CONTRACT (§6.1): value null means AUTO, so origin
+    // MUST reset to undefined - this unified formula is what removes the stale
+    // 'profile' a per-branch expression would leave behind on unbind-to-null.
+    next.contextWindowOrigin = next.contextWindow === null ? undefined
+        : live.contextWindowOrigin === 'env' ? 'env'
+            : 'contextWindow' in patch ? 'config'
+                : projection.mainOwnsContextWindow ? 'profile' : 'config';
     next.liveToolOutput = live.liveToolOutput;
     for (const role of ['main', 'fast']) {
         if (!draft.activateRoles.includes(role) && !connectionChanged(draft, role)

package/dist/agent/reducer.js

@@ -183,18 +183,20 @@ export function reduceEvent(event, cost, patches) {
 }
 /**
  * Tools that render their own transcript entry, so the generic tool card would
- * be a duplicate. EXACTLY ONE MEMBER; read todo-plan-execution §3.8 before
- * adding a second.
+ * be a duplicate. `task` is the second member (F-8 A'3): every dispatch opens
+ * a TeamCard immediately after, so the pair read as TWO dispatches - the first
+ * failed, the second succeeded - when there was only ever one. `todo_write`'s
+ * reasons (todo-plan-execution §3.8 / D-10) carry over unchanged.
  *
  * SUPPRESSING ONLY `tool_call_start` IS SUFFICIENT, AND THAT IS THE WHOLE REASON
  * THIS IS SAFE (D-10). Every later handler — `toolCallDelta` / `toolCallEnd` /
  * `toolExecStart` / `toolExecEnd` — resolves its target through
  * `findToolEntryId`, which returns `undefined` when no entry was created, and
- * every one of them already early-returns on that. There is no fifth handler. A
- * second suppression rule would be four more places to get the id matching
- * wrong, for nothing.
+ * every one of them already early-returns on that. There is no fifth handler.
+ * Blast radius verified: headless and `exec` writers key on EVENTS, not tool
+ * names, and `transcript-text` renders `kind:'team'` on its own.
  */
-const SELF_RENDERING_TOOLS = new Set(['todo_write']);
+const SELF_RENDERING_TOOLS = new Set(['todo_write', 'task']);
 function reduceStreamEvent(se) {
     switch (se.type) {
         case 'thinking_start':

package/dist/commands/builtins.js

@@ -16,6 +16,8 @@ import { buildContinuationMessage } from '../todo/follow-through.js';
 import { MODE_LABEL, nextMode } from '../agent/agent-mode.js';
 import { loadSession, resolveSessionPath, saveSession } from '../session/persist.js';
 import { getLogger } from '../logging/logger.js';
+import { resolveFastTier } from '../fast/resolve.js';
+import { subagentModelSummary } from '../ui/overlays/model-profile-state.js';
 import { registerQueueCommand } from './queue-command.js';
 import { perfCommand } from './perf.js';
 import { runCompactCommand } from '../compaction/command.js';
@@ -90,6 +92,66 @@ function formatUpdateStatus(port, override) {
     lines.push(`  next      ${next ? new Date(next).toLocaleTimeString() : 'not scheduled'}`);
     return lines.join('\n');
 }
+/**
+ * `/team model [main|fast|<profile-name>]` (subagent-model §3.6 / 5.2).
+ *
+ * The user-owned switch's command-line surface. `main` and `fast` are verbs
+ * compared case-insensitively; a NAME is matched EXACTLY, case and all,
+ * against the library - the same identity the picker shows - and a miss lists
+ * the names that would work rather than failing silently. The summary reuses
+ * the exact wording the settings root row renders, so the two surfaces can
+ * never describe the same setting differently.
+ */
+function runTeamModelCommand(ctx, subject) {
+    const controller = ctx.controller;
+    const config = controller.getConfig();
+    const tier = resolveFastTier(config, (id, role) => controller.hasApiKey(id, role));
+    const summary = subagentModelSummary({
+        mode: config.team.model,
+        modelProfileId: config.team.modelProfileId,
+        mainConnection: `${config.provider}:${config.model}`,
+        fastConnection: tier.ok ? `${tier.ref.providerId}:${tier.ref.modelId}` : null,
+        profiles: config.modelProfiles,
+        hasKey: (id) => controller.hasApiKey(id),
+    });
+    if (!subject) {
+        // The status readout: mode + the connection actually in effect + the
+        // disclosed reason when the request had to fall back (D-4).
+        const resolution = controller.subagentModelResolution();
+        const note = resolution.fallbackReason
+            ? ` Unavailable (${resolution.fallbackReason}); children run on the main model.`
+            : resolution.effective === 'main'
+                ? ''
+                : ` Children run on ${resolution.ref.providerId}:${resolution.ref.modelId}.`;
+        ctx.notify('info', `Subagent model: ${summary}.${note}`);
+        return;
+    }
+    if (subject.toLowerCase() === 'main' || subject.toLowerCase() === 'fast') {
+        const mode = subject.toLowerCase();
+        const applied = controller.setTeamConfig({ model: mode });
+        ctx.persistConfig({ team: { model: applied.model } });
+        if (mode === 'fast' && !tier.ok) {
+            ctx.notify('warn', 'The fast tier is not available now; subagents will fall back to the main model.');
+        }
+        ctx.toast('success', `Subagent model: ${mode === 'main' ? 'same as main' : 'fast tier'}.`);
+        return;
+    }
+    const entries = config.modelProfiles?.entries ?? [];
+    const matches = entries.filter((entry) => entry.name === subject);
+    if (matches.length === 0) {
+        const names = entries.slice(0, 8).map((entry) => entry.name);
+        ctx.notify('warn', `No profile named "${subject}".`
+            + (names.length ? ` Available: ${names.join(', ')}` : ' Create one in /settings first.'));
+        return;
+    }
+    const entry = matches[0];
+    const disambiguation = matches.length > 1 ? ` (#${entry.id.slice(0, 8)})` : '';
+    const applied = controller.setTeamConfig({ model: 'profile', modelProfileId: entry.id });
+    const teamPatch = { model: applied.model,
+        modelProfileId: applied.modelProfileId };
+    ctx.persistConfig({ team: teamPatch });
+    ctx.toast('success', `Subagent model: ${entry.name}${disambiguation} (${entry.provider}:${entry.model}).`);
+}
 function formatFastStatus(controller) {
     const status = controller.getFastStatus();
     const cfg = controller.getFastConfig();
@@ -254,7 +316,7 @@ const COMMANDS = [
     },
     {
         name: 'team',
-        description: 'Team subagents: status | on | off | max <n> | overseer on|off',
+        description: 'Team subagents: status | on | off | max <n> | overseer on|off | model [main|fast|<profile-name>]',
         /**
          * `/team on` CANNOT turn team mode on in a session that started without it,
          * and this command must not pretend otherwise (§4.5 / D-17 / P0-2).
@@ -275,6 +337,12 @@ const COMMANDS = [
             const arg = ctx.args.trim().toLowerCase();
             const [verb, value] = arg.split(/\s+/, 2);
             const controller = ctx.controller;
+            // RAW TOKENS, KEPT BESIDE THE LOWERCASED PAIR (V-2): `model` matches a
+            // profile NAME case-sensitively - like the picker and the library -
+            // and the whole-args toLowerCase() above would have made
+            // `/team model MyProfile` silently unmatchable. Only the VERB token is
+            // compared through the lowercased pair.
+            const rawTokens = ctx.args.trim().split(/\s+/).filter((token) => token.length > 0);
             // Refused mid-dispatch for the reason `/reload` already records: half a
             // dispatch under one ceiling and half under another produces a result
             // nothing can afterwards explain.
@@ -284,6 +352,7 @@ const COMMANDS = [
             }
             if (!verb || verb === 'status') {
                 const cfg = controller.getTeamConfig();
+                const config = controller.getConfig();
                 const state = !controller.isTeamRegistered()
                     ? 'off for this session (started with --no-team)'
                     : controller.isTeamEnabled()
@@ -293,9 +362,20 @@ const COMMANDS = [
                 const roster = snapshot
                     ? `\nRunning: ${snapshot.runs.map((r) => `${r.label} (${r.phase})`).join(', ')}`
                     : '';
+                const tier = resolveFastTier(controller.getConfig(), (id, role) => controller.hasApiKey(id, role));
+                const subagent = subagentModelSummary({
+                    mode: cfg.model,
+                    modelProfileId: cfg.modelProfileId,
+                    mainConnection: `${config.provider}:${config.model}`,
+                    fastConnection: tier.ok ? `${tier.ref.providerId}:${tier.ref.modelId}` : null,
+                    profiles: config.modelProfiles,
+                    hasKey: (id) => controller.hasApiKey(id),
+                });
                 ctx.notify('info', `Team mode: ${state}. Max subagents: ${cfg.maxSubagents}, ` +
                     `concurrent: ${cfg.maxConcurrent}, supervisor: ${cfg.overseer ? 'on' : 'off'}` +
-                    `${cfg.overseer ? ' (never kills; checks the fast tier each tick)' : ' (legacy hard timeouts)'}.${roster}`);
+                    `${cfg.overseer ? ' (never kills; checks the fast tier each tick)' : ' (legacy hard timeouts)'}.` +
+                    `\nSubagent model: ${subagent}` +
+                    `${roster}`);
                 return;
             }
             if (verb === 'on' || verb === 'off') {
@@ -332,6 +412,10 @@ const COMMANDS = [
                 ctx.toast('success', `Max subagents set to ${applied.maxSubagents}.`);
                 return;
             }
+            if (verb === 'model') {
+                runTeamModelCommand(ctx, rawTokens[1]);
+                return;
+            }
             if (verb === 'overseer') {
                 if (value !== 'on' && value !== 'off') {
                     ctx.notify('warn', 'Usage: /team overseer on|off');
@@ -347,7 +431,7 @@ const COMMANDS = [
                 ctx.toast('success', `Dispatch supervisor ${applied.overseer ? 'on' : 'off'}.`);
                 return;
             }
-            ctx.notify('warn', `Unknown argument "${verb}" - use /team [status|on|off|max <n>|overseer on|off].`);
+            ctx.notify('warn', `Unknown argument "${verb}" - use /team [status|on|off|max <n>|overseer on|off|model main|fast|<profile-name>].`);
         },
     },
     {

package/dist/compaction/child.d.ts

@@ -56,6 +56,21 @@ export interface ChildCompactionDeps {
 export interface ChildContextManagerRequest {
     role?: ModelRole;
     label: string;
+    /**
+     * REQ-LEVEL key resolvers, preferred over the lead's (V-4).
+     *
+     * `wiring.ts::childFactory` injects the LEAD layer's `getApiKey`/`hasKey`,
+     * which cannot see a fast-only key (`role='fast'` only) or a profile's
+     * dedicated key (it exists only inside the entry). `Compactor.resolveSummarizer`
+     * gates its main choice on `hasKey(provider, role)` and reads the call key
+     * through `getApiKey`, so without these a `team.model: fast/profile` child
+     * whose only key lives in one of those places would have its compaction
+     * SILENTLY DISABLED - the summarizer resolves to null and nobody says why.
+     * `subagent.ts` passes the same profile-aware closures the child's own
+     * `Agent` gets; ABSENT keeps today's behavior byte for byte (matrix 24).
+     */
+    getApiKey?: (providerId: string, role?: ModelRole) => string | undefined;
+    hasKey?: (providerId: string, role?: ModelRole) => boolean;
     /**
      * The model the child actually runs on.
      *

package/dist/compaction/child.js

@@ -65,6 +65,10 @@ export function createChildContextManager(req, deps) {
             baseUrl: req.model.baseUrl,
             // The lead's manual limit describes its connection, not every worker.
             contextWindow: sameConnection ? base.contextWindow : null,
+            // THE ORIGIN FOLLOWS THE VALUE through the same gate, honoring the iff
+            // contract (value null => origin undefined): a child on a different
+            // connection inherits neither the number nor its provenance.
+            contextWindowOrigin: sameConnection ? base.contextWindowOrigin : undefined,
             // NO POLICY OVERLAY (main-agent parity): `keepRecentTurns`, `onFailure`,
             // `useFastTier` and `archive` are read LIVE from the base config, so the
             // child compacts under exactly the policy the lead compacts under. A
@@ -81,15 +85,22 @@ export function createChildContextManager(req, deps) {
             return source === undefined ? deps.isPricedModel(req.model) : source !== 'fallback';
         },
         getWindowOverride: () => childConfig().contextWindow,
+        getWindowOverrideOrigin: () => childConfig().contextWindowOrigin,
     });
     let overflowNotified = false;
+    // REQ-LEVEL RESOLVERS WIN (V-4), and the fallback is the lead layer - so a
+    // request built without them behaves exactly as before, while one built by
+    // `subagent.ts` under `team.model: fast/profile` lets the summarizer see
+    // the same keys the child itself authenticates with.
+    const hasKey = req.hasKey ?? deps.hasKey;
+    const getApiKey = req.getApiKey ?? deps.getApiKey;
     const compactor = new Compactor({
         getConfig: childConfig,
         meter,
         mainRole: req.role ?? 'main',
-        resolveFastCandidate: () => resolveFastTier(deps.getConfig(), deps.hasKey),
-        hasKey: deps.hasKey,
-        getApiKey: deps.getApiKey,
+        resolveFastCandidate: () => resolveFastTier(deps.getConfig(), hasKey),
+        hasKey,
+        getApiKey,
         getModelInfoFor: deps.getModelInfoFor,
         isPricedModel: deps.isPricedModel,
         getMessages: req.getMessages,

package/dist/compaction/compactor.js

@@ -403,6 +403,7 @@ export class Compactor {
                 return source === undefined ? this.deps.isPricedModel(this.mainRef()) : source !== 'fallback';
             },
             getWindowOverride: () => this.deps.getConfig().contextWindow,
+            getWindowOverrideOrigin: () => this.deps.getConfig().contextWindowOrigin,
         });
         return this.ownMeter;
     }

package/dist/compaction/context-command.js

@@ -87,7 +87,16 @@ function occupancyLine(usage) {
  */
 function windowLine(usage) {
     if (usage.windowOverridden) {
-        return `${usage.window}   from contextWindow configuration (overrides API, model table, and model-windows.json)`;
+        // THREE NAMED SURFACES (profile-context-window §6.2). `undefined` falls
+        // into the config branch ON PURPOSE: that string is what every pre-feature
+        // test asserts, and a meter built without the origin dep must report
+        // byte-identically (invariant 6).
+        const named = usage.windowOverrideOrigin === 'profile'
+            ? 'from your model profile'
+            : usage.windowOverrideOrigin === 'env'
+                ? 'from ARAGON_CONTEXT_WINDOW'
+                : 'from contextWindow configuration';
+        return `${usage.window}   ${named} (overrides API, model table, and model-windows.json)`;
     }
     if (!usage.windowKnown) {
         return (`UNKNOWN - no verified model limit; internal fallback budget ${usage.window}; ` +

package/dist/compaction/meter.d.ts

@@ -28,6 +28,7 @@
  * `current()` carries a structural belt for a missed invalidation site.
  */
 import type { AgentEvent, Message, ModelInfo, TokenUsage } from '@aragon-agent/core';
+import type { ContextWindowOrigin } from '../config/schema.js';
 import type { ContextUsageSnapshot, Pressure } from './types.js';
 /**
  * How long a scheduled publish waits.
@@ -62,6 +63,14 @@ export interface ContextMeterDeps {
      * measurement, so a live settings edit moves the denominator without a relaunch.
      */
     getWindowOverride(): number | null;
+    /**
+     * WHERE the override above came from, read live beside it
+     * (profile-context-window §6.1). Optional so every existing construction
+     * site stays untouched: a meter built without it reports an overridden
+     * window with origin `undefined`, and `/context` falls back to its
+     * pre-feature "from contextWindow configuration" wording byte-for-byte.
+     */
+    getWindowOverrideOrigin?(): ContextWindowOrigin | undefined;
     getRequestVersion?(): number;
 }
 export type ContextUsageListener = (usage: ContextUsageSnapshot) => void;

package/dist/compaction/meter.js

@@ -63,6 +63,9 @@ export function toContextUsage(p) {
         deltaTokens: p.deltaTokens,
         windowKnown: p.windowKnown,
         windowOverridden: p.windowOverridden,
+        ...(p.windowOverrideOrigin !== undefined
+            ? { windowOverrideOrigin: p.windowOverrideOrigin }
+            : {}),
     };
 }
 /** The reading a meter with nothing measured yet reports. */
@@ -426,6 +429,10 @@ export class ContextMeter {
             window.contextWindow !== last?.contextWindow ||
             window.windowKnown !== last?.windowKnown ||
             window.windowOverridden !== last?.windowOverridden ||
+            // ORIGIN IS PART OF THE DIRTY CHECK: a save that keeps the same number
+            // but moves it from the legacy key to the bound profile would otherwise
+            // leave `/context` naming the old source until the next measurement.
+            window.windowOverrideOrigin !== last?.windowOverrideOrigin ||
             this.deps.getModelInfo().contextWindowSource !== last?.windowSource)
             this.dirty = true;
     }
@@ -451,6 +458,9 @@ export class ContextMeter {
             windowSource: this.deps.getModelInfo().contextWindowSource,
             windowKnown: window.windowKnown,
             windowOverridden: window.windowOverridden,
+            ...(window.windowOverrideOrigin !== undefined
+                ? { windowOverrideOrigin: window.windowOverrideOrigin }
+                : {}),
             ...(this.estimateOffset !== undefined ? { estimateOffset: this.estimateOffset } : {}),
             ...(this.measuredPrefixLength !== undefined
                 ? { measuredPrefixLength: this.measuredPrefixLength }
@@ -469,7 +479,12 @@ export class ContextMeter {
     resolveWindow() {
         const override = this.deps.getWindowOverride();
         if (override !== null && Number.isFinite(override) && override > 0) {
-            return { contextWindow: override, windowKnown: true, windowOverridden: true };
+            // ORIGIN ONLY RIDES THE OVERRIDE BRANCH: the AUTO branch names its own
+            // source through `getModelInfo().contextWindowSource`, and the two must
+            // never both claim the denominator (profile-context-window §6.1).
+            const origin = this.deps.getWindowOverrideOrigin?.();
+            return { contextWindow: override, windowKnown: true, windowOverridden: true,
+                ...(origin !== undefined ? { windowOverrideOrigin: origin } : {}) };
         }
         return {
             contextWindow: this.deps.getModelInfo().contextWindow,

package/dist/compaction/pressure.d.ts

@@ -93,6 +93,13 @@ export interface PressureInput {
      * construction site is unchanged and the arithmetic below is untouched.
      */
     windowOverridden?: boolean;
+    /**
+     * Which override surface supplied `contextWindow` (profile-context-window).
+     * PURE PASS-THROUGH like `windowOverridden` - the arithmetic below never
+     * reads it; it only travels with the number so `/context` can name the
+     * source honestly.
+     */
+    windowOverrideOrigin?: Pressure['windowOverrideOrigin'];
     /** Carried from the last `turn_end`; applied to every ESTIMATED figure. */
     estimateOffset?: number;
     /**

package/dist/compaction/pressure.js

@@ -174,6 +174,11 @@ export function computePressure(input) {
         // context-usage-gauge-accuracy; the flag only travels with the number so
         // `/context` can name the denominator's source (I-6).
         windowOverridden: input.windowOverridden ?? false,
+        // CONDITIONAL SPREAD: with no origin there is no KEY, keeping a
+        // feature-off Pressure byte-identical to the pre-feature one (I-6).
+        ...(input.windowOverrideOrigin !== undefined
+            ? { windowOverrideOrigin: input.windowOverrideOrigin }
+            : {}),
         ...(input.estimateOffset !== undefined ? { estimateOffset: input.estimateOffset } : {}),
         deltaTokens,
     };

package/dist/compaction/types.d.ts

@@ -20,6 +20,7 @@
  * possible change.
  */
 import type { ModelRef, TokenUsage } from '@aragon-agent/core';
+import type { ContextWindowOrigin } from '../config/schema.js';
 /** Billing belongs to the actual call, including fallback and late responses. */
 export interface CompactionBilling {
     modelRef?: ModelRef;
@@ -92,6 +93,14 @@ export interface Pressure {
      * one being wrong is exactly the case that most needs to be visible.
      */
     windowOverridden: boolean;
+    /**
+     * WHICH override surface supplied the denominator (profile-context-window
+     * §5.2): env / legacy top-level key / the profile bound as Main. Optional
+     * and PURE PASS-THROUGH - present only when `windowOverridden` is true, and
+     * `undefined` there means "config", the pre-feature answer. The arithmetic
+     * never reads it; only `/context`'s window line does.
+     */
+    windowOverrideOrigin?: ContextWindowOrigin;
     /**
      * `occupiedTokens(lastUsage) - estimateOf(the history that produced it)`,
      * clamped to `>= 0`, or `undefined` before the first `turn_end`.
@@ -150,6 +159,8 @@ export interface ContextUsageSnapshot {
     windowKnown: boolean;
     /** Whether the denominator came from the user's `contextWindow` (I-6). */
     windowOverridden: boolean;
+    /** Which override surface named the denominator; see `Pressure` (pass-through). */
+    windowOverrideOrigin?: ContextWindowOrigin;
 }
 /** One compaction, in whatever state it reached. Rendered by `CompactionCard`. */
 export interface CompactionRecord {

package/dist/compaction/wiring.js

@@ -105,6 +105,7 @@ export class CompactionWiring {
                         return source === undefined ? deps.isPricedModel(this.mainRef()) : source !== 'fallback';
                     },
                     getWindowOverride: () => deps.getConfig().contextWindow,
+                    getWindowOverrideOrigin: () => deps.getConfig().contextWindowOrigin,
                 });
         this.compactor = new Compactor({
             getConfig: deps.getConfig,

package/dist/config/cli-commands.d.ts

@@ -20,7 +20,7 @@ export declare const LOG_CONFIG_SET_KEYS: readonly ["log.level", "log.toFile", "
  * switch falls through every case, writes nothing, and still prints
  * `Set team.enabled = false`.
  */
-export declare const TEAM_CONFIG_SET_KEYS: readonly ["team.enabled", "team.maxSubagents", "team.maxConcurrent", "team.subagentTimeoutMs", "team.dispatchTimeoutMs", "team.maxTurnsPerSubagent", "team.overseer", "team.overseerIntervalMs"];
+export declare const TEAM_CONFIG_SET_KEYS: readonly ["team.enabled", "team.maxSubagents", "team.maxConcurrent", "team.subagentTimeoutMs", "team.dispatchTimeoutMs", "team.maxTurnsPerSubagent", "team.overseer", "team.overseerIntervalMs", "team.model", "team.modelProfileId"];
 /**
  * Config keys under `retry.` that `aragon config set` accepts
  * (llm-api-retry-backoff §8.2).