sneakoscope 0.7.56 → 0.7.57

package/README.md

@@ -42,7 +42,7 @@ sks selftest --mock
 
 | Area | What it does |
 | --- | --- |
-| CLI runtime | Bare `sks` opens or reuses the default tmux Codex CLI workspace. `sks tmux open` remains the explicit form for session/workspace flags, and `sks --mad` launches the explicit full-access high-reasoning profile. |
+| CLI runtime | Bare `sks` opens or reuses the default tmux Codex CLI workspace. `sks tmux open` remains the explicit form for session/workspace flags, and `sks --mad` launches a multi-pane MAD tmux cockpit with the explicit full-access high-reasoning profile. |
 | Codex App commands | Installs generated skills so `$Team`, `$From-Chat-IMG`, `$DFix`, `$QA-LOOP`, `$PPT`, `$Image-UX-Review`, `$UX-Review`, `$Goal`, `$DB`, `$Wiki`, `$Help`, and related routes are visible in prompt workflows. `sks codex-app remote-control` wraps Codex CLI 0.130.0+ headless remote control without falling back to older app-server internals. |
 | OpenClaw agents | Generates an OpenClaw skill package so OpenClaw agents can attach `sneakoscope-codex`, enable the `shell` tool, and discover/use SKS commands from the target repo root. |
 | Pipeline plans | Writes `pipeline-plan.json` for stateful routes so the runtime lane, kept stages, skipped stages, verification commands, and no-unrequested-fallback invariant are visible with `sks pipeline plan`. |
@@ -53,7 +53,7 @@ sks selftest --mock
 | PPT pipeline | Uses `$PPT` for simple, restrained, information-first HTML/PDF presentation artifacts, first asking delivery context, audience profile, STP strategy, decision context, and 3+ pain-point to solution/aha mappings before source research, design-system work, HTML/PDF export, and render QA. Independent strategy/render/file-write phases run in parallel where inputs allow and are recorded in `ppt-parallel-report.json`; editable source HTML is preserved under `source-html/`, PPT-only temporary build files are cleaned after completion, installed skills/MCPs outside the `$PPT` allowlist are ignored, generated image assets must use real `$imagegen`/`gpt-image-2` output when sealed in the contract, and `ppt-style-tokens.json` records the design SSOT plus fused source inputs. |
 | Image UX Review | Uses `$Image-UX-Review` / `$UX-Review` for UI/UX audits where source screenshots are first turned into generated annotated review images through Codex App `$imagegen`/`gpt-image-2`; those generated images are then read back into `image-ux-issue-ledger.json`, optional requested fixes are rechecked, and missing generated review images or text-only screenshot critique cannot pass `image-ux-review-gate.json`. |
 | Computer Use fast lane | Uses `$Computer-Use` / `$CU` for UI/browser/visual work that needs maximum speed: skip Team debate and upfront TriWiki loops, use Codex Computer Use directly, then refresh/validate TriWiki and run Honest Mode at final closeout. |
-| Goal | Provides a fast SKS bridge overlay for Codex native persisted `/goal` create, pause, resume, and clear controls; implementation continues through the selected SKS execution route. |
+| Goal | Provides a fast SKS bridge overlay for Codex native persisted `/goal` create, pause, resume, and clear controls; an ambient non-disruptive Goal continuation overlay is also recorded in normal pipeline plans while implementation continues through the selected SKS execution route. |
 | TriWiki voxels | Maintains `.sneakoscope/wiki/context-pack.json` as the context SSOT with coordinate anchors, voxel metadata, `attention.use_first`, `attention.hydrate_first`, and prompt-bound mistake recall ledgers. |
 | Context7 | Requires current docs for external packages, APIs, MCPs, SDKs, and framework/runtime behavior when correctness depends on current guidance. |
 | Design SSOT | Treats `design.md` as the only design decision source of truth. `docs/Design-Sys-Prompt.md` is the builder prompt; getdesign.md, official getdesign docs, and curated DESIGN.md examples from `VoltAgent/awesome-design-md` are source inputs that must be fused into `design.md` or route-local style tokens instead of becoming parallel authorities. |
@@ -80,7 +80,7 @@ The default `sks` runtime checks npm for newer `sneakoscope` and `@openai/codex`
 - Checks npm for newer `sneakoscope` and `@openai/codex` versions before launch and asks whether to update when the terminal can answer y/n.
 - Installs the latest Codex CLI with `npm i -g @openai/codex@latest` when it is missing and you approve or pass `--yes`.
 - Requires tmux 3.x or newer before opening the session.
-- Creates or reuses a named detached tmux session and prints only the session, gate, attach, and blocker details needed to act.
+- Creates a named detached tmux cockpit with multiple panes and prints only the session, gate, attach, and blocker details needed to act.
 
 ## Installation
 
@@ -167,7 +167,7 @@ sks tmux check
 sks tmux status --once
 ```
 
-Bare `sks` creates or reuses the default named tmux session for Codex CLI and attaches to it in an interactive terminal. By default it launches Codex in the SKS fast-high runtime (`--model gpt-5.5 -c model_reasoning_effort="high"`) with a static SKS 3D ASCII intro inside tmux; the animated intro is reserved for non-tmux unauthenticated Codex launches and can be disabled with `SKS_TMUX_LOGO_ANIMATION=0`. SKS always forces the model to `gpt-5.5`; `SKS_CODEX_MODEL` and `SKS_CODEX_FAST_HIGH=0` cannot downgrade or remove that model pin. You can still set `SKS_CODEX_REASONING` to change reasoning effort. Use `sks tmux open` when you need explicit `--workspace` / `--session` flags, `sks tmux check` for readiness without launching, and `sks help` for CLI help. Use `--no-attach` or `SKS_TMUX_NO_AUTO_ATTACH=1` when you only want SKS to create/reuse the session and print the manual attach command.
+Bare `sks` creates or reuses the default named tmux session for Codex CLI and attaches to it in an interactive terminal. By default it launches Codex in Fast service tier with `--model gpt-5.5`, `-c service_tier="fast"`, and the selected `model_reasoning_effort` with a static SKS 3D ASCII intro inside tmux; the animated intro is reserved for non-tmux unauthenticated Codex launches and can be disabled with `SKS_TMUX_LOGO_ANIMATION=0`. SKS always forces the model to `gpt-5.5`; `SKS_CODEX_MODEL` and `SKS_CODEX_FAST_HIGH=0` cannot downgrade or remove that model pin. You can still set `SKS_CODEX_REASONING` to change reasoning effort. Use `sks tmux open` when you need explicit `--workspace` / `--session` flags, `sks tmux check` for readiness without launching, and `sks help` for CLI help. Use `--no-attach` or `SKS_TMUX_NO_AUTO_ATTACH=1` when you only want SKS to create/reuse the session and print the manual attach command.
 
 Before opening tmux, SKS checks the installed Codex CLI against npm `@openai/codex@latest`. If a newer version exists, it asks `Y/n`; answering `y` updates automatically with `npm i -g @openai/codex@latest` and then opens tmux with the updated Codex CLI.
 
@@ -211,7 +211,7 @@ sks --mad
 sks --mad --yes
 ```
 
-This syncs existing codex-lb/Codex CLI auth before launch, creates/uses the `sks-mad-high` Codex profile for a one-shot full-access, high-reasoning tmux session with `sandbox_mode = "danger-full-access"` and `approval_policy = "never"`, opens an active MAD-SKS permission gate for that tmux run, then launches Codex with `--sandbox danger-full-access --ask-for-approval never` and attaches to the session in an interactive terminal. If codex-lb is configured and no explicit `--workspace`/`--session` was passed, SKS opens a fresh tmux session so the repaired key is loaded by the Codex process immediately. While the gate is active, live server work, Supabase MCP database writes, direct SQL, targeted DML, schema cleanup, Supabase MCP `apply_migration`, and required Supabase CLI migration application such as `supabase migration up` or `supabase db push` are allowed. Catastrophic database wipe/all-row/project-management safeguards remain active. Repeat launches reuse the same named SKS MAD tmux session unless auth repair requires a fresh codex-lb session.
+This syncs existing codex-lb/Codex CLI auth before launch, creates/uses the `sks-mad-high` Codex profile for a one-shot full-access, high-reasoning tmux cockpit with `sandbox_mode = "danger-full-access"` and `approval_policy = "never"`, opens an active MAD-SKS permission gate for that tmux run, then launches a tiled cockpit: Codex CLI, MAD permission gate status, and live guide panes. The cockpit recreates the named session on launch so stale single-pane sessions do not hide the split layout, then attaches in an interactive terminal. If codex-lb is configured and no explicit `--workspace`/`--session` was passed, SKS opens a fresh tmux session so the repaired key is loaded by the Codex process immediately. While the gate is active, live server work, Supabase MCP database writes, direct SQL, targeted DML, schema cleanup, Supabase MCP `apply_migration`, and required Supabase CLI migration application such as `supabase migration up` or `supabase db push` are allowed. Catastrophic database wipe/all-row/project-management safeguards remain active.
 
 MAD does not disable the pipeline contract: stages, executors, reviewers, and auto-review policy still must not invent unrequested fallback implementation code. If the requested path cannot be implemented, SKS should block with evidence rather than add substitute behavior.
 
@@ -239,10 +239,12 @@ sks team log latest
 
 By default, Team missions keep at least five QA/reviewer lanes active. Use explicit role counts only when you need to raise or otherwise pin the lane mix for a specific mission.
 
-Team mode prepares the mission, records live events, compiles runtime tasks and worker inboxes, writes schema-backed effort/work-order/dashboard artifacts, and opens a named tmux Team session with split live lanes when tmux is available. The default terminal output stays compact: mission id, agent count, role count, tmux status, watch command, and artifact directory. `sks team dashboard` renders the cockpit panes for mission overview, agent lanes, task DAG, QA/dogfood, artifacts/evidence, and performance.
+Team mode prepares the mission, records live events, compiles runtime tasks and worker inboxes, writes schema-backed effort/work-order/dashboard artifacts, and opens a named tmux Team session with split live lanes by default when tmux is available. Use `--no-open-tmux` for artifact-only mission creation. The default terminal output stays compact: mission id, agent count, role count, tmux status, watch command, and artifact directory. `sks team dashboard` renders the cockpit panes for mission overview, agent lanes, task DAG, QA/dogfood, artifacts/evidence, and performance.
 
 The tmux Team launch is a live orchestration screen in one tmux window: the first pane follows `sks team watch <mission-id> --follow` as the mission overview, and neighboring split panes follow individual `sks team lane <mission-id> --agent <name> --follow` views. Pane headers show only mission, lane, phase, follow command, and cleanup command. SKS gives lanes role-specific colors, labels, and terminal titles, so scouts, planning/debate voices, executors, reviewers, and safety lanes are visually distinct while detailed evidence is mirrored into `team-transcript.jsonl`, `team-live.md`, and `team-dashboard.json`.
 
+Team roster and runtime artifacts now include per-agent Fast reasoning metadata. Simple bounded Team lanes can use low reasoning, tool-heavy runtime/CLI/tmux work uses medium, and knowledge, current-docs, safety, DB, release, commit, or research-heavy lanes use high or xhigh as appropriate instead of opening every scout at high.
+
 Agent sessions communicate through the bounded Team transcript. Use `sks team message <mission-id|latest> --from <agent> --to <agent|all> --message "..."` to add direct or broadcast messages; lane panes show messages addressed to that agent plus the fallback global tail.
 
 When the Team route reaches `session_cleanup`, SKS marks the tmux session record complete and asks `watch --follow` / `lane --follow` panes to show a cleanup summary and stop. You can also run `sks team cleanup-tmux <mission-id|latest>` manually, or `sks team cleanup-tmux latest --close` to kill the recorded tmux session.
@@ -412,7 +414,7 @@ Use these inside Codex App or another agent prompt. They are prompt commands, no
 | --- | --- |
 | `$Team` | You want implementation, code changes, or substantial repo work. |
 | `$From-Chat-IMG` | You have a chat screenshot plus original attachments and want each visible request mapped to work. |
-| `$DFix` | You need a tiny design/content edit such as copy, label, color, spacing, or translation, with no Team/TriWiki/reflection recording and only a one-line DFix Honest check. |
+| `$DFix` | You need Direct Fix work: tiny copy/config/docs/labels/spacing/translation/simple mechanical edits, with broad implementation still routed to Team and UI design specifics handled by the relevant UI/design route rules. |
 | `$Answer` | You want an answer only and no implementation should start. |
 | `$SKS` | You need setup, status, usage, or workflow help. |
 | `$QA-LOOP` | You want UI/API dogfooding, safe fixes, and rechecks. |

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "sneakoscope",
   "displayName": "ㅅㅋㅅ",
-  "version": "0.7.56",
+  "version": "0.7.57",
   "description": "Sneakoscope Codex: database-safe Codex CLI/App harness with Team, Goal, AutoResearch, TriWiki, and Honest Mode.",
   "type": "module",
   "homepage": "https://github.com/mandarange/Sneakoscope-Codex#readme",

package/src/cli/install-helpers.mjs

@@ -994,8 +994,8 @@ export async function selftestCodexLb(tmp) {
   if (!codexLbLaunch.includes('sks-codex-lb.env')) throw new Error('selftest failed: tmux launch command does not source codex-lb env file');
   if (!codexLbLaunch.includes("'--model' 'gpt-5.5'")) throw new Error('selftest failed: tmux launch command without args did not force GPT-5.5');
   if (!codexLbLaunch.includes('SKS_TMUX_LOGO_ANIMATION') || !codexLbLaunch.includes('SNEAKOSCOPE CODEX')) throw new Error('selftest failed: tmux launch command does not include the animated SKS logo intro');
-  const madLaunchSource = await safeReadText(path.join(packageRoot(), 'src', 'cli', 'main.mjs'));
-  if (!madLaunchSource.includes('const lb = await maybePromptCodexLbSetupForLaunch(args)') || !madLaunchSource.includes("const launchLb = lb.status === 'present'") || !madLaunchSource.includes('codexLbImmediateLaunchOpts(cleanArgs, launchLb')) throw new Error('selftest failed: MAD launch does not sync codex-lb auth and fresh-session launch options');
+  const madLaunchSource = await safeReadText(path.join(packageRoot(), 'src', 'cli', 'maintenance-commands.mjs'));
+  if (!madLaunchSource.includes('const lb = await deps.maybePromptCodexLbSetupForLaunch(args)') || !madLaunchSource.includes("const launchLb = lb.status === 'present'") || !madLaunchSource.includes('codexLbImmediateLaunchOpts(cleanArgs, launchLb')) throw new Error('selftest failed: MAD launch does not sync codex-lb auth and fresh-session launch options');
 
 }
 

package/src/cli/main.mjs

@@ -75,11 +75,11 @@ import { GOAL_WORKFLOW_ARTIFACT } from '../core/goal-workflow.mjs';
 import { CODEX_APP_DOCS_URL, codexAppIntegrationStatus, formatCodexAppStatus } from '../core/codex-app.mjs';
 import { codexAppRemoteControlCommand } from './codex-app-command.mjs';
 import { OPENCLAW_SKILL_NAME, installOpenClawSkill } from '../core/openclaw.mjs';
-import { buildTmuxLaunchPlan, buildTmuxOpenArgs, codexLaunchCommand, createTmuxSession, isTmuxShellSession, runTmuxLaunchPlanSyntaxCheck, shouldAutoAttachTmux, tmuxReadiness, tmuxStatusKind, defaultTmuxSessionName, formatTmuxBanner, launchTmuxTeamView, launchTmuxUi, platformTmuxInstallHint, runTmuxStatus, sanitizeTmuxSessionName, teamLaneStyle } from '../core/tmux-ui.mjs';
+import { buildTmuxLaunchPlan, buildTmuxOpenArgs, codexLaunchCommand, createTmuxSession, defaultCodexLaunchArgs, isTmuxShellSession, runTmuxLaunchPlanSyntaxCheck, shouldAutoAttachTmux, tmuxReadiness, tmuxStatusKind, defaultTmuxSessionName, formatTmuxBanner, launchMadTmuxUi, launchTmuxTeamView, launchTmuxUi, platformTmuxInstallHint, runTmuxStatus, sanitizeTmuxSessionName, teamLaneStyle } from '../core/tmux-ui.mjs';
 import { autoReviewProfileName, autoReviewStatus, autoReviewSummary, enableAutoReview, disableAutoReview, enableMadHighProfile, madHighProfileName } from '../core/auto-review.mjs';
 import { context7Command } from './context7-command.mjs';
 import { askPostinstallQuestion, checkContext7, checkRequiredSkills, codexLbStatus, configureCodexLb, ensureCodexCliTool, ensureGlobalCodexSkillsDuringInstall, ensureProjectContext7Config, ensureRelatedCliTools, ensureSksCommandDuringInstall, ensureTmuxCliTool, globalCodexSkillsRoot, maybePromptCodexLbSetupForLaunch, maybePromptCodexUpdateForLaunch, postinstall, postinstallBootstrapDecision, repairCodexLbAuth, selftestCodexLb, shouldAutoApproveInstall } from './install-helpers.mjs';
-import { buildTeamPlan, codeStructureCommand, dbCommand, defaultBeta, defaultVGraph, evalCommand, gcCommand, goalCommand, gxCommand, harnessCommand, hproofCommand, memoryCommand, migrateWikiContextPack, parseTeamCreateArgs, perfCommand, profileCommand, projectWikiClaims, proofFieldCommand, qaLoopCommand, quickstartCommand, researchCommand, skillDreamCommand, statsCommand, team, teamWorkflowMarkdown, validateArtifactsCommand, wikiCommand, wikiVoxelRowCount, writeWikiContextPack } from './maintenance-commands.mjs';
+import { buildTeamPlan, codeStructureCommand, dbCommand, defaultBeta, defaultVGraph, evalCommand, gcCommand, goalCommand, gxCommand, harnessCommand, hproofCommand, madHighCommand as runMadHighCommand, memoryCommand, migrateWikiContextPack, parseTeamCreateArgs, perfCommand, profileCommand, projectWikiClaims, proofFieldCommand, qaLoopCommand, quickstartCommand, researchCommand, skillDreamCommand, statsCommand, team, teamWorkflowMarkdown, validateArtifactsCommand, wikiCommand, wikiVoxelRowCount, writeWikiContextPack } from './maintenance-commands.mjs';
 import { openClawCommand } from './openclaw-command.mjs';
 
 const flag = (args, name) => args.includes(name);
@@ -96,7 +96,7 @@ function installScopeFromArgs(args = [], fallback = 'global') {
 }
 
 export async function main(args) {
-  if (isMadHighLaunch(args)) return madHighCommand(args);
+  if (isMadHighLaunch(args)) return runMadHighCommand(args, { maybePromptSksUpdateForLaunch, maybePromptCodexUpdateForLaunch, ensureMadLaunchDependencies, printDepsInstallAction, maybePromptCodexLbSetupForLaunch, packageVersion: PACKAGE_VERSION });
   if (isAutoReviewFlag(args[0])) return autoReviewCommand('start', args.slice(1));
   const [cmd, sub, ...rest] = args;
   const tail = sub === undefined ? [] : [sub, ...rest];
@@ -143,6 +143,16 @@ async function defaultTmuxCommand(args = []) {
   return launchTmuxUi(args, codexLbImmediateLaunchOpts(args, lb, { conciseBlockers: true }));
 }
 
+function codexLbImmediateLaunchOpts(args = [], lb = {}, opts = {}) {
+  if (!lb?.ok || lb.status !== 'configured') return opts;
+  if (readOption(args, '--session', null) || readOption(args, '--workspace', null)) return opts;
+  const root = readOption(args, '--root', process.cwd());
+  const session = sanitizeTmuxSessionName(`sks-codex-lb-${Date.now().toString(36)}-${defaultTmuxSessionName(root)}`);
+  console.log(`codex-lb key loaded for this launch: ${lb.env_path}`);
+  console.log(`Using fresh tmux session: ${session}`);
+  return { ...opts, session };
+}
+
 function help(args = []) {
   const topic = args[0];
   if (topic) return usage([topic]);
@@ -338,7 +348,7 @@ async function updateCheck(args = []) {
   if (result.update_available) console.log('Run:     npm i -g sneakoscope');
 }
 
-const DOLLAR_DEFAULT_PIPELINE_TEXT = 'Default pipeline: questions -> $Answer, small design/content -> $DFix, presentation/PDF artifacts -> $PPT, image-generation UI/UX reviews -> $Image-UX-Review/$UX-Review, Computer Use UI/browser speed work -> $Computer-Use, code -> $Team. Use $From-Chat-IMG only for chat screenshot plus original attachments. Use $MAD-SKS only as an explicit scoped DB authorization modifier that can be combined with another $ route. No route may invent unrequested fallback implementation code.';
+const DOLLAR_DEFAULT_PIPELINE_TEXT = 'Default pipeline: questions -> $Answer, tiny Direct Fix edits -> $DFix, presentation/PDF artifacts -> $PPT, image-generation UI/UX reviews -> $Image-UX-Review/$UX-Review, Computer Use UI/browser speed work -> $Computer-Use, code -> $Team. Use $From-Chat-IMG only for chat screenshot plus original attachments. Use $MAD-SKS only as an explicit scoped DB authorization modifier that can be combined with another $ route. No route may invent unrequested fallback implementation code.';
 
 function commands(args = []) {
   if (flag(args, '--json')) return console.log(JSON.stringify({ aliases: ['sks', 'sneakoscope'], dollar_commands: DOLLAR_COMMANDS, app_skill_aliases: DOLLAR_COMMAND_ALIASES, commands: COMMAND_CATALOG }, null, 2));
@@ -397,18 +407,20 @@ function dollarCommandNames() {
 }
 
 function dfixHelp() {
-  console.log(`SKS DFix Mode
+  console.log(`SKS Direct Fix Mode
 
 Prompt command:
-  $DFix <small design/content request>
+  $DFix <tiny direct fix request>
 
 Examples:
   $DFix 글자 색 파란색으로 바꿔줘
   $DFix 내용을 영어로 바꿔줘
   $DFix Change the CTA label to "Start"
+  $DFix Fix the README typo
+  $DFix Update the package version
 
 Purpose:
-  Fast design/content fixes only. DFix bypasses the general SKS prompt pipeline and uses an ultralight, no-record task list.
+  Fast tiny direct edits only. Direct Fix bypasses the general SKS prompt pipeline and uses an ultralight, no-record task list.
 
 Rules:
   List the exact micro-edits, inspect only needed files, apply only those edits.
@@ -1151,130 +1163,6 @@ async function codexLbCommand(action = 'status', args = []) {
   process.exitCode = 1;
 }
 
-function codexLbImmediateLaunchOpts(args = [], lb = {}, opts = {}) {
-  if (!lb?.ok || lb.status !== 'configured') return opts;
-  if (readOption(args, '--session', null) || readOption(args, '--workspace', null)) return opts;
-  const root = readOption(args, '--root', process.cwd());
-  const session = sanitizeTmuxSessionName(`sks-codex-lb-${Date.now().toString(36)}-${defaultTmuxSessionName(root)}`);
-  console.log(`codex-lb key loaded for this launch: ${lb.env_path}`);
-  console.log(`Using fresh tmux session: ${session}`);
-  return { ...opts, session };
-}
-
-async function madHighCommand(args = []) {
-  const cleanArgs = args.filter((arg) => !['--mad', '--MAD', '--mad-sks', '--high', '--no-auto-install-tmux'].includes(arg));
-  if (flag(args, '--json')) {
-    const profile = await enableMadHighProfile();
-    return console.log(JSON.stringify(profile, null, 2));
-  }
-  const update = await maybePromptSksUpdateForLaunch(args, { label: 'MAD launch' });
-  if (update.status === 'updated') {
-    console.log(`SKS updated from ${PACKAGE_VERSION} to ${update.latest}. Rerun: sks --mad`);
-    return;
-  }
-  if (update.status === 'failed') {
-    console.error(`SKS update failed: ${update.error}`);
-    process.exitCode = 1;
-    return;
-  }
-  const codexUpdate = await maybePromptCodexUpdateForLaunch(args, { label: 'MAD launch' });
-  if (codexUpdate.status === 'failed' || codexUpdate.status === 'updated_not_reflected') {
-    console.error(`Codex CLI update failed: ${codexUpdate.error || 'updated version was not visible on PATH'}`);
-    process.exitCode = 1;
-    return;
-  }
-  const deps = await ensureMadLaunchDependencies(args);
-  if (!deps.ready) {
-    console.error('SKS MAD launch blocked by missing dependencies.');
-    for (const action of deps.actions) printDepsInstallAction(action);
-    process.exitCode = 1;
-    return;
-  }
-  const lb = await maybePromptCodexLbSetupForLaunch(args);
-  if (lb.status === 'missing_api_key') {
-    process.exitCode = 1;
-    return;
-  }
-  const profile = await enableMadHighProfile();
-  const madLaunch = await activateMadTmuxPermissionState(process.cwd());
-  console.log(`SKS MAD ready: ${madHighProfileName()} | gate ${madLaunch.mission_id}`);
-  console.log('Live full-access active; catastrophic DB wipe/all-row/project-management guards remain.');
-  const launchLb = lb.status === 'present' ? { ...lb, status: 'configured' } : lb;
-  const launchOpts = codexLbImmediateLaunchOpts(cleanArgs, launchLb, {
-    codexArgs: profile.launch_args,
-    autoInstallTmux: !flag(args, '--no-auto-install-tmux'),
-    conciseBlockers: true
-  });
-  const workspace = readOption(cleanArgs, '--workspace', readOption(cleanArgs, '--session', launchOpts.session || `sks-mad-${defaultTmuxSessionName(process.cwd())}`));
-  return launchTmuxUi([...cleanArgs, '--workspace', workspace], {
-    ...launchOpts,
-    codexArgs: profile.launch_args,
-    autoInstallTmux: !flag(args, '--no-auto-install-tmux'),
-    conciseBlockers: true
-  });
-}
-
-async function activateMadTmuxPermissionState(cwd = process.cwd()) {
-  const root = await sksRoot();
-  if (!(await exists(path.join(root, '.sneakoscope')))) await initProject(root, {});
-  const { id, dir } = await createMission(root, { mode: 'mad-sks', prompt: 'sks --mad tmux live full-access session' });
-  const gate = {
-    schema_version: 1,
-    passed: false,
-    mad_sks_permission_active: true,
-    permissions_deactivated: false,
-    live_server_writes_allowed: true,
-    supabase_mcp_schema_cleanup_allowed: true,
-    direct_execute_sql_allowed: true,
-    normal_db_writes_allowed: true,
-    migration_apply_allowed: true,
-    catastrophic_safety_guard_active: true,
-    permission_profile: permissionGateSummary(),
-    activated_by: 'sks --mad',
-    cwd: path.resolve(cwd || process.cwd())
-  };
-  await writeJsonAtomic(path.join(dir, 'mad-sks-gate.json'), gate);
-  await writeJsonAtomic(path.join(dir, 'route-context.json'), {
-    route: 'MadSKS',
-    command: '$MAD-SKS',
-    mode: 'MADSKS',
-    task: gate.activated_by,
-    mad_sks_authorization: true,
-    tmux_launch: true,
-    permission_profile: gate.permission_profile
-  });
-  await appendJsonlBounded(path.join(dir, 'events.jsonl'), {
-    ts: nowIso(),
-    type: 'mad_sks.tmux_permission_opened',
-    route: 'MadSKS',
-    live_server_writes_allowed: true,
-    catastrophic_safety_guard_active: true
-  });
-  await setCurrent(root, {
-    mission_id: id,
-    route: 'MadSKS',
-    route_command: '$MAD-SKS',
-    mode: 'MADSKS',
-    phase: 'MADSKS_TMUX_PERMISSION_ACTIVE',
-    questions_allowed: false,
-    implementation_allowed: true,
-    mad_sks_active: true,
-    mad_sks_modifier: true,
-    mad_sks_gate_file: 'mad-sks-gate.json',
-    mad_sks_gate_ready: true,
-    live_server_writes_allowed: true,
-    supabase_mcp_schema_cleanup_allowed: true,
-    direct_execute_sql_allowed: true,
-    normal_db_writes_allowed: true,
-    migration_apply_allowed: true,
-    catastrophic_safety_guard_active: true,
-    permission_profile: gate.permission_profile,
-    stop_gate: 'mad-sks-gate.json',
-    prompt: gate.activated_by
-  });
-  return { mission_id: id, dir, gate };
-}
-
 async function maybePromptSksUpdateForLaunch(args = [], opts = {}) {
   if (flag(args, '--json') || flag(args, '--skip-update-check') || process.env.SKS_SKIP_UPDATE_CHECK === '1') return { status: 'skipped' };
   const latest = await npmPackageVersion('sneakoscope');
@@ -1712,7 +1600,7 @@ async function setup(args) {
   console.log(`Codex App: .codex/config.toml, .codex/hooks.json, .agents/skills, .codex/agents, .codex/SNEAKOSCOPE.md`);
   console.log(`Global $:  ${globalSkills.status === 'installed' ? 'ok' : globalSkills.status} ${globalSkills.root || ''}`.trimEnd());
   console.log(`App tools: ${appRuntime.ok ? 'ok' : 'needs setup'} Codex App=${appRuntime.app.installed ? 'ok' : 'missing'} Browser Use=${appRuntime.mcp.has_browser_use ? 'ok' : 'missing'} Computer Use=${appRuntime.mcp.has_computer_use ? 'ok' : 'missing'} Image Gen=${appRuntime.features?.image_generation ? 'ok' : 'missing'}`);
-  console.log(`Prompt:    intent-first routing, $Answer fact-check route, $DFix ultralight design/content route, $PPT HTML/PDF presentation route, Context7 gate`);
+  console.log(`Prompt:    intent-first routing, $Answer fact-check route, $DFix ultralight Direct Fix route, $PPT HTML/PDF presentation route, Context7 gate`);
   console.log(`Skills:    .agents/skills`);
   console.log(`Next:      sks context7 check; sks selftest --mock; sks commands; sks dollar-commands`);
   if (cliTools.codex.status === 'failed') console.log(`\nCodex CLI install failed. Run manually: npm i -g @openai/codex. ${cliTools.codex.error || ''}`.trim());
@@ -2266,7 +2154,7 @@ async function selftest() {
   const madProfilePath = path.join(tmp, 'mad-codex-config.toml');
   const madProfile = await enableMadHighProfile({ configPath: madProfilePath });
   const madProfileText = await safeReadText(madProfilePath);
-  if (madProfile.profile_name !== 'sks-mad-high' || !madProfileText.includes('sandbox_mode = "danger-full-access"') || !madProfileText.includes('approval_policy = "never"') || !madProfileText.includes('approvals_reviewer = "auto_review"') || !madProfile.launch_args.includes('--sandbox') || !madProfile.launch_args.includes('danger-full-access') || !madProfile.launch_args.includes('--ask-for-approval') || !madProfile.launch_args.includes('never') || !madProfileText.includes('model_reasoning_effort = "high"') || !madProfileText.includes('unrequested fallback implementation code')) throw new Error('selftest failed: MAD high profile is not Codex full-access high with fallback-code guard');
+  if (madProfile.profile_name !== 'sks-mad-high' || !madProfileText.includes('sandbox_mode = "danger-full-access"') || !madProfileText.includes('approval_policy = "never"') || !madProfileText.includes('approvals_reviewer = "auto_review"') || !madProfileText.includes('service_tier = "fast"') || !madProfile.launch_args.includes('--sandbox') || !madProfile.launch_args.includes('danger-full-access') || !madProfile.launch_args.includes('--ask-for-approval') || !madProfile.launch_args.includes('never') || !madProfileText.includes('model_reasoning_effort = "high"') || !madProfileText.includes('unrequested fallback implementation code')) throw new Error('selftest failed: MAD high profile is not Codex full-access high with fallback-code guard');
   if (!isMadHighLaunch(['--mad', '--high']) || isMadHighLaunch(['db', '--mad'])) throw new Error('selftest failed: MAD high launch flag parsing is not top-level only');
   const workspacePlan = { session: 'sks-mad-selftest', root: tmp, codexArgs: madProfile.launch_args };
   const tmuxSyntax = runTmuxLaunchPlanSyntaxCheck(workspacePlan);
@@ -2274,11 +2162,11 @@ async function selftest() {
   const tmuxOpenArgs = buildTmuxOpenArgs(workspacePlan);
   if (tmuxOpenArgs.join(' ') !== 'attach-session -t sks-mad-selftest') throw new Error('selftest failed: MAD tmux attach args are not stable by session name');
   const defaultFastHighPlan = await buildTmuxLaunchPlan({ root: tmp, tmux: { ok: true, bin: 'tmux', version: '3.4' }, codex: { bin: 'codex', version: 'codex-cli 99.0.0' }, app: { ok: true } });
-  if (defaultFastHighPlan.codexArgs.join(' ') !== '--model gpt-5.5 -c model_reasoning_effort="high"') throw new Error('selftest failed: default sks tmux launch is not fast-high');
+  if (defaultFastHighPlan.codexArgs.join(' ') !== '--model gpt-5.5 -c service_tier="fast" -c model_reasoning_effort="high"') throw new Error('selftest failed: default sks tmux launch is not fast-high');
   const forcedModelPlan = await buildTmuxLaunchPlan({ root: tmp, env: { SKS_CODEX_MODEL: 'gpt-5.4-mini', SKS_CODEX_FAST_HIGH: '0', SKS_CODEX_REASONING: 'medium' }, tmux: { ok: true, bin: 'tmux', version: '3.4' }, codex: { bin: 'codex', version: 'codex-cli 99.0.0' }, app: { ok: true } });
-  if (forcedModelPlan.codexArgs.includes('gpt-5.4-mini') || forcedModelPlan.codexArgs.join(' ') !== '--model gpt-5.5 -c model_reasoning_effort="medium"') throw new Error('selftest failed: sks tmux launch allowed a non-GPT-5.5 model override');
+  if (forcedModelPlan.codexArgs.includes('gpt-5.4-mini') || forcedModelPlan.codexArgs.join(' ') !== '--model gpt-5.5 -c service_tier="fast" -c model_reasoning_effort="medium"') throw new Error('selftest failed: sks tmux launch allowed a non-GPT-5.5 model override');
   const explicitBadModelPlan = await buildTmuxLaunchPlan({ root: tmp, codexArgs: ['--profile', 'legacy-5.4', '--model', 'gpt-5.4-mini', '-c', 'model="gpt-5.4"', '-c', 'model_reasoning_effort="low"'], tmux: { ok: true, bin: 'tmux', version: '3.4' }, codex: { bin: 'codex', version: 'codex-cli 99.0.0' }, app: { ok: true } });
-  if (explicitBadModelPlan.codexArgs.join(' ').includes('gpt-5.4') || explicitBadModelPlan.codexArgs.join(' ') !== '--model gpt-5.5 --profile legacy-5.4 -c model_reasoning_effort="low"') throw new Error('selftest failed: explicit tmux model override was not forced back to GPT-5.5');
+  if (explicitBadModelPlan.codexArgs.join(' ').includes('gpt-5.4') || explicitBadModelPlan.codexArgs.join(' ') !== '--model gpt-5.5 -c service_tier="fast" --profile legacy-5.4 -c model_reasoning_effort="low"') throw new Error('selftest failed: explicit tmux model override was not forced back to GPT-5.5');
   const codexExecArgs = buildCodexExecArgs({ root: tmp, prompt: 'model guard selftest', profile: 'legacy-5.4', extraArgs: ['--model=gpt-5.4-mini', '--config', 'model = "gpt-5.4"', '-c', 'model_reasoning_effort="medium"'] });
   if (codexExecArgs.join(' ').includes('gpt-5.4') || !codexExecArgs.includes('gpt-5.5') || codexExecArgs.includes('--model=gpt-5.4-mini')) throw new Error('selftest failed: codex exec args allowed a non-GPT-5.5 model override');
   await selftestCodexLb(tmp);
@@ -2545,8 +2433,27 @@ async function selftest() {
   if (routePrompt('[$research](/tmp/research/SKILL.md) Codex Computer Use 도구 노출 문제를 QA루프 관점에서 연구')?.id !== 'Research') throw new Error('selftest failed: markdown-linked $Research was not treated as explicit route');
   if (routePrompt('$WikiRefresh 갱신')) throw new Error('selftest failed: deprecated $WikiRefresh route still resolved');
   if (routePrompt('$MAD-SKS Supabase MCP main 작업')?.id !== 'MadSKS') throw new Error('selftest failed: $MAD-SKS route did not resolve');
+  if (routePrompt('$MAD-SKS 버튼 라벨만 바꿔줘')?.id === 'DFix') throw new Error('selftest failed: $MAD-SKS tiny label fix incorrectly routed to DFix');
   if (routePrompt('$MAD-SKS $Team Supabase MCP main 작업')?.id !== 'Team') throw new Error('selftest failed: $MAD-SKS did not compose with $Team');
+  if (routePrompt('$MAD-SKS $Team 버튼 라벨만 바꿔줘')?.id !== 'Team') throw new Error('selftest failed: $MAD-SKS $Team tiny fix did not stay on Team route');
   if (routePrompt('$DB Supabase 점검 $MAD-SKS')?.id !== 'DB') throw new Error('selftest failed: trailing $MAD-SKS changed primary route');
+  if (routePrompt('Fix the typo in README')?.id !== 'DFix') throw new Error('selftest failed: inferred typo Direct Fix did not route to DFix');
+  if (routePrompt('Update the package version to 1.2.3')?.id !== 'DFix') throw new Error('selftest failed: inferred package-version Direct Fix did not route to DFix');
+  if (routePrompt('package.json version만 1.2.3으로 바꿔줘')?.id !== 'DFix') throw new Error('selftest failed: inferred package.json version Direct Fix did not route to DFix');
+  if (routePrompt('How do I fix the typo in README?')?.id !== 'Answer') throw new Error('selftest failed: how-to Direct Fix question did not route to Answer');
+  if (routePrompt('How do I change README title?')?.id !== 'Answer') throw new Error('selftest failed: how-to README title question did not route to Answer');
+  if (routePrompt('How do I make a settings page?')?.id !== 'Answer') throw new Error('selftest failed: how-to create question did not route to Answer');
+  if (routePrompt('How to create a new form component?')?.id !== 'Answer') throw new Error('selftest failed: how-to form component question did not route to Answer');
+  if (routePrompt('How can I build a modal?')?.id !== 'Answer') throw new Error('selftest failed: how-can-I build question did not route to Answer');
+  if (routePrompt('Make a button')?.id !== 'Team') throw new Error('selftest failed: create-style button work did not route to Team');
+  if (routePrompt('Make a button that submits the form')?.id !== 'Team') throw new Error('selftest failed: form button creation did not route to Team');
+  if (routePrompt('Change button to submit the form')?.id !== 'Team') throw new Error('selftest failed: form button behavior change did not route to Team');
+  if (routePrompt('버튼이 폼 제출하게 바꿔줘')?.id !== 'Team') throw new Error('selftest failed: Korean form button behavior change did not route to Team');
+  if (routePrompt('Can you change the button to submit the form?')?.id !== 'Team') throw new Error('selftest failed: polite form button behavior request did not route to Team');
+  if (routePrompt('Change button label to Submit')?.id !== 'DFix') throw new Error('selftest failed: button label Direct Fix did not route to DFix');
+  if (routePrompt('Change button text to Submit')?.id !== 'DFix') throw new Error('selftest failed: button text Direct Fix did not route to DFix');
+  if (routePrompt('Can you change the button label to Save?')?.id !== 'DFix') throw new Error('selftest failed: polite button label Direct Fix did not route to DFix');
+  if (routePrompt('Make README generator work')?.id !== 'Team') throw new Error('selftest failed: README generator implementation did not route to Team');
   const imageUxRoute = routePrompt('$Image-UX-Review localhost 화면 검수');
   if (imageUxRoute?.id !== 'ImageUXReview') throw new Error('selftest failed: $Image-UX-Review did not route to ImageUXReview');
   if (routePrompt('$UX-Review 스크린샷 gpt-image-2 콜아웃 리뷰')?.id !== 'ImageUXReview') throw new Error('selftest failed: $UX-Review did not route to ImageUXReview');
@@ -2939,6 +2846,16 @@ async function selftest() {
   if (await exists(path.join(hookDfixTmp, '.sneakoscope', 'state', 'light-route-stop.json'))) throw new Error('selftest failed: $DFix hook created persistent light-route state');
   const hookDfixState = await readJson(stateFile(hookDfixTmp), {});
   if (String(hookDfixState.phase || '').includes('CLARIFICATION_AWAITING_ANSWERS')) throw new Error('selftest failed: $DFix state entered clarification gate');
+  const explicitDfixDirectTmp = tmpdir();
+  await initProject(explicitDfixDirectTmp, {});
+  const explicitDfixDirectPayload = JSON.stringify({ cwd: explicitDfixDirectTmp, prompt: '$DFix Update the docs config wording to Direct Fix' });
+  const explicitDfixDirectResult = await runProcess(process.execPath, [hookBin, 'hook', 'user-prompt-submit'], { cwd: explicitDfixDirectTmp, input: explicitDfixDirectPayload, env: { SKS_DISABLE_UPDATE_CHECK: '1' }, timeoutMs: 15000, maxOutputBytes: 128 * 1024 });
+  if (explicitDfixDirectResult.code !== 0) throw new Error(`selftest failed: explicit Direct Fix docs/config hook exited ${explicitDfixDirectResult.code}: ${explicitDfixDirectResult.stderr}`);
+  const explicitDfixDirectJson = JSON.parse(explicitDfixDirectResult.stdout);
+  const explicitDfixDirectContext = explicitDfixDirectJson.hookSpecificOutput?.additionalContext || '';
+  if (!explicitDfixDirectContext.includes('DFix ultralight pipeline active')) throw new Error('selftest failed: explicit Direct Fix docs/config request did not use ultralight hook');
+  if (explicitDfixDirectContext.includes('SKS skill-first pipeline active') || explicitDfixDirectContext.includes('Mission:')) throw new Error('selftest failed: explicit Direct Fix docs/config request leaked general pipeline context');
+  if (await exists(path.join(explicitDfixDirectTmp, '.sneakoscope', 'state', 'light-route-stop.json'))) throw new Error('selftest failed: explicit Direct Fix docs/config hook created persistent light-route state');
   const inferredDfixPayload = JSON.stringify({ cwd: hookTeamTmp, prompt: '버튼 라벨 바꿔줘' });
   const inferredDfixResult = await runProcess(process.execPath, [hookBin, 'hook', 'user-prompt-submit'], { cwd: hookTeamTmp, input: inferredDfixPayload, env: { SKS_DISABLE_UPDATE_CHECK: '1' }, timeoutMs: 15000, maxOutputBytes: 128 * 1024 });
   if (inferredDfixResult.code !== 0) throw new Error(`selftest failed: inferred DFix hook exited ${inferredDfixResult.code}: ${inferredDfixResult.stderr}`);
@@ -3196,6 +3113,15 @@ async function selftest() {
   if (teamPlan.role_counts.executor !== 3 || teamPlan.role_counts.user !== 1 || teamPlan.role_counts.reviewer !== 5) throw new Error('selftest failed: team default role counts invalid');
   if (teamPlan.codex_config_required?.features?.hooks !== true || teamPlan.codex_config_required?.features?.codex_hooks === true) throw new Error('selftest failed: team plan Codex config still uses legacy hooks feature flag');
   if (!teamPlan.review_gate?.passed || teamPlan.review_gate.required_reviewer_lanes !== 5) throw new Error('selftest failed: team review policy gate did not pass default plan');
+  if (teamPlan.codex_config_required?.service_tier !== 'fast' || teamPlan.reasoning?.service_tier !== 'fast') throw new Error('selftest failed: team plan did not require Fast service tier');
+  if (!teamPlan.goal_continuation?.enabled || teamPlan.goal_continuation?.mode !== 'ambient_codex_native_goal_overlay') throw new Error('selftest failed: Team plan did not include ambient Goal continuation');
+  if (!teamPlan.roster.analysis_team.every((agent) => agent.service_tier === 'fast' && agent.reasoning_effort && agent.reasoning_profile)) throw new Error('selftest failed: analysis scouts missing dynamic Fast reasoning metadata');
+  const simpleTeamPlan = buildTeamPlan(teamId, '$Team 간단한 코드 수정');
+  if (!simpleTeamPlan.roster.analysis_team.some((agent) => agent.reasoning_effort === 'low')) throw new Error('selftest failed: simple Team prompt did not allow low-reasoning scouts');
+  const toolingTeamPlan = buildTeamPlan(teamId, '$Team tmux CLI tool-calling runtime fix');
+  if (!toolingTeamPlan.roster.analysis_team.some((agent) => agent.reasoning_effort === 'medium')) throw new Error('selftest failed: tool-heavy Team prompt did not assign medium-reasoning scouts');
+  const researchTeamPlan = buildTeamPlan(teamId, '$Team external library research and current docs update');
+  if (!researchTeamPlan.roster.analysis_team.some((agent) => agent.reasoning_effort === 'high')) throw new Error('selftest failed: research/docs Team prompt did not assign high-reasoning scouts');
   const underProvisionedReviewCount = 2;
   const blockedReviewGate = evaluateTeamReviewPolicyGate({ roleCounts: { reviewer: underProvisionedReviewCount }, agentSessions: 3, roster: { validation_team: [{ id: 'reviewer_1', role: 'reviewer' }] } });
   if (blockedReviewGate.passed || !blockedReviewGate.blockers.includes('validation_team_reviewers_below_required')) throw new Error('selftest failed: team review policy gate did not block under-provisioned review');
@@ -3237,6 +3163,8 @@ async function selftest() {
   if (roleParsed.roleCounts.executor !== 5 || roleParsed.roleCounts.reviewer !== 6 || roleParsed.agentSessions !== 6 || roleParsed.prompt !== '작업') throw new Error('selftest failed: team role-count parsing');
   const openTmuxFlagParsed = parseTeamCreateArgs(['--open-tmux', '작업']);
   if (openTmuxFlagParsed.prompt !== '작업') throw new Error('selftest failed: team --open-tmux leaked into prompt');
+  const noOpenTmuxFlagParsed = parseTeamCreateArgs(['--no-open-tmux', '작업']);
+  if (noOpenTmuxFlagParsed.prompt !== '작업') throw new Error('selftest failed: team --no-open-tmux leaked into prompt');
   const roleTeamPlan = buildTeamPlan(teamId, '역할 팀 테스트', { roleCounts: roleParsed.roleCounts });
   if (roleTeamPlan.roster.debate_team.length !== 5) throw new Error('selftest failed: executor role count not reflected in debate team size');
   if (roleTeamPlan.roster.analysis_team.length !== 5) throw new Error('selftest failed: executor role count not reflected in analysis scout team');
@@ -3251,6 +3179,28 @@ async function selftest() {
   if (teamLaneStyle('analysis_scout_1').role !== 'scout' || teamLaneStyle('executor_1').role !== 'execution' || teamLaneStyle('reviewer_1').role !== 'review') throw new Error('selftest failed: Team tmux role palette did not classify lane roles');
   if (!String(tmuxTeam.cleanup_policy || '').includes('mark-complete') || !tmuxTeam.lanes.every((entry) => entry.style?.color && entry.title)) throw new Error('selftest failed: Team tmux view did not expose color/title metadata and cleanup policy');
   if (tmuxTeam.session !== `sks-team-${teamId}` || !tmuxTeam.attach_command?.includes(`sks-team-${teamId}`)) throw new Error('selftest failed: Team tmux session is not named for visibility');
+  const fakeTmuxDir = path.join(tmp, 'fake-tmux');
+  await ensureDir(fakeTmuxDir);
+  const fakeTmuxLog = path.join(fakeTmuxDir, 'tmux.log');
+  const fakeTmuxBin = path.join(fakeTmuxDir, 'tmux');
+  await writeTextAtomic(fakeTmuxBin, `#!/usr/bin/env node\nconst fs = require('node:fs');\nconst log = process.env.SKS_FAKE_TMUX_LOG;\nif (log) fs.appendFileSync(log, process.argv.slice(2).join(' ') + '\\n');\nconst cmd = process.argv[2];\nif (cmd === 'has-session') process.exit(0);\nif (cmd === 'kill-session') process.exit(0);\nif (cmd === 'new-session') { console.log('%1'); process.exit(0); }\nif (cmd === 'split-window') { console.log('%2'); process.exit(0); }\nprocess.exit(0);\n`);
+  await fsp.chmod(fakeTmuxBin, 0o755);
+  const previousFakeTmuxLog = process.env.SKS_FAKE_TMUX_LOG;
+  process.env.SKS_FAKE_TMUX_LOG = fakeTmuxLog;
+  const recreatedTmux = await createTmuxSession({ root: tmp, session: 'sks-existing-selftest', tmux: { bin: fakeTmuxBin }, codex: { bin: process.execPath } }, [
+    { cwd: tmp, command: 'pwd', role: 'overview' },
+    { cwd: tmp, command: 'pwd', role: 'lane' }
+  ], { recreate: true });
+  const fakeTmuxLogText = await safeReadText(fakeTmuxLog);
+  if (!recreatedTmux.ok || !fakeTmuxLogText.includes('kill-session -t sks-existing-selftest') || !fakeTmuxLogText.includes('new-session') || !fakeTmuxLogText.includes('split-window')) throw new Error('selftest failed: tmux recreate did not replace stale existing session with split panes');
+  await writeTextAtomic(fakeTmuxLog, '');
+  const madCockpit = await launchMadTmuxUi(['--workspace', 'sks-mad-selftest-ui', '--no-attach'], { root: tmp, tmux: { ok: true, bin: fakeTmuxBin, version: '3.4' }, codex: { bin: process.execPath }, app: { ok: true, guidance: [] }, missionId: 'M-MAD-SELFTEST' });
+  const madTmuxLogText = await safeReadText(fakeTmuxLog);
+  if (!madCockpit.created || madCockpit.opened?.panes?.length !== 3 || !madTmuxLogText.includes('new-session') || !madTmuxLogText.includes('split-window')) throw new Error('selftest failed: MAD tmux launch did not create a multi-pane cockpit');
+  if (previousFakeTmuxLog === undefined) delete process.env.SKS_FAKE_TMUX_LOG;
+  else process.env.SKS_FAKE_TMUX_LOG = previousFakeTmuxLog;
+  const codexLaunchArgs = defaultCodexLaunchArgs({ SKS_CODEX_REASONING: 'low' }).join(' ');
+  if (!codexLaunchArgs.includes('service_tier="fast"') || !codexLaunchArgs.includes('model_reasoning_effort="low"')) throw new Error('selftest failed: Codex tmux launch args do not force Fast service tier plus dynamic reasoning');
   await initTeamLive(teamId, teamDir, '역할 팀 테스트', { agentSessions: roleTeamPlan.agent_session_count, roleCounts: roleTeamPlan.role_counts, roster: roleTeamPlan.roster });
   const teamWatch = await renderTeamWatch(teamDir, { missionId: teamId });
   if (!roleTeamPlan.roster.analysis_team.every((agent) => teamWatch.includes(`- ${agent.id}:`))) throw new Error('selftest failed: Team watch overview collapsed numbered analysis scout lanes');
@@ -3258,6 +3208,9 @@ async function selftest() {
   if (routeReasoning(routePrompt('$From-Chat-IMG 채팅 이미지 작업'), '$From-Chat-IMG 채팅 이미지 작업').effort !== 'xhigh') throw new Error('selftest failed: From-Chat-IMG reasoning not xhigh');
   if (routeReasoning(routePrompt('$Computer-Use localhost UI smoke'), '$Computer-Use localhost UI smoke').effort !== 'low') throw new Error('selftest failed: Computer Use fast lane reasoning not low');
   if (routeReasoning(routePrompt('$DB migration'), '$DB migration').effort !== 'high') throw new Error('selftest failed: logical reasoning not high');
+  if (routeReasoning(routePrompt('$Team 간단한 코드 수정'), '$Team 간단한 코드 수정').effort !== 'low') throw new Error('selftest failed: simple Team reasoning not low');
+  if (routeReasoning(routePrompt('$Team tmux CLI tool-calling fix'), '$Team tmux CLI tool-calling fix').effort !== 'medium') throw new Error('selftest failed: tool-heavy Team reasoning not medium');
+  if (routeReasoning(routePrompt('$Team library research current docs'), '$Team library research current docs').effort !== 'high') throw new Error('selftest failed: research/docs Team reasoning not high');
   const lowReasoning = routeReasoning({ id: 'LowSmoke', reasoningPolicy: 'low' }, 'small metadata read');
   if (lowReasoning.effort !== 'low' || lowReasoning.profile !== 'sks-task-low') throw new Error('selftest failed: low reasoning did not route to sks-task-low');
   const forensicEffort = selectEffort({ mission_id: 'selftest', task_id: 'TASK-IMG', route: 'from-chat-img', prompt: '$From-Chat-IMG screenshot match' });
@@ -3296,6 +3249,8 @@ async function selftest() {
   if (routePrompt('근데 왜 팀원 구성을 안하고 작업을 하는 경우가 이렇게 많지?')?.id !== 'Team') throw new Error('selftest failed: question-shaped Team complaint did not route to Team');
   if (routePrompt('$DF button label')) throw new Error('selftest failed: deprecated $DF route still resolved');
   if (routePrompt('implement feature')?.id !== 'Team') throw new Error('selftest failed: implementation prompt did not default to Team');
+  const broadMadTeamGoalPrompt = 'sks --mad tmux multi pane scout reasoning commit push $team $goal';
+  if (routePrompt(broadMadTeamGoalPrompt)?.id !== 'Team') throw new Error('selftest failed: broad MAD/Team/Goal tmux request was misrouted away from Team');
   if (routePrompt('$SKS implement feature')?.id !== 'Team') throw new Error('selftest failed: $SKS implementation prompt did not promote to Team');
   if (routePrompt('$From-Chat-IMG 채팅 기록 이미지와 첨부 이미지로 고객사 요청 수정 작업 수행해줘')?.id !== 'Team') throw new Error('selftest failed: explicit chat capture client work did not promote to Team');
   if (routePrompt('$Computer-Use localhost 화면 빠르게 검증해줘')?.id !== 'ComputerUse') throw new Error('selftest failed: $Computer-Use did not route to ComputerUse fast lane');