@lazyingart/agintiflow 0.20.181 → 0.20.183

package/README.md

@@ -153,7 +153,7 @@ aginti --language de
 | Save provider keys | `aginti auth`, `/auth`, `/login` |
 | Review current repo | `/review [focus]` |
 | Toggle SCS quality gate | `/scs` |
-| Use SCS only for complex work | `/scs auto` or `aginti --scs auto "task"` |
+| Keep default auto SCS routing | `/scs auto` or `aginti --scs auto "task"` |
 | Disable SCS for simple work | `/scs off` or `aginti --no-scs "task"` |
 | Control dynamic step budgets | `--dynamic-steps auto\|on\|off` |
 | Work with AAPS workflows | `aginti aaps status`, `/aaps validate` |

package/docs/runtime-modes-and-autonomy.md

@@ -138,6 +138,22 @@ aginti "start a tmux session named demo, run ls in it, keep it open, and tell me
 
 The agent should use `tmux_start_session`, `tmux_send_keys`, and `tmux_capture_pane`, not Docker `run_command`. In Docker sandbox mode, tmux commands must stay inside the project. In host mode, tmux command text is still governed by host shell policy; if a tmux command is blocked, present the suggested rerun path instead of trying tmux as a workaround. For trusted whole-host tmux work, use the host recipe above.
 
+Long download or long command:
+
+```bash
+aginti "download this large zip with wget -c, verify the final byte count, and do not burn model steps polling it"
+```
+
+The agent should use `start_long_job` once instead of repeatedly calling `wait`, `run_command`, or `tmux_capture_pane`. The tool creates `.aginti/long-jobs/<job-id>/status.json`, stdout/stderr logs, a supervisor log, and an optional status card. For resumable downloads, the agent should first determine `Content-Length` when practical, then call `start_long_job` with:
+
+- `command`: a resumable transfer such as `wget -c URL -O file.zip`
+- `expectedOutputPath`: the downloaded file
+- `expectedSizeBytes`: the expected byte count
+- `verifyCommand`: a deterministic check such as `unzip -t file.zip` or `sha256sum -c`
+- `restartOnFailure`: `true` when the transfer command is safe to resume
+
+After `start_long_job` succeeds, the model loop should finish with the job id and status path. Later status checks should use `long_job_status` or direct shell inspection of the status JSON. This keeps multi-hour I/O under a shell supervisor instead of consuming model tokens and step budget.
+
 ## Future Persistent Container Mode
 
 A useful next runtime mode is a service container:

package/docs/student-committee-supervisor.md

@@ -61,13 +61,13 @@ The current implementation is deliberately bounded:
 
 ## Auto Mode
 
-`/scs auto` and `--scs auto` are still available when a user wants cheaper routing for simple turns. Auto mode activates SCS for complex, risky, or long-running work. Signals include:
+`/scs auto` and `--scs auto` are the default for users who want cheap simple turns without losing strict validation on risky work. Auto mode activates SCS for high-risk, evidence-heavy, or long-running work. Signals include:
 
-- high smart-routing complexity score;
-- profiles such as code, app, Android/iOS, large-codebase, GitHub, maintenance, security, LaTeX, research, and supervision;
-- prompts mentioning multi-file work, regressions, builds, deployments, tmux, emulators, Docker, PDFs, or similar high-friction workflows.
+- very high smart-routing complexity score;
+- profiles such as app, Android/iOS, large-codebase, GitHub, maintenance, QA, security, supervision, pipeline, and website;
+- prompts mentioning failing tests, repo-wide changes, browser uploads, releases, deployments, tmux, emulators, Docker, PDFs, or similar high-friction workflows.
 
-Auto mode stays off for simple turns, but it is not the default.
+Auto mode is the default. It stays off for simple turns, uses the main model for moderate complexity, and activates SCS only for high-risk or evidence-heavy turns.
 
 ## Dynamic Step Budgets
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lazyingart/agintiflow",
-  "version": "0.20.181",
+  "version": "0.20.183",
   "type": "module",
   "description": "AgInTiFlow is a project-aware agent workspace for hybrid wet-dry R&D, hardware-aware intelligence, software automation, and industrial workflows.",
   "license": "Apache-2.0",
@@ -63,6 +63,7 @@
     "scripts/smoke-capabilities.js",
     "scripts/smoke-auto-update.js",
     "scripts/smoke-inbox.js",
+    "scripts/smoke-long-jobs.js",
     "scripts/smoke-mcp.js",
     "scripts/fixtures/mcp-stdio-smoke-server.mjs",
     "scripts/smoke-model-roles.js",
@@ -106,6 +107,7 @@
     "smoke:skillmesh": "node scripts/smoke-skillmesh.js",
     "smoke:toolchain-docker": "node scripts/smoke-toolchain-docker.js",
     "smoke:inbox": "node scripts/smoke-inbox.js",
+    "smoke:long-jobs": "node scripts/smoke-long-jobs.js",
     "smoke:mcp": "node scripts/smoke-mcp.js",
     "smoke:model-roles": "node scripts/smoke-model-roles.js",
     "smoke:platform": "node scripts/smoke-platform.js",
@@ -125,7 +127,7 @@
     "storage:migrate": "node bin/aginti-cli.js storage migrate",
     "publish:env": "node scripts/npm-publish-from-env.js publish --access public",
     "publish:env:whoami": "node scripts/npm-publish-from-env.js whoami",
-    "test": "npm run check && npm run smoke:runtime-compat && npm run smoke:autoupdate && npm run smoke:web-api && npm run smoke:web-ui && npm run smoke:web-autostart && npm run smoke:webapp-command && npm run smoke:web-port-fallback && npm run smoke:docker-command && npm run smoke:coding-tools && npm run smoke:dynamic-step-budget && npm run smoke:aaps-adapter && npm run smoke:auxiliary-tools && npm run smoke:perception-research && npm run smoke:auth && npm run smoke:canvas-artifacts && npm run smoke:capabilities && npm run smoke:mcp && npm run smoke:model-roles && npm run smoke:platform && npm run smoke:permission-modes && npm run smoke:skills && npm run smoke:skillmesh && npm run smoke:tmux-tools && npm run smoke:cli-chat && npm run smoke:inbox",
+    "test": "npm run check && npm run smoke:runtime-compat && npm run smoke:autoupdate && npm run smoke:web-api && npm run smoke:web-ui && npm run smoke:web-autostart && npm run smoke:webapp-command && npm run smoke:web-port-fallback && npm run smoke:docker-command && npm run smoke:coding-tools && npm run smoke:dynamic-step-budget && npm run smoke:aaps-adapter && npm run smoke:auxiliary-tools && npm run smoke:perception-research && npm run smoke:auth && npm run smoke:canvas-artifacts && npm run smoke:capabilities && npm run smoke:mcp && npm run smoke:model-roles && npm run smoke:platform && npm run smoke:permission-modes && npm run smoke:skills && npm run smoke:skillmesh && npm run smoke:tmux-tools && npm run smoke:long-jobs && npm run smoke:cli-chat && npm run smoke:inbox",
     "pack:dry-run": "npm pack --dry-run",
     "smoke:capabilities": "node scripts/smoke-capabilities.js"
   },

package/public/app.js

@@ -1402,7 +1402,7 @@ function syncQuickModeControls() {
   if (aapsModeToggle) aapsModeToggle.checked = (taskProfileField?.value || "auto") === "aaps";
   if (veniceModeToggle) veniceModeToggle.checked = quickVeniceModeActive;
   if (quickModeStatusEl) {
-    const scs = enableScsField?.value || "on";
+    const scs = enableScsField?.value || "auto";
     const profile = taskProfileField?.value || "auto";
     const route = `${routeProviderField?.value || "deepseek"}/${fieldValue(routeModelField) || "auto"}`;
     const main = `${mainProviderField?.value || "deepseek"}/${fieldValue(mainModelField) || "auto"}`;
@@ -1451,8 +1451,8 @@ function applyVeniceMode(enabled) {
   syncQuickModeControls();
 }
 
-function applyScsMode(mode = enableScsField?.value || "on") {
-  if (enableScsField) enableScsField.value = ["on", "auto", "off"].includes(mode) ? mode : "on";
+function applyScsMode(mode = enableScsField?.value || "auto") {
+  if (enableScsField) enableScsField.value = ["on", "auto", "off"].includes(mode) ? mode : "auto";
   if (allowParallelScoutsField && enableScsField?.value === "on") {
     allowParallelScoutsField.checked = false;
   }
@@ -1707,7 +1707,7 @@ function formPayload() {
     allowWebSearch: allowWebSearchField?.checked ?? true,
     allowMcpTools: allowMcpToolsField?.checked ?? true,
     allowParallelScouts: allowParallelScoutsField?.checked ?? true,
-    enableScs: enableScsField?.value || "on",
+    enableScs: enableScsField?.value || "auto",
     dynamicSteps: dynamicStepsField?.value || "auto",
     veniceMode: quickVeniceModeActive,
     parallelScoutCount: Math.min(Math.max(Number(parallelScoutCountField?.value) || 3, 1), 10),
@@ -3818,7 +3818,7 @@ async function loadConfig() {
   if (allowWebSearchField) allowWebSearchField.checked = prefs.allowWebSearch ?? true;
   if (allowMcpToolsField) allowMcpToolsField.checked = prefs.allowMcpTools ?? true;
   if (allowParallelScoutsField) allowParallelScoutsField.checked = prefs.allowParallelScouts ?? true;
-  if (enableScsField) enableScsField.value = prefs.enableScs || "on";
+  if (enableScsField) enableScsField.value = prefs.enableScs || "auto";
   if (dynamicStepsField) dynamicStepsField.value = prefs.dynamicSteps || "auto";
   if (parallelScoutCountField) parallelScoutCountField.value = String(prefs.parallelScoutCount || 3);
   allowWrapperToolsField.checked = prefs.allowWrapperTools ?? false;

package/public/index.html

@@ -264,8 +264,8 @@
             <label class="quick-mode-select">
               <span data-i18n="scsModeLabel">SCS</span>
               <select id="enableScs" name="enableScs">
-                <option value="on" data-i18n="scsOnOption">On</option>
                 <option value="auto" data-i18n="scsAutoOption">Auto</option>
+                <option value="on" data-i18n="scsOnOption">On</option>
                 <option value="off" data-i18n="scsOffOption">Off</option>
               </select>
             </label>

package/references/aginti-generalization-audit-2026-05-20.md

@@ -60,7 +60,8 @@ npm run smoke:model-roles
 node scripts/smoke-skills.js
 node scripts/smoke-dynamic-step-budget.js
 node scripts/smoke-coding-tools.js
-rg -n --hidden --glob '!node_modules/**' --glob '!.git/**' --glob '!dist/**' --glob '!coverage/**' 'xiaoyunque|小云雀|xyq|jianying|LALACHAN|啦啦|拉拉|阿芽|飒飒|Seedance|沉浸式短片|Trio\.png|display\.png|R1\.jpg|patchwork-leather|duanpian' .
+rg -n --hidden --glob '!node_modules/**' --glob '!.git/**' --glob '!dist/**' --glob '!coverage/**' '<private-project-term-pattern>' .
 ```
 
-Result: checks passed, and the repository search found no remaining private workflow terms in AgInTiFlow runtime code, tests, or general references after excluding this audit's own search expression.
+Result: checks passed, and the repository search found no remaining private
+workflow terms in AgInTiFlow runtime code, tests, or general references.

package/references/agintiflow-npm-publication-memory.md

@@ -6,6 +6,18 @@ This note records the working publication route for AgInTiFlow so future release
 
 ## Published Evidence
 
+Latest verified release:
+
+- Version: `0.20.182`
+- Commit: `b5492f7`
+- GitHub Actions run: `https://github.com/lazyingart/AgInTiFlow/actions/runs/26198123298`
+- Workflow result: success
+- npm registry check: `npm view @lazyingart/agintiflow version` returned `0.20.182`
+- Installed verification: `npm install -g @lazyingart/agintiflow@0.20.182` then `aginti --version` returned `0.20.182`
+- Webapp verification: `aginti webapp restart --port 3210` then `curl -fsS http://127.0.0.1:3210/health` returned `version":"0.20.182"` from the global npm package path.
+
+Previous reference release:
+
 - Package: `@lazyingart/agintiflow`
 - Version: `0.20.169`
 - Commit: `6ef3768`

package/references/bilingual-book-supervision-lessons.md

@@ -1,7 +1,7 @@
 # Bilingual Book Supervision Lessons
 
-This note records lessons from supervising AgInTiFlow on the ZhJpBook bilingual
-interlinear book pipeline. The goal is not to make AgInTiFlow a book-specific
+This note records lessons from supervising AgInTiFlow on a multilingual
+annotated book pipeline. The goal is not to make AgInTiFlow a book-specific
 tool. The goal is to identify which failures were project workflow issues and
 which reusable capabilities AgInTiFlow needs in order to complete large artifact
 projects from raw inputs to final outputs with less external guidance.
@@ -19,25 +19,25 @@ Observed AgInTiFlow primitives:
 - `docs/self-healing-pipelines.md`
 - `docs/auxiliary-image-generation.md`
 
-Observed ZhJpBook workflow pieces:
+Observed project-local workflow pieces:
 
-- `scripts/interlinear/aginti_write_chunks.py`
-- `scripts/interlinear/aginti_dynamic_review_chunks.py`
-- `scripts/interlinear/start_aginti_parallel_json_writers.sh`
-- `scripts/interlinear/watch_aginti_compile.py`
-- `scripts/interlinear/compile_prepared_book_both_previews.sh`
+- source ingestion and cleanup scripts
+- chunk writer runners
+- dynamic review scripts
+- parallel structured-output workers
+- compile and progress-watch scripts
 
-Observed Sishu run evidence:
+Observed large-run evidence:
 
-- The book reached `2480/2480` manifest chunks valid with no missing chunks.
-- The final compile produced four PDFs: ZH-main and JP-main, color and
-  blackwhite.
-- Page counts were `2387` for ZH-main and `2065` for JP-main.
+- The book reached full manifest coverage with no missing chunks.
+- The final compile produced multiple language, direction, and style variants.
+- Page counts differed by output variant and had to be verified against the
+  current manifest.
 - A stale compile-watch status still reported older failed counts after the
   canonical progress report was clean, showing that multiple status snapshots
   can diverge unless one source of truth is enforced.
 - AgInTi image generation was useful for cover backgrounds, but exact title,
-  author, furigana, and curation text still needed deterministic composition.
+  author, annotations, and curation text still needed deterministic composition.
 
 ## Main Finding
 
@@ -66,11 +66,12 @@ artifact-pipeline lifecycle that AgInTi can instantiate and supervise:
 These belong in the target repository or generated project scripts, not in
 AgInTiFlow core:
 
-- Book-specific source schemas and bilingual chunk schemas.
-- Chinese/Japanese rendering decisions, TeX macros, page size, grammar colors,
-  ruby placement, table of contents, and cover typography.
-- Exact validators for source preservation, Hanzi tokenization, furigana/pinyin
-  policy, grammar roles, and line-based interlinear layout.
+- Book-specific source schemas and annotated chunk schemas.
+- Language-pair rendering decisions, TeX or renderer macros, page size,
+  grammar colors, reading placement, table of contents, and cover typography.
+- Exact validators for source preservation, script-specific tokenization,
+  reading or romanization policy, grammar roles, and line-based interlinear
+  layout.
 - Book plans, source bundles, source-specific OCR cleanup rules, and known
   editions/translations.
 - The concrete writer prompt for one literary/classical task.
@@ -132,8 +133,8 @@ Recommended core behavior:
 ### 2. Restarting From Zero Destroyed Efficiency
 
 Large book work is valuable even when partial. The most expensive artifacts were
-translation/alignment/ruby/grammar chunks. Retuning chunk size or prompt
-instructions should not discard reviewed chunks.
+translation, alignment, reading, and grammar chunks. Retuning chunk size or
+prompt instructions should not discard reviewed chunks.
 
 Recommended core behavior:
 
@@ -164,7 +165,7 @@ Recommended core behavior:
 
 The deterministic reviewer fixed schema/render issues. The user also wanted
 review that notices OCR corruption, missing text, bad line alignment, all-one
-color grammar pages, kana-only Japanese, source drift, and repeated filler.
+color grammar pages, wrong-script output, source drift, and repeated filler.
 
 Recommended core behavior:
 
@@ -177,8 +178,8 @@ Recommended core behavior:
 
 ### 5. OCR and Source Ingestion Were Underestimated
 
-For scanned Chinese books, bad OCR flowed into later high-cost annotation. The
-right place to catch this is before bilingual generation.
+For scanned or OCR-heavy books, bad OCR flowed into later high-cost annotation.
+The right place to catch this is before multilingual generation.
 
 Recommended core behavior:
 
@@ -201,7 +202,7 @@ Recommended core behavior:
 
 - Cover/poster workflows should default to "image model creates background;
   deterministic renderer adds exact text" when exact titles, author names, URLs,
-  ruby, or branding are required.
+  reading marks, annotations, or branding are required.
 - The image-generation tool should save raw prompt/result metadata, but project
   scripts should own final typography.
 

package/references/codex-session-cross-references.md

@@ -65,11 +65,11 @@ Project-local skill produced from this direction:
 
 - `.aginti/skills/<project-browser-workflow>/SKILL.md` in the task repository
 
-### ZhJpBook And AgInTiFlow Implementation Session
+### Multilingual Book Pipeline And AgInTiFlow Implementation Session
 
 - Session: `019e1f99-289e-7711-986a-d41047f5ed21`
-- Status context shown by Codex: `~/ProjectsLFS/ZhJpBook`
-- Primary task repo involved: `/home/lachlan/ProjectsLFS/ZhJpBook`
+- Status context shown by Codex: a project-local multilingual book repository
+- Primary task repo involved: project-local, outside AgInTiFlow core
 - AgInTiFlow implementation repo: `/home/lachlan/ProjectsLFS/Agent/AgInTiFlow`
 - Role: implementation and validation session.
 
@@ -129,9 +129,9 @@ These sessions should be read together:
   owner.
 - The browser workflow session supplies the concrete failure mode: browser automation and media tasks
   can look successful while the visible external state is wrong or unverified.
-- ZhJpBook supplies the long-running pipeline pressure: the agent must keep
-  moving through scripts, monitors, validators, and artifacts without accepting
-  unsupported finish claims.
+- The multilingual book pipeline supplies the long-running pipeline pressure:
+  the agent must keep moving through scripts, monitors, validators, and
+  artifacts without accepting unsupported finish claims.
 - AAPS supplies the project/workflow/block/program front-end contract where
   AgInTiFlow may act as a backend adapter but must not mutate AAPS scope or
   semantics unexpectedly.

package/references/dynamic-step-budget-and-scs-auto.md

@@ -1,6 +1,6 @@
 # Dynamic Step Budget And SCS Auto Routing
 
-This note documents the implemented design for dynamic `maxSteps` extension and SCS routing. SCS now defaults to `on`; auto routing remains available with `--scs auto` or `/scs auto` when a user wants SCS only for complex turns.
+This note documents the implemented design for dynamic `maxSteps` extension and SCS routing. SCS defaults to `auto`: simple turns use the fast route, moderate work can use the main model without heavy validation, and high-risk evidence-bearing work activates the Student-Committee-Supervisor gates.
 
 ## Current State
 
@@ -15,11 +15,11 @@ Relevant paths:
 - `src/step-budget-controller.js`: owns deterministic progress/blocker checks, caps, serialization, and extension application.
 - `src/scs-controller.js`: supports `off`, `on`, `auto`, and the SCS student step-budget gate.
 
-SCS auto is supported but is not the default:
+SCS auto is the default:
 
 - CLI: `--scs auto`.
 - Interactive: `/scs auto`.
-- Config: `enableScs: "on"` by default, or `"auto"` when explicitly requested.
+- Config: `enableScs: "auto"` by default, or `"on"`/`"off"` when explicitly requested.
 - Router policy: `shouldActivateScs("auto", { goal, taskProfile, complexityScore })`.
 - Active model policy: when SCS is active, committee/student/supervisor use the main model.
 
@@ -86,11 +86,11 @@ SCS mode allows a slightly larger budget because the student gate is stricter:
 
 ### SCS Auto Mode
 
-`/scs auto` means the router may activate SCS for a specific turn. Signals include:
+`/scs auto` means the router may activate SCS for a specific turn. It deliberately has a higher bar than smart model routing, so a turn can use the main model without paying SCS overhead. Signals include:
 
-- high smart-routing complexity score;
-- high-friction profiles such as Android, app, code, large-codebase, GitHub, maintenance, QA, research, review, security, supervision, or website;
-- prompt language such as debug, failing, migration, emulator, deploy, PDF, refactor, monitor, or long-running.
+- very high smart-routing complexity score;
+- high-friction profiles such as Android, app, large-codebase, GitHub, maintenance, QA, security, supervision, pipeline, or website;
+- prompt language such as failing tests, fix the build, migration, deploy, browser automation, upload, PDF compilation, publish, monitor, or long-running.
 
 If SCS auto activates, dynamic step extension uses the SCS student gate. If SCS auto does not activate, dynamic step extension uses the normal deterministic gate.
 
@@ -212,7 +212,7 @@ Defaults:
 Routing semantics:
 
 - `/scs off`: never use SCS; dynamic budget uses normal deterministic monitoring if enabled.
-- `/scs on`: always use SCS; this is the default, and dynamic budget uses the student budget gate.
+- `/scs on`: always use SCS, and dynamic budget uses the student budget gate.
 - `/scs auto`: router decides per run; if activated, dynamic budget uses the student gate.
 
 For observability, state preserves:

package/references/student-committee-supervisor-mode.md

@@ -187,7 +187,7 @@ Recommended modes:
 | --- | --- |
 | `off` | Current behavior. |
 | `on` | SCS gates all non-trivial work in the session. |
-| `auto` | SCS only activates for complex tasks, write actions, multi-step shell work, GitHub, system maintenance, Android/iOS, LaTeX/PDF, and long-running tasks. |
+| `auto` | Default. Simple turns stay linear, moderate tasks may use the main model without SCS, and SCS activates for high-risk/evidence-heavy work such as failing builds, repo-wide edits, GitHub/release work, browser uploads, system maintenance, Android/iOS, LaTeX/PDF compilation, and long-running supervision. |
 
 `/scs` should toggle SCS on/off in interactive CLI. `/scs auto` keeps the mature complex-task gate for users who want quality control without paying the SCS cost on simple turns.
 

package/scripts/smoke-cli-chat.js

@@ -795,12 +795,12 @@ try {
     throw new Error("interactive /review did not launch the bounded review workflow");
   }
   const scsDefaultStatusResult = await runChat("/scs status\n");
-  if (!scsDefaultStatusResult.stdout.includes("SCS mode: on")) {
-    throw new Error("interactive /scs status did not show default on mode");
+  if (!scsDefaultStatusResult.stdout.includes("SCS mode: auto")) {
+    throw new Error("interactive /scs status did not show default auto mode");
   }
   const statusFooterResult = await runChat("/status\n");
   if (
-    !statusFooterResult.stdout.includes(`cwd=${tempRoot} scs=on aaps=off venice=off`) ||
+    !statusFooterResult.stdout.includes(`cwd=${tempRoot} scs=auto aaps=off venice=off`) ||
     !statusFooterResult.stdout.includes("permission=")
   ) {
     throw new Error("interactive /status did not include scs/aaps/venice mode state on the cwd line");

package/scripts/smoke-dynamic-step-budget.js

@@ -218,7 +218,8 @@ try {
       sessionId: "dynamic-step-budget-smoke",
     }
   );
-  assert(config.enableScs === "on", "runtime config should default SCS mode to on");
+  assert(config.enableScs === "auto", "runtime config should default SCS mode to auto");
+  assert(config.scsActive === false, "simple mock write should not activate SCS in auto mode");
   const run = await runAgent(config);
   assert(!run.stopped, "mock run stopped instead of using dynamic extension");
   const written = await fs.readFile(path.join(workspace, "notes/dynamic-budget.md"), "utf8");

package/scripts/smoke-long-jobs.js

@@ -0,0 +1,104 @@
+#!/usr/bin/env node
+import assert from "node:assert/strict";
+import fs from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
+import { checkToolUse } from "../src/guardrails.js";
+import { longJobStatus, startLongJob } from "../src/long-job-tools.js";
+import { tmuxAvailable } from "../src/tmux-tools.js";
+
+const workspace = await fs.mkdtemp(path.join(os.tmpdir(), "agintiflow-long-job-"));
+const config = {
+  allowShellTool: true,
+  allowDestructive: true,
+  allowPasswords: true,
+  commandCwd: workspace,
+  sandboxMode: "host",
+  useDockerSandbox: false,
+};
+const dockerConfig = {
+  ...config,
+  allowDestructive: false,
+  allowPasswords: false,
+  useDockerSandbox: true,
+  sandboxMode: "docker-workspace",
+  packageInstallPolicy: "allow",
+};
+
+function sleep(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+
+try {
+  if (!(await tmuxAvailable())) {
+    console.log(JSON.stringify({ ok: true, skipped: true, reason: "tmux is not installed" }, null, 2));
+    process.exit(0);
+  }
+
+  const noShell = checkToolUse({
+    toolName: "start_long_job",
+    args: { command: "printf ok > out.txt" },
+    config: { ...config, allowShellTool: false },
+  });
+  assert.equal(noShell.allowed, false, "start_long_job should require shell capability");
+
+  const outsidePath = path.join(os.tmpdir(), "aginti-long-job-outside.txt");
+  const outsideGuard = checkToolUse({
+    toolName: "start_long_job",
+    args: { name: "outside", command: `cat ${outsidePath}` },
+    config: dockerConfig,
+  });
+  assert.equal(outsideGuard.allowed, false, "Docker-mode long jobs should block outside host absolute paths");
+
+  const start = await startLongJob(
+    {
+      name: "smoke-download-style",
+      command: "printf 'abc123' > download.bin",
+      expectedOutputPath: "download.bin",
+      expectedSizeBytes: 6,
+      verifyCommand: "grep -q abc download.bin",
+      pollIntervalSeconds: 5,
+      note: "Smoke test for durable long command handoff.",
+    },
+    config
+  );
+  assert.equal(start.ok, true, start.error || start.reason);
+  assert.equal(start.background, true, "long job should return as a background handoff");
+  assert.match(start.statusPath, /^\.aginti\/long-jobs\//, "status path should be project-local");
+
+  let status = null;
+  for (let i = 0; i < 30; i += 1) {
+    await sleep(250);
+    status = await longJobStatus({ jobId: start.jobId }, config);
+    assert.equal(status.ok, true, status.reason);
+    if (["completed", "failed"].includes(status.state)) break;
+  }
+  assert.equal(status?.state, "completed", `long job did not complete: ${JSON.stringify(status)}`);
+  assert.equal(status.outputBytes, 6, "expected-size progress was not recorded");
+  assert.equal(status.verifyExitCode, 0, "verify command did not pass");
+
+  const output = await fs.readFile(path.join(workspace, "download.bin"), "utf8");
+  assert.equal(output, "abc123", "job output content mismatch");
+  await fs.access(path.join(workspace, start.statusPath));
+  await fs.access(path.join(workspace, start.stdoutPath));
+  await fs.access(path.join(workspace, start.stderrPath));
+  await fs.access(path.join(workspace, start.supervisorLogPath));
+
+  console.log(
+    JSON.stringify(
+      {
+        ok: true,
+        workspace,
+        jobId: start.jobId,
+        statusPath: start.statusPath,
+        state: status.state,
+        outputBytes: status.outputBytes,
+        verifyExitCode: status.verifyExitCode,
+      },
+      null,
+      2
+    )
+  );
+} finally {
+  await fs.rm(workspace, { recursive: true, force: true }).catch(() => {});
+}

package/scripts/smoke-model-roles.js

@@ -16,9 +16,11 @@ import {
   buildSupervisorInstruction,
   deterministicPlanActionContradiction,
   reviewScsFinish,
+  shouldActivateScs,
   shouldRequestScsReplan,
 } from "../src/scs-controller.js";
 import { buildScsEvidenceLedger, deriveScsTaskContract, evaluateScsEvidence } from "../src/scs-evidence.js";
+import { resolveRuntimeConfig } from "../src/config.js";
 
 const repoRoot = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
 
@@ -116,6 +118,86 @@ const fastRoute = selectModelRoute({
 });
 assert(fastRoute.model === "deepseek-v4-flash", "fast route did not use route model override");
 
+const simpleSmartRoute = selectModelRoute({
+  routingMode: "smart",
+  provider: "deepseek",
+  goal: "say hello",
+  taskProfile: "auto",
+});
+assert(simpleSmartRoute.model === "deepseek-v4-flash", "simple smart route should use the route model");
+assert(
+  !shouldActivateScs("auto", {
+    goal: "say hello",
+    taskProfile: "auto",
+    complexityScore: simpleSmartRoute.complexityScore,
+  }),
+  "SCS auto should stay inactive for simple turns"
+);
+
+const moderateSmartRoute = selectModelRoute({
+  routingMode: "smart",
+  provider: "deepseek",
+  goal: "implement a focused refactor design for one module",
+  taskProfile: "auto",
+});
+assert(moderateSmartRoute.model === "deepseek-v4-pro", "moderate smart route should use the main model");
+assert(
+  !shouldActivateScs("auto", {
+    goal: "implement a focused refactor design for one module",
+    taskProfile: "auto",
+    complexityScore: moderateSmartRoute.complexityScore,
+  }),
+  "SCS auto should not gate every main-model turn"
+);
+
+const highRiskSmartRoute = selectModelRoute({
+  routingMode: "smart",
+  provider: "deepseek",
+  goal: "debug failing tests and fix the build in a large repo, then commit and push",
+  taskProfile: "auto",
+});
+assert(highRiskSmartRoute.model === "deepseek-v4-pro", "high-risk smart route should use the main model");
+assert(
+  shouldActivateScs("auto", {
+    goal: "debug failing tests and fix the build in a large repo, then commit and push",
+    taskProfile: "auto",
+    complexityScore: highRiskSmartRoute.complexityScore,
+  }),
+  "SCS auto should activate for high-risk evidence-bearing work"
+);
+assert(
+  !shouldActivateScs("auto", { goal: "explain this function", taskProfile: "code", complexityScore: 0 }),
+  "code profile alone should not force SCS auto"
+);
+
+const simpleRuntimeConfig = resolveRuntimeConfig({
+  goal: "say hello",
+  provider: "deepseek",
+  routingMode: "smart",
+  taskProfile: "auto",
+});
+assert(simpleRuntimeConfig.enableScs === "auto", "runtime config should default SCS mode to auto");
+assert(simpleRuntimeConfig.scsActive === false, "simple runtime config should not activate SCS");
+assert(simpleRuntimeConfig.model === "deepseek-v4-flash", "simple runtime config should use route model");
+
+const moderateRuntimeConfig = resolveRuntimeConfig({
+  goal: "implement a focused refactor design for one module",
+  provider: "deepseek",
+  routingMode: "smart",
+  taskProfile: "auto",
+});
+assert(moderateRuntimeConfig.scsActive === false, "moderate main-model work should not automatically activate SCS");
+assert(moderateRuntimeConfig.model === "deepseek-v4-pro", "moderate runtime config should use main model");
+
+const highRiskRuntimeConfig = resolveRuntimeConfig({
+  goal: "debug failing tests and fix the build in a large repo, then commit and push",
+  provider: "deepseek",
+  routingMode: "smart",
+  taskProfile: "auto",
+});
+assert(highRiskRuntimeConfig.scsActive === true, "high-risk runtime config should activate SCS");
+assert(highRiskRuntimeConfig.model === "deepseek-v4-pro", "SCS runtime config should use main model");
+
 assert(MODEL_PROVIDER_GROUPS["venice-gpt"].provider === "venice", "venice-gpt group missing");
 assert(modelsForProviderGroup("venice").some((item) => item.id === "venice-uncensored-1-2"), "venice group missing Venice 1.2");
 assert(modelsForProviderGroup("venice-gemma").some((item) => item.id === "google-gemma-4-31b-it"), "venice-gemma bucket missing Gemma 4 instruct");

package/scripts/smoke-web-api.js

@@ -117,7 +117,7 @@ try {
   if (config.preferences?.packageInstallPolicy !== "allow") throw new Error("web did not default to Docker package installs");
   if (config.preferences?.permissionMode !== "normal") throw new Error("web did not default to normal permission mode");
   if (config.preferences?.workspaceWritePolicy !== "allow") throw new Error("web did not default to workspace writes allowed");
-  if (config.preferences?.enableScs !== "on") throw new Error("web did not default to CLI-aligned SCS on mode");
+  if (config.preferences?.enableScs !== "auto") throw new Error("web did not default to CLI-aligned SCS auto mode");
   if (config.preferences?.dynamicSteps !== "auto") throw new Error("web did not default to CLI-aligned dynamic steps auto mode");
   if (config.preferences?.veniceMode !== false) throw new Error("web should default Venice shortcut mode off");
   if (Number(config.preferences?.maxSteps) < 24) throw new Error("web default max steps is too low");

package/skills/bilingual-interlinear-book/SKILL.md

@@ -9,11 +9,11 @@ triggers:
   - paired language
   - annotated book
   - parallel text
-  - ruby
-  - furigana
-  - pinyin
-  - xelatex
-  - pocket book
+  - aligned annotations
+  - glossed text
+  - reading annotations
+  - translation notes
+  - source commentary
 tools:
   - read_file
   - write_file
@@ -46,7 +46,7 @@ This is a general workflow skill. It must not hard-code a specific book, languag
 Keep annotation policies project-defined:
 
 - Token shape, reading placement, grammar tags, and alignment granularity must come from the project schema.
-- For scripts that need per-character readings, validators should enforce that locally instead of relying on a prompt.
+- For scripts that need per-token or per-character readings, validators should enforce that locally instead of relying on a prompt.
 - Placeholder translations, empty commentary, duplicated readings, copied modern paraphrases, or source drift should be detected by validators or reviewers.
 - Translation/reference quality gates must be derived from the project schema, language profile, and source audit. Do not bake one book's vocabulary, source names, or forbidden-token list into this built-in skill or AgInTiFlow core.
 - If a reference is unreliable for the requested target language, do not feed it to the writer as trusted translation. Either omit it, label it clearly as alignment/source-only context, or ask the project to supply a reliable reference.