@lazyingart/agintiflow 0.3.0 → 0.4.0

package/README.md

@@ -38,8 +38,12 @@ Install the published CLI:
 
 ```bash
 npm install -g @lazyingart/agintiflow
+cd /path/to/your-project
+aginti init
+aginti doctor
 aginti --list-routes
-aginti --sandbox-status --sandbox-mode docker-readonly --cwd "$PWD"
+aginti --list-profiles
+aginti --sandbox-status --sandbox-mode docker-readonly
 ```
 
 Launch the local web UI from an installed package:
@@ -49,10 +53,23 @@ aginti web --port 3210
 # then open http://127.0.0.1:3210
 ```
 
+`aginti web` uses the folder it is launched from as the project root, default working directory, session store, and settings database. CLI and web runs share the same project-local `.sessions/` folder.
+
 Run the installed CLI without a live provider key by using the local mock route:
 
 ```bash
-aginti --provider mock --routing manual --allow-shell --cwd "$PWD" "Report the current directory"
+aginti --provider mock --routing manual --allow-file-tools "Create notes/hello.md with a smoke-test note"
+```
+
+Useful project commands:
+
+```bash
+aginti keys status
+printf '%s' "$DEEPSEEK_API_KEY" | aginti keys set deepseek --stdin
+aginti sessions list
+aginti sessions show <session-id>
+aginti resume <session-id> "continue with a short follow-up"
+aginti --profile code --provider mock --routing manual "Create notes/hello.md"
 ```
 
 Run from a source checkout:
@@ -101,6 +118,9 @@ The package exposes both `aginti` and `aginti-cli`; they run the same CLI entryp
 The web app includes:
 
 - Routing dropdown for smart, fast, complex, and manual model selection.
+- Project folder indicator showing project root, command cwd, session folder, and session database.
+- First-run provider setup panel with mock fallback and project-local DeepSeek/OpenAI key save.
+- Task profile dropdown for code, writing, design docs, Python, shell, Node, AAPS, LaTeX, and system maintenance workflows.
 - Provider dropdown for DeepSeek, OpenAI, and local mock mode when manual routing is needed.
 - Language dropdown with 11 persisted UI locales.
 - Editable model field, with DeepSeek v4 flash as the fast default and DeepSeek v4 pro as the complex route.
@@ -162,6 +182,7 @@ PACKAGE_INSTALL_POLICY=prompt
 USE_DOCKER_SANDBOX=true
 DOCKER_SANDBOX_IMAGE=agintiflow-sandbox:latest
 COMMAND_CWD=/home/lachlan/ProjectsLFS/Agent
+AGINTI_TASK_PROFILE=auto
 ```
 
 Defaults:
@@ -180,6 +201,15 @@ Provider credentials:
 | OpenAI | `OPENAI_API_KEY` | `https://api.openai.com/v1` |
 | DeepSeek | `DEEPSEEK_API_KEY` | `https://api.deepseek.com/v1` |
 
+Project-local credentials can be stored without committing secrets:
+
+```bash
+aginti init
+printf '%s' "$DEEPSEEK_API_KEY" | aginti keys set deepseek --stdin
+```
+
+This writes `.aginti/.env` with `0600` permissions and adds safe `.gitignore` entries. APIs and logs expose only key presence, never raw values.
+
 ## Agent Wrappers
 
 AgInTiFlow can expose external coding agents as advisory tools when `ALLOW_WRAPPER_TOOLS=true` or the web UI toggle is enabled. Wrappers are not a replacement for the core runner; they are used for second opinions, codebase analysis, or planning when they are installed and authenticated. The preferred wrapper defaults to Codex and can be changed with `PREFERRED_WRAPPER=codex`, `aginti --allow-wrappers --wrapper codex`, or the web UI dropdown.

package/docs/project-agent-lessons.md

@@ -0,0 +1,41 @@
+# Project-Agent UX Lessons
+
+Round 8 distilled implementation lessons from local agent projects and applied the highest-value subset to AgInTiFlow.
+
+## Codex
+
+- Treat the launch directory as the working project unless the user overrides it.
+- Keep patch/file tools deterministic and auditable: edits should be workspace-local, compactly diffed, and easy to inspect.
+- Resume should be a first-class CLI flow, not only a hidden runtime detail.
+
+## Claude Code
+
+- Terminal UX should be simple: install, `cd` into a project, run the agent.
+- Project context matters more than global state for everyday coding work.
+- Natural-language prompts should map to file edits, shell checks, and git-aware workflows without extra ceremony.
+
+## Gemini CLI
+
+- Session history should be project-scoped so runs from different folders do not mix.
+- Checkpoints/resume flows need discoverable commands and clear current-project boundaries.
+- Extensibility works best as named profiles/skills rather than hardcoded one-off prompts.
+
+## Copilot SDK
+
+- BYOK status should report only presence/absence and env-var names, never raw key material.
+- Session APIs should expose metadata, logs, workspace state, and model/provider choices for web clients.
+- Tool and permission hooks should be visible enough that users can understand why actions were allowed or blocked.
+
+## Claw Code
+
+- First-run `init` and `doctor` commands reduce setup ambiguity.
+- Container/sandbox readiness should be diagnosed before complex tasks.
+- Parity and smoke checks should run against isolated project folders, not only the source repo.
+
+## Applied In AgInTiFlow
+
+- `aginti web` now defaults command execution to the folder it was launched from.
+- CLI and web share `.sessions/` in the project root.
+- `aginti init`, `aginti doctor`, `aginti keys`, `aginti sessions`, and `aginti resume` provide basic project lifecycle control.
+- Task profiles wire lightweight skill prompts into CLI/web runs while keeping the LLM responsible for the main plan.
+- Project-local `.aginti/.env` can store provider keys safely with 0600 permissions and ignored git entries.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lazyingart/agintiflow",
-  "version": "0.3.0",
+  "version": "0.4.0",
   "type": "module",
   "description": "AgInTiFlow is a resumable Playwright website-control agent with OpenAI-compatible tool calling.",
   "license": "Apache-2.0",

package/public/app.js

@@ -19,6 +19,18 @@ const translations = {
     languageLabel: "Language",
     intro:
       "Web-first agent platform with smart model routing, resumable runs, guarded tools, and optional external agent wrappers.",
+    projectStatusTitle: "Project folder",
+    setupTitle: "Provider setup",
+    setupHelp:
+      "DeepSeek/OpenAI keys are missing. Use mock mode, export an env var, or save a project-local DeepSeek key.",
+    setupEnvHelp:
+      "Env vars: DEEPSEEK_API_KEY, OPENAI_API_KEY, or LLM_API_KEY. Mock mode remains available for local tests.",
+    setupProviderLabel: "Provider",
+    setupKeyLabel: "API key",
+    saveKeyButton: "Save local key",
+    keySavedStatus: "Local key saved. Raw values are never returned by the API.",
+    keySaveFailed: "Failed to save local key.",
+    taskProfileLabel: "Task profile",
     routingModeLabel: "Routing policy",
     routingSmartOption: "Smart: flash/pro/wrappers",
     routingFastOption: "DeepSeek v4 flash",
@@ -617,6 +629,13 @@ const providerField = document.querySelector("#provider");
 const modelField = document.querySelector("#model");
 const routingHintEl = document.querySelector("#routing-hint");
 const modelRouteStatusEl = document.querySelector("#model-route-status");
+const projectStatusEl = document.querySelector("#project-status");
+const setupCardEl = document.querySelector("#setup-card");
+const setupProviderField = document.querySelector("#setup-provider");
+const setupApiKeyField = document.querySelector("#setup-api-key");
+const saveApiKeyButton = document.querySelector("#save-api-key");
+const setupStatusEl = document.querySelector("#setup-status");
+const taskProfileField = document.querySelector("#taskProfile");
 const sandboxModeField = document.querySelector("#sandboxMode");
 const packageInstallPolicyField = document.querySelector("#packageInstallPolicy");
 const packageWarningEl = document.querySelector("#package-warning");
@@ -673,6 +692,8 @@ const defaults = {
 
 let currentLanguage = "en";
 let routingPresets = {};
+let taskProfiles = [];
+let projectInfo = null;
 let currentSessionId = "";
 let pollTimer = null;
 let saveTimer = null;
@@ -713,6 +734,28 @@ function renderKeyStatus(status = lastKeyStatus) {
   } · DeepSeek ${status.deepseek ? t("availableLabel") : t("missingLabel")} · ${t("mockLabel")} ${
     status.mock ? t("availableLabel") : t("missingLabel")
   }`;
+  if (setupCardEl) setupCardEl.hidden = Boolean(status.openai || status.deepseek);
+}
+
+function renderProjectStatus(info = projectInfo) {
+  projectInfo = info;
+  if (!projectStatusEl || !info) return;
+  projectStatusEl.textContent = [
+    `root=${info.root || ""}`,
+    `cwd=${info.commandCwd || ""}`,
+    `sessions=${info.sessionsDir || ""}`,
+    `db=${info.sessionDbPath || ""}`,
+    `shared=${info.sharedSessionFolder ? "yes" : "no"}`,
+  ].join(" · ");
+}
+
+function renderTaskProfiles(selected = "auto") {
+  if (!taskProfileField) return;
+  const profiles = taskProfiles.length ? taskProfiles : [{ id: "auto", label: "Auto" }];
+  taskProfileField.innerHTML = profiles
+    .map((profile) => `<option value="${escapeHtml(profile.id)}">${escapeHtml(profile.label || profile.id)}</option>`)
+    .join("");
+  taskProfileField.value = profiles.some((profile) => profile.id === selected) ? selected : "auto";
 }
 
 function renderWrapperStatus(wrappers = lastWrappers) {
@@ -893,6 +936,7 @@ function applyLanguage(language, { persist = true } = {}) {
   }
 
   renderKeyStatus();
+  renderProjectStatus();
   renderWrapperStatus();
   renderWorkspacePanel();
   renderSandboxStatus();
@@ -922,6 +966,7 @@ function formPayload() {
     allowFileTools: document.querySelector("#allowFileTools").checked,
     allowWrapperTools: allowWrapperToolsField.checked,
     preferredWrapper: preferredWrapperField.value,
+    taskProfile: taskProfileField?.value || "auto",
     useDockerSandbox: sandboxModeField.value !== "host",
     dockerSandboxImage: document.querySelector("#dockerSandboxImage").value.trim(),
     allowPasswords: document.querySelector("#allowPasswords").checked,
@@ -1673,6 +1718,30 @@ sandboxModeField.addEventListener("change", updatePackageWarning);
 packageInstallPolicyField.addEventListener("change", updatePackageWarning);
 allowWrapperToolsField.addEventListener("change", () => renderWrapperStatus());
 preferredWrapperField.addEventListener("change", () => renderWrapperStatus());
+taskProfileField?.addEventListener("change", schedulePreferenceSave);
+
+saveApiKeyButton?.addEventListener("click", async () => {
+  const provider = setupProviderField.value || "deepseek";
+  const apiKey = setupApiKeyField.value.trim();
+  if (!apiKey) {
+    setupStatusEl.textContent = t("setupKeyLabel");
+    return;
+  }
+  setupStatusEl.textContent = t("sendingStatus");
+  const response = await fetch(`/api/keys/${encodeURIComponent(provider)}`, {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify({ apiKey }),
+  });
+  const data = await response.json().catch(() => ({}));
+  setupApiKeyField.value = "";
+  if (!response.ok) {
+    setupStatusEl.textContent = data.error || t("keySaveFailed");
+    return;
+  }
+  setupStatusEl.textContent = t("keySavedStatus");
+  renderKeyStatus(data.keyStatus);
+});
 
 form.addEventListener("input", schedulePreferenceSave);
 form.addEventListener("change", schedulePreferenceSave);
@@ -1785,6 +1854,8 @@ async function loadConfig() {
   const data = await response.json();
   const prefs = data.preferences || {};
   routingPresets = data.routing?.presets || {};
+  taskProfiles = data.taskProfiles || [];
+  projectInfo = data.project || null;
   defaults.openai = data.defaults?.openai?.model || defaults.openai;
   defaults.deepseek = routingPresets.fast?.model || data.defaults?.deepseek?.model || defaults.deepseek;
   defaults.mock = data.defaults?.mock?.model || defaults.mock;
@@ -1794,9 +1865,10 @@ async function loadConfig() {
   routingModeField.value = prefs.routingMode || "smart";
   providerField.value = prefs.provider || "deepseek";
   modelField.value = prefs.model || defaults[providerField.value] || "deepseek-v4-flash";
+  renderTaskProfiles(prefs.taskProfile || "auto");
   document.querySelector("#startUrl").value = prefs.startUrl || "";
   document.querySelector("#allowedDomains").value = prefs.allowedDomains || "";
-  document.querySelector("#commandCwd").value = prefs.commandCwd || "/home/lachlan/ProjectsLFS/Agent";
+  document.querySelector("#commandCwd").value = prefs.commandCwd || data.project?.root || "";
   document.querySelector("#headless").checked = prefs.headless ?? data.defaults.headless;
   document.querySelector("#maxSteps").value = prefs.maxSteps ?? data.defaults.maxSteps;
   sandboxModeField.value = prefs.sandboxMode || (prefs.useDockerSandbox ? "docker-workspace" : "host");
@@ -1810,6 +1882,7 @@ async function loadConfig() {
   document.querySelector("#allowDestructive").checked = prefs.allowDestructive ?? false;
 
   renderKeyStatus(data.keyStatus);
+  renderProjectStatus(data.project);
   renderWrapperStatus(data.wrappers || []);
   renderWorkspacePanel(data.workspace, []);
   await refreshWorkspaceChanges();

package/public/index.html

@@ -36,6 +36,39 @@
         <p class="subtle" data-i18n="intro">
           Browser agent with provider selection, resumable runs, persisted settings, and an optional guarded shell tool with Docker sandbox support.
         </p>
+        <section class="project-card">
+          <strong data-i18n="projectStatusTitle">Project folder</strong>
+          <p id="project-status" class="subtle">Loading project context...</p>
+        </section>
+
+        <section id="setup-card" class="setup-card" hidden>
+          <div>
+            <strong data-i18n="setupTitle">Provider setup</strong>
+            <p class="subtle" data-i18n="setupHelp">
+              DeepSeek/OpenAI keys are missing. Use mock mode, export an env var, or save a project-local DeepSeek key.
+            </p>
+            <p class="subtle" data-i18n="setupEnvHelp">
+              Env vars: DEEPSEEK_API_KEY, OPENAI_API_KEY, or LLM_API_KEY. Mock mode remains available for local tests.
+            </p>
+          </div>
+          <div class="grid">
+            <label>
+              <span data-i18n="setupProviderLabel">Provider</span>
+              <select id="setup-provider">
+                <option value="deepseek">DeepSeek</option>
+                <option value="openai">OpenAI</option>
+              </select>
+            </label>
+            <label>
+              <span data-i18n="setupKeyLabel">API key</span>
+              <input id="setup-api-key" type="password" autocomplete="off" placeholder="Stored in .aginti/.env" />
+            </label>
+          </div>
+          <div class="actions">
+            <button id="save-api-key" type="button" class="secondary" data-i18n="saveKeyButton">Save local key</button>
+            <span id="setup-status" class="subtle"></span>
+          </div>
+        </section>
 
         <form id="run-form">
           <label>
@@ -67,6 +100,13 @@
           <p id="routing-hint" class="subtle route-hint"></p>
           <p id="model-route-status" class="subtle route-hint"></p>
 
+          <label>
+            <span data-i18n="taskProfileLabel">Task profile</span>
+            <select id="taskProfile" name="taskProfile">
+              <option value="auto">Auto</option>
+            </select>
+          </label>
+
           <label>
             <span data-i18n="goalLabel">Goal</span>
             <textarea

package/public/styles.css

@@ -120,6 +120,27 @@ h1 {
   line-height: 1.45;
 }
 
+.project-card,
+.setup-card {
+  display: grid;
+  min-width: 0;
+  gap: 10px;
+  margin: 14px 0;
+  padding: 13px;
+  border: 1px solid rgba(15, 118, 110, 0.18);
+  border-radius: 16px;
+  background: linear-gradient(135deg, rgba(204, 251, 241, 0.44), rgba(255, 247, 237, 0.9));
+}
+
+.project-card p,
+.setup-card p {
+  overflow-wrap: anywhere;
+}
+
+.setup-card[hidden] {
+  display: none;
+}
+
 form,
 .checks {
   display: grid;

package/scripts/smoke-coding-tools.js

@@ -17,7 +17,7 @@ function assert(condition, message) {
   if (!condition) throw new Error(message);
 }
 
-async function runMock(goal, sessionId) {
+async function runMock(goal, sessionId, { resume = false } = {}) {
   const config = resolveRuntimeConfig(
     {
       provider: "mock",
@@ -26,6 +26,7 @@ async function runMock(goal, sessionId) {
       goal,
       commandCwd: workspace,
       maxSteps: 5,
+      resume: resume ? sessionId : "",
     },
     {
       baseDir: runtimeDir,
@@ -38,7 +39,7 @@ async function runMock(goal, sessionId) {
       allowFileTools: true,
       sandboxMode: "host",
       packageInstallPolicy: "block",
-      sessionId,
+      sessionId: resume ? "" : sessionId,
     }
   );
 
@@ -78,11 +79,23 @@ try {
     "repaired DeepSeek history retained an orphan stale tool message"
   );
 
-  const writeRun = await runMock("Create file: notes/mock-output.txt with a short coding smoke message.", "coding-write");
-  const written = await fs.readFile(path.join(workspace, "notes/mock-output.txt"), "utf8");
+  const writeRun = await runMock("Create notes/hello.md with a short coding smoke message.", "coding-write");
+  const written = await fs.readFile(path.join(workspace, "notes/hello.md"), "utf8");
   assert(written.includes("Created by AgInTiFlow mock mode."), "mock write did not create expected file");
   assert(writeRun.events.some((event) => event.type === "file.changed"), "write run did not persist file.changed event");
 
+  await runMock("Create notes/resume.md with resumed session content.", "coding-write", { resume: true });
+  const resumed = await fs.readFile(path.join(workspace, "notes/resume.md"), "utf8");
+  assert(resumed.includes("Created by AgInTiFlow mock mode."), "mock resume did not create a new requested file");
+
+  let duplicateFailed = false;
+  try {
+    await runMock("Create notes/hello.md with duplicate content.", "coding-write-duplicate");
+  } catch (error) {
+    duplicateFailed = /File already exists|Mock tool failed/.test(String(error));
+  }
+  assert(duplicateFailed, "duplicate mock write did not fail safely");
+
   await runMock("Create file: /workspace/virtual-output.txt with virtual Docker path support.", "coding-write-virtual");
   const virtualWritten = await fs.readFile(path.join(workspace, "virtual-output.txt"), "utf8");
   assert(virtualWritten.includes("Created by AgInTiFlow mock mode."), "virtual /workspace path was not mapped safely");
@@ -123,6 +136,8 @@ try {
         checks: [
           "deepseek_history_repair",
           "write_file",
+          "duplicate_write_failed",
+          "resume_session_write",
           "virtual_workspace_path",
           "apply_patch",
           "block_env",

package/scripts/smoke-web-api.js

@@ -74,6 +74,23 @@ try {
   if (!config.keyStatus?.mock) throw new Error("mock provider is not advertised by /api/config");
   if (!config.workspace?.enabled) throw new Error("workspace file tools are not advertised by /api/config");
   if (config.preferences?.preferredWrapper !== "codex") throw new Error("Codex is not the default preferred wrapper");
+  if (config.project?.root !== runtimeDir) throw new Error("web project root did not default to launch directory");
+  if (config.preferences?.commandCwd !== runtimeDir) throw new Error("commandCwd did not default to project root");
+  if (!Array.isArray(config.taskProfiles) || !config.taskProfiles.some((profile) => profile.id === "latex")) {
+    throw new Error("task profiles are not advertised by /api/config");
+  }
+
+  const keyStatus = await fetchJson("/api/keys/status");
+  if (typeof keyStatus.keyStatus?.deepseek !== "boolean") throw new Error("key status endpoint is invalid");
+  if ("localEnvPath" in keyStatus.keyStatus) throw new Error("key status leaked a local env path");
+  const savedKey = await fetchJson("/api/keys/deepseek", {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify({ apiKey: "test-deepseek-key-not-real" }),
+  });
+  if (!savedKey.ok || !savedKey.keyStatus?.deepseek || "apiKey" in savedKey || "key" in savedKey) {
+    throw new Error("local key save endpoint returned invalid or sensitive data");
+  }
 
   const status = await fetchJson("/api/sandbox/status");
   if (!status.status?.workspaceReadable) throw new Error("sandbox status did not report a readable workspace");
@@ -111,6 +128,30 @@ try {
   if (run.status !== "finished") throw new Error(`mock run failed: ${run.error || "unknown error"}`);
   if (!/Mock run complete/.test(run.result)) throw new Error("mock run did not return the expected result");
 
+  const fileRunStart = await fetchJson("/api/runs", {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify({
+      provider: "mock",
+      routingMode: "manual",
+      model: "mock-agent",
+      goal: "Create notes/hello.md with safe web API content.",
+      commandCwd: runtimeDir,
+      sandboxMode: "host",
+      packageInstallPolicy: "block",
+      allowShellTool: false,
+      allowFileTools: true,
+      preferredWrapper: "codex",
+      maxSteps: 4,
+      headless: true,
+      taskProfile: "code",
+    }),
+  });
+  const fileRun = await waitForRun(fileRunStart.sessionId);
+  if (fileRun.status !== "finished") throw new Error(`mock file run failed: ${fileRun.error || "unknown error"}`);
+  const hello = await fs.readFile(path.join(runtimeDir, "notes", "hello.md"), "utf8");
+  if (!hello.includes("Created by AgInTiFlow mock mode.")) throw new Error("mock file run did not create requested path");
+
   const chat = await fetchJson(`/api/sessions/${encodeURIComponent(runStart.sessionId)}/chat`);
   if (!Array.isArray(chat.chat) || chat.chat.length < 2) throw new Error("chat history was not persisted");
 
@@ -189,6 +230,8 @@ try {
         ok: true,
         endpoints: [
           "/api/config",
+          "/api/keys/status",
+          "POST /api/keys/:provider",
           "/api/sandbox/status",
           "/api/sandbox/preflight",
           "/api/runs",

package/src/agent-runner.js

@@ -13,6 +13,7 @@ import { evaluateCommandPolicy } from "./command-policy.js";
 import { redactSensitiveText, redactValue } from "./redaction.js";
 import { executeWorkspaceTool, summarizeWorkspaceTools, WORKSPACE_TOOL_NAMES } from "./workspace-tools.js";
 import { normalizeCanvasPayload } from "./artifact-tunnel.js";
+import { getTaskProfile } from "./task-profiles.js";
 
 const exec = promisify(execCallback);
 const BROWSER_TOOLS = new Set(["open_url", "click", "type", "scroll", "press", "back"]);
@@ -108,6 +109,7 @@ export function repairModelMessageHistory(state, config = {}) {
 
 function createInitialState(config, sessionId) {
   const now = new Date().toISOString();
+  const taskProfile = getTaskProfile(config.taskProfile);
   return {
     sessionId,
     createdAt: now,
@@ -151,6 +153,7 @@ function createInitialState(config, sessionId) {
           config.allowWrapperTools
             ? `External coding-agent wrappers are available as advisory tools only. Use the selected wrapper only: ${normalizeWrapperName(config.preferredWrapper)}. Wrapper status: ${wrapperStatusText()}.`
             : "External coding-agent wrappers are disabled.",
+          `Task profile: ${taskProfile.label}. ${taskProfile.prompt}`,
           "A frontend canvas/artifacts tunnel exists. Use send_to_canvas when important markdown, diffs, screenshots, images, or workspace files should be highlighted in the UI. It is optional and ordinary final text can still go directly to finish.",
           "For visual-output requests such as draw, plot, graph, chart, diagram, figure, image, or visualization, proactively publish a canvas artifact even when the user does not mention canvas. If workspace file tools are enabled, prefer creating a small SVG or markdown artifact and call send_to_canvas with selected=true.",
           "For LaTeX/PDF requests, create the needed source/assets, compile with the available allowlisted TeX toolchain, and publish the resulting PDF through send_to_canvas. For subfolder documents, keep outputs beside the source. Use pdflatex-compatible figure formats such as PDF or PNG.",
@@ -172,6 +175,7 @@ function createInitialState(config, sessionId) {
           config.allowWrapperTools
             ? `Agent wrappers: selected=${normalizeWrapperName(config.preferredWrapper)}; ${wrapperStatusText()}`
             : "",
+          `Task profile: ${taskProfile.label}. ${taskProfile.prompt}`,
           "Canvas/artifacts tunnel: available through send_to_canvas for optional frontend rendering.",
           "Visual-output requests should produce a canvas artifact without requiring the user to ask for canvas explicitly.",
           "LaTeX/PDF requests should produce source artifacts and, when possible, a compiled PDF artifact. For subfolder documents, keep outputs beside the source. For figure-in-document tasks, create PDF/PNG figures that pdflatex can include.",
@@ -248,6 +252,7 @@ function appendChatEntry(state, role, content) {
 function applyContinuationPrompt(state, config, observers) {
   if (!config.resume || !config.goal) return;
 
+  const taskProfile = getTaskProfile(config.taskProfile);
   ensureChatState(state);
   state.goal = config.goal;
   state.provider = config.provider;
@@ -271,6 +276,7 @@ function applyContinuationPrompt(state, config, observers) {
       config.allowWrapperTools
         ? `Agent wrappers: selected=${normalizeWrapperName(config.preferredWrapper)}; ${wrapperStatusText()}`
         : "",
+      `Task profile: ${taskProfile.label}. ${taskProfile.prompt}`,
     ]
       .filter(Boolean)
       .join("\n"),
@@ -754,6 +760,7 @@ export async function runAgent(config) {
       model: config.model,
       routingMode: config.routingMode,
       routeReason: config.routeReason,
+      taskProfile: config.taskProfile,
       commandCwd: config.commandCwd,
       allowShellTool: config.allowShellTool,
       allowWrapperTools: config.allowWrapperTools,
@@ -817,6 +824,7 @@ export async function runAgent(config) {
           plan: state.plan || "",
           suggestedStartUrl: config.startUrl || "",
           canvasArtifactsAvailable: true,
+          taskProfile: getTaskProfile(config.taskProfile),
         })}`,
       });
 
@@ -899,6 +907,12 @@ export async function runAgent(config) {
           });
         }
 
+        if (config.provider === "mock" && toolResult.ok === false && !toolResult.blocked) {
+          throw new Error(
+            `Mock tool failed: ${toolResult.error || toolResult.reason || `${toolResult.toolName || "tool"} returned ok=false`}`
+          );
+        }
+
         if (toolResult.done) {
           state.stepsCompleted = step;
           state.updatedAt = new Date().toISOString();