micro-models-agent 0.63.0 → 0.63.3

package/CHANGELOG.md

@@ -4,6 +4,33 @@ All notable changes to Micro Models Agent (MMA) will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/), and this project adheres to [Semantic Versioning](https://semver.org/).
 
+## [0.63.3] - 2026-09-17
+
+### Fixed
+- **Plan-alignment false positives hard-blocked legitimate writes**: `checkPlanAlignment` (`src/modules/execution/module.ts`) could accumulate 3 "outside the current step" warnings and block a tool call the agent needed for the current step (observed in `ses_mu5wed4r`: the current step's own `src/main.ts` was blocked). Causes fixed:
+  - **Path separators were not normalized** (root cause): model tool calls pass absolute Windows paths (`C:\…\src\main.ts`) while plan steps name files relatively (`src/main.ts`); the raw substring comparison never intersected backslash vs forward slash, so *every* write was treated as off-path. Both sides now go through `planPathToken` (forward slashes + lowercase) before matching.
+  - The completed-step rewrite exemption compared paths one-directionally (`finishedPaths.some((s) => s.includes(p))`), so a done step naming `main.ts` did not exempt a rewrite of `src/main.ts` — the overlap check is now bidirectional, mirroring the current-step comparison.
+  - Read-only diagnostic/structural tools were not exempt: `lsp_check` (mandated by the system prompt after every write) and `project_map` now join the allow-list (`PLAN_ALIGNMENT_EXEMPT_TOOLS`), alongside the other filesystem-reads, web/memory/session reads, process introspection and interactive prompts. Only file-mutating tools are pinned to the current step's file tokens.
+- **Plugin-blocked calls were treated as real tool failures**: a guard refusal (`onBeforeTool` → `false`/string) is now flagged `blocked` on `ToolResult` and propagated to the context message (`src/tools/executor.ts`, `src/core/agent/tool-batch.ts`, `src/modules/context/manager.ts`). Blocked calls no longer feed the compaction "Already tried & failed — do NOT repeat" memory, the per-tool repeated-failure rule, or the consecutive-failure recovery counter — so a transient guard (or a guard bug) cannot convince the agent to never retry a correct action.
+
+### Added
+- **Browser launch overrides**: `MMA_BROWSER_EXECUTABLE` (absolute path) and `MMA_BROWSER_CHANNEL` (`chrome`/`msedge`/`chromium`) let the browser tool use an installed system browser instead of Playwright's pinned Chromium. When the bundled revision is missing (offline / blocked CDN) it now falls back to `chrome`, then `msedge`, automatically (`src/modules/browser/driver.ts`, `bridge-server.mjs`).
+
+## [0.63.2] - 2026-09-17
+
+### Fixed
+- **Long answers were silently cut by the completion cap**: `max_tokens` (default 4096) is sent on every request; when the model hit it while already streaming text, `finish_reason: "length"` was ignored and the truncated text was accepted as the final answer. The provider now emits a non-fatal `warning` chunk (`src/llm/openai-compat.ts`, streaming and non-streaming paths) that the agent renders in the chat — the answer still arrives, but the user sees it was cut off. Tool-call/empty-response truncation keeps the existing recoverable-error path.
+- **Resumed sessions lost their history**: `saveAssistantMessage` and `logToolResult` (`src/core/session-logger.ts`) persisted only the first 500 chars of each assistant/tool message, so `/resume` seeded the model with truncated context. Full content is stored now; the context budget still trims what actually goes to the LLM.
+
+## [0.63.1] - 2026-09-17
+
+### Added
+- **`ui.verbose`**: opt-in chatty mode for internal orchestration status. MoE planning / re-plan / execution / scope one-liners (`src/core/agent-moe.ts`) are now behind this flag (default `false`) instead of always being written to the chat.
+
+### Changed
+- **Failed tool output is no longer echoed verbatim**: the chat used to print the full model-facing `result.output` of any tool without a `display` — including `Hint:` / recovery directives / "Do NOT …" instructions written for the agent. Only the first non-empty line (the gist of the error) is shown now, in red (`compactToolError`, `src/core/agent/tool-output.ts`).
+- **Live busy spinners for tool runs**: the renderer animates while `read_file` / `write_file` / `edit_file` / `bash` / … execute and restarts the spinner on every LLM "thinking" phase. Previously the spinner stopped at the first tool call and the rest of the run looked frozen. Interactive tools (`question`, `approve`) never spin; dead `BUSY_TOOLS` set removed.
+
 ## [0.63.0] - 2026-09-17
 
 ### Added

package/dist/i18n/en.json

@@ -44,6 +44,7 @@
   "error.llm_stream_idle_toolcall": "LLM stream stalled after a tool_call started — no data for {timeout}ms. The provider most likely buffers SSE instead of streaming tool-call argument deltas (seen with LM Studio). Raise \"retry.noDataTimeoutMs\" in ~/.mma/config.json (restart required)",
   "error.llm_truncated": "Response hit the completion token limit ({tokens}) and was cut off before any content arrived. Split the task into smaller outputs or raise \"maxCompletionTokens\" in the config",
   "error.llm_truncated_toolcall": "Response hit the completion token limit ({tokens}) in the middle of a tool_call — its arguments were cut off. Write the file in smaller chunks (several write_file/edit_file calls) or raise \"maxCompletionTokens\" in the config",
+  "error.llm_output_truncated": "The response hit the completion token limit ({tokens}) and was cut off — this answer is incomplete. Raise \"maxCompletionTokens\" in the config to allow longer replies",
   "error.llm_provider_stream_error": "Provider returned an error mid-stream: {error}",
   "error.llm_timeout": "LLM request timed out ({timeout}ms)",
   "env.runtime_node": "Running under Node (v{version}) — clipboard image paste, subagent performance and LSP spawn on Windows degrade. Install Bun (https://bun.sh) for full features.",

package/dist/i18n/ru.json

@@ -43,6 +43,7 @@
   "error.llm_stream_idle_toolcall": "Поток LLM завис после начала tool_call — нет данных {timeout}мс. Скорее всего провайдер буферизирует SSE вместо потоковой передачи аргументов tool_call (наблюдается в LM Studio). Увеличьте \"retry.noDataTimeoutMs\" в ~/.mma/config.json (нужен рестарт)",
   "error.llm_truncated": "Ответ упёрся в лимит токенов генерации ({tokens}) и был обрезан до какого-либо содержимого. Разбейте задачу на меньшие порции вывода или увеличьте \"maxCompletionTokens\" в конфиге",
   "error.llm_truncated_toolcall": "Ответ упёрся в лимит токенов генерации ({tokens}) посреди tool_call — его аргументы обрезаны. Пишите файл частями (несколько вызовов write_file/edit_file) или увеличьте \"maxCompletionTokens\" в конфиге",
+  "error.llm_output_truncated": "Ответ упёрся в лимит токенов генерации ({tokens}) и был обрезан — этот ответ неполный. Увеличьте \"maxCompletionTokens\" в конфиге, чтобы разрешить длинные ответы",
   "error.llm_provider_stream_error": "Провайдер вернул ошибку посреди стрима: {error}",
   "error.llm_timeout": "Время запроса LLM истекло ({timeout}мс)",
   "env.runtime_node": "Запущено под Node (v{version}) — вставка изображений из буфера и LSP на Windows работают урезанно. Установите Bun (https://bun.sh) для полного функционала.",

package/dist/main.js

@@ -2373,7 +2373,8 @@ var init_defaults = __esm(() => {
     ui: {
       spinner: true,
       showContextStats: false,
-      showCompaction: true
+      showCompaction: true,
+      verbose: false
     },
     mcpServers: {
       context7: {
@@ -2484,6 +2485,7 @@ The path was joined onto the working directory because it does not exist as give
     "error.llm_stream_idle_toolcall": 'LLM stream stalled after a tool_call started — no data for {timeout}ms. The provider most likely buffers SSE instead of streaming tool-call argument deltas (seen with LM Studio). Raise "retry.noDataTimeoutMs" in ~/.mma/config.json (restart required)',
     "error.llm_truncated": 'Response hit the completion token limit ({tokens}) and was cut off before any content arrived. Split the task into smaller outputs or raise "maxCompletionTokens" in the config',
     "error.llm_truncated_toolcall": 'Response hit the completion token limit ({tokens}) in the middle of a tool_call — its arguments were cut off. Write the file in smaller chunks (several write_file/edit_file calls) or raise "maxCompletionTokens" in the config',
+    "error.llm_output_truncated": 'The response hit the completion token limit ({tokens}) and was cut off — this answer is incomplete. Raise "maxCompletionTokens" in the config to allow longer replies',
     "error.llm_provider_stream_error": "Provider returned an error mid-stream: {error}",
     "error.llm_timeout": "LLM request timed out ({timeout}ms)",
     "env.runtime_node": "Running under Node (v{version}) — clipboard image paste, subagent performance and LSP spawn on Windows degrade. Install Bun (https://bun.sh) for full features.",
@@ -3320,6 +3322,7 @@ var init_ru = __esm(() => {
     "error.llm_stream_idle_toolcall": 'Поток LLM завис после начала tool_call — нет данных {timeout}мс. Скорее всего провайдер буферизирует SSE вместо потоковой передачи аргументов tool_call (наблюдается в LM Studio). Увеличьте "retry.noDataTimeoutMs" в ~/.mma/config.json (нужен рестарт)',
     "error.llm_truncated": 'Ответ упёрся в лимит токенов генерации ({tokens}) и был обрезан до какого-либо содержимого. Разбейте задачу на меньшие порции вывода или увеличьте "maxCompletionTokens" в конфиге',
     "error.llm_truncated_toolcall": 'Ответ упёрся в лимит токенов генерации ({tokens}) посреди tool_call — его аргументы обрезаны. Пишите файл частями (несколько вызовов write_file/edit_file) или увеличьте "maxCompletionTokens" в конфиге',
+    "error.llm_output_truncated": 'Ответ упёрся в лимит токенов генерации ({tokens}) и был обрезан — этот ответ неполный. Увеличьте "maxCompletionTokens" в конфиге, чтобы разрешить длинные ответы',
     "error.llm_provider_stream_error": "Провайдер вернул ошибку посреди стрима: {error}",
     "error.llm_timeout": "Время запроса LLM истекло ({timeout}мс)",
     "env.runtime_node": "Запущено под Node (v{version}) — вставка изображений из буфера и LSP на Windows работают урезанно. Установите Bun (https://bun.sh) для полного функционала.",
@@ -6361,6 +6364,10 @@ class OpenAICompatProvider {
             tokens: maxTokens
           }), { terminal: true, recoverable: true });
         }
+        yield {
+          type: "warning",
+          content: t("error.llm_output_truncated", { tokens: maxTokens })
+        };
       }
     } finally {
       cleanup();
@@ -6420,12 +6427,13 @@ class OpenAICompatProvider {
     };
   }
   async doNonStreaming(messages, tools, signal, options) {
+    const maxTokens = options?.maxTokens ?? this.config.maxCompletionTokens ?? 4096;
     const body = buildRequestBody({
       model: this.model,
       messages,
       tools,
       stream: false,
-      maxTokens: options?.maxTokens ?? this.config.maxCompletionTokens ?? 4096,
+      maxTokens,
       reasoningEffort: options?.reasoningEffort,
       reasoningStrategy: options?.reasoningStrategy,
       ...this.cacheHints()
@@ -6475,6 +6483,12 @@ class OpenAICompatProvider {
           });
         }
       }
+      if (choice.finish_reason === "length" && msg.content && !msg.tool_calls) {
+        chunks.push({
+          type: "warning",
+          content: t("error.llm_output_truncated", { tokens: maxTokens })
+        });
+      }
       if (data.usage) {
         chunks.push({
           type: "done",
@@ -7285,6 +7299,7 @@ class ToolExecutor {
             const reason = typeof proceed === "string" ? proceed : undefined;
             return {
               success: false,
+              blocked: true,
               output: reason ? t("tool.blocked_reason", { plugin: pluginName, reason }) : t("tool.blocked", { plugin: pluginName }),
               toolCallId: call.id
             };
@@ -10419,7 +10434,7 @@ class SessionLogger {
     const caller = this.getCaller?.();
     this.session?.appendMessage({
       role: "assistant",
-      content: content.slice(0, 500),
+      content,
       timestamp: new Date().toISOString(),
       provider: caller?.provider,
       model: caller?.model
@@ -10448,7 +10463,7 @@ class SessionLogger {
     const caller = this.getCaller?.();
     this.session?.appendMessage({
       role: "tool",
-      content: sanitizeLogMessage(result.output.slice(0, 500)),
+      content: sanitizeLogMessage(result.output),
       name: call.name,
       timestamp: new Date().toISOString(),
       provider: caller?.provider,
@@ -13892,6 +13907,11 @@ async function runWithMoE(deps, input, fallback, opts = {}) {
     if (signal?.aborted)
       throw newAbortError();
   };
+  const verbose = config.ui?.verbose === true;
+  const status = (message) => {
+    if (verbose)
+      onMeta?.(message);
+  };
   ensureNotAborted();
   let orchestrator = deps.orchestratorOverride;
   if (!orchestrator) {
@@ -13909,7 +13929,7 @@ async function runWithMoE(deps, input, fallback, opts = {}) {
     logger.warn("MoE enabled but no orchestrator model configured — falling back to single-agent. Set orchestrator.model to activate MoE.");
     return fallback();
   }
-  onMeta?.(`\uD83E\uDD16 Planning with MoE mode...
+  status(`\uD83E\uDD16 Planning with MoE mode...
 `);
   onPhase?.("thinking");
   const planResult = await orchestrator.plan(input, undefined, signal);
@@ -13931,7 +13951,7 @@ async function runWithMoE(deps, input, fallback, opts = {}) {
   for (let cycle = 1;cycle <= maxCycles; cycle++) {
     ensureNotAborted();
     if (cycle > 1) {
-      onMeta?.(t("moe.replan_started", { cycle, max: maxCycles }) + `
+      status(t("moe.replan_started", { cycle, max: maxCycles }) + `
 `);
       emit("moe_replan", {
         cycle,
@@ -13946,7 +13966,7 @@ async function runWithMoE(deps, input, fallback, opts = {}) {
       if (!retry.valid) {
         logger.warn(`MoE plan validation failed: ${retry.errors.join("; ")}`);
         if (cycle === 1) {
-          onMeta?.(`⚠️ Plan validation failed. Falling back to single-agent mode.
+          status(`⚠️ Plan validation failed. Falling back to single-agent mode.
 `);
           return fallback();
         }
@@ -13954,7 +13974,7 @@ async function runWithMoE(deps, input, fallback, opts = {}) {
       }
       currentPlan = applied;
       if (validation.autoFixes.length > 0) {
-        onMeta?.(`\uD83D\uDD27 Auto-fixed ${validation.autoFixes.length} plan issues.
+        status(`\uD83D\uDD27 Auto-fixed ${validation.autoFixes.length} plan issues.
 `);
       }
     }
@@ -13971,19 +13991,19 @@ async function runWithMoE(deps, input, fallback, opts = {}) {
       onScopeRequest: async (subtaskId, req) => {
         const decision = await orchestrator.resolveScopeRequest(subtaskId, req, signal);
         if (decision.action === "approve") {
-          onMeta?.(t("moe.scope_approved", {
+          status(t("moe.scope_approved", {
             subtask: subtaskId,
             files: [...decision.write, ...decision.read].join(", ")
           }) + `
 `);
         } else {
-          onMeta?.(t("moe.scope_rejected", { subtask: subtaskId }) + `
+          status(t("moe.scope_rejected", { subtask: subtaskId }) + `
 `);
         }
         return decision;
       }
     });
-    onMeta?.(t("moe.executing", { count: String(currentPlan.subtasks.length), cycle: String(cycle) }) + `
+    status(t("moe.executing", { count: String(currentPlan.subtasks.length), cycle: String(cycle) }) + `
 `);
     const skipIds = new Set;
     const carriedResults = [];
@@ -14019,7 +14039,7 @@ async function runWithMoE(deps, input, fallback, opts = {}) {
       mergedResults.set(r.subtaskId, r);
     }
     const succeeded = [...mergedResults.values()].filter((r) => r.success).length;
-    onMeta?.(`✅ ${t("moe.execution_complete", { succeeded: String(succeeded), total: String(mergedResults.size) })}
+    status(`✅ ${t("moe.execution_complete", { succeeded: String(succeeded), total: String(mergedResults.size) })}
 `);
     const verifier = deps.verifierOverride ?? new StepVerifier(baseDir);
     const knownTags = collectKnownToolTags(toolExecutor);
@@ -14691,6 +14711,17 @@ function summarizeToolArgs(args) {
     return ` (${s})`;
   return ` (${s.slice(0, TOOL_ARGS_SUMMARY_MAX_CHARS)}…(${s.length} chars total))`;
 }
+function compactToolError(output, max = 200) {
+  const text = String(output ?? "");
+  for (const line of text.split(`
+`)) {
+    const trimmed = line.trim();
+    if (!trimmed)
+      continue;
+    return trimmed.length > max ? `${trimmed.slice(0, max - 1)}…` : trimmed;
+  }
+  return "";
+}
 var init_tool_output = __esm(() => {
   init_i18n();
 });
@@ -14766,7 +14797,7 @@ class ToolBatchExecutor {
       if (this.deps.isShutdownRequested())
         break;
       const duration = Date.now() - startTime;
-      if (!result.success && !infoTools.has(call.name))
+      if (!result.success && !result.blocked && !infoTools.has(call.name))
         anyToolFailed = true;
       if (result.success && call.name === "plan") {
         const action = String(call.arguments.action ?? "");
@@ -14788,10 +14819,17 @@ class ToolBatchExecutor {
 ` + result.display + `
 `);
         }
-      } else {
+      } else if (result.success) {
         const metaOut = pluginManager.runOnMeta({ iteration, logger, contextManager }, result.output);
         onMeta?.(`
 ` + pc2.dim(metaOut) + `
+`);
+      } else {
+        const line = compactToolError(result.output);
+        const metaOut = line ? pluginManager.runOnMeta({ iteration, logger, contextManager }, line) : "";
+        if (metaOut)
+          onMeta?.(`
+` + pc2.red(metaOut) + `
 `);
       }
       if (result.diff) {
@@ -14808,6 +14846,7 @@ class ToolBatchExecutor {
         name: call.name,
         tool_call_id: call.id,
         success: result.success,
+        blocked: result.blocked,
         arguments: call.arguments
       });
       answeredToolCallIds.add(call.id);
@@ -14826,7 +14865,7 @@ class ToolBatchExecutor {
         slog?.logToolResult(call, result, duration, iteration);
       }
       this.deps.compactionService.compactAfterTool(state);
-      if (!result.success && !infoTools.has(call.name)) {
+      if (!result.success && !result.blocked && !infoTools.has(call.name)) {
         const key = call.name;
         const prev = state.toolFailureCounts.get(key) ?? { count: 0, error: "" };
         prev.count++;
@@ -15697,6 +15736,13 @@ class Agent {
           if (chunk.type === "done" && chunk.usage) {
             tokenTracker.recordApiUsage(state, baseline, chunk.usage);
           }
+          if (chunk.type === "warning" && chunk.content) {
+            logger.warn(`LLM warning: ${chunk.content}`);
+            onChunk?.(`
+
+> ⚠️ ${chunk.content}
+`);
+          }
         }
       } catch (err) {
         if (this.shutdownRequested || err?.name === "AbortError") {
@@ -16189,6 +16235,8 @@ function extractTriedAndFailed(messages) {
   const failures = new Map;
   for (const msg of messages) {
     if (msg.role === "tool" && msg.name && msg.success === false) {
+      if (msg.blocked)
+        continue;
       const key = `${msg.name}:${summarizeArgs(msg.arguments)}`;
       const existing = failures.get(key);
       const errorText = truncate(typeof msg.content === "string" ? msg.content : getMessageText(msg.content), 100);
@@ -19543,6 +19591,8 @@ class BridgeDriver {
       headless: cfg.headless,
       viewport: cfg.viewport,
       timeoutMs: cfg.timeoutMs,
+      executablePath: cfg.overrides?.executablePath,
+      channel: cfg.overrides?.channel,
       maxConsoleLineChars: this.maxConsoleLineChars
     });
   }
@@ -19665,6 +19715,41 @@ var init_bridge_client = __esm(() => {
 
 // src/modules/browser/driver.ts
 import { chromium } from "playwright";
+function resolveBrowserLaunchOverrides(env = process.env) {
+  const executablePath = env.MMA_BROWSER_EXECUTABLE?.trim();
+  const channel = env.MMA_BROWSER_CHANNEL?.trim();
+  const out = {};
+  if (executablePath)
+    out.executablePath = executablePath;
+  if (channel)
+    out.channel = channel;
+  return out;
+}
+function isMissingBrowserError(err) {
+  return /Executable doesn't exist|is not found|looks like Playwright was just installed/i.test(String(err?.message ?? err));
+}
+async function launchChromiumWithFallback(launch, cfg, overrides = {}) {
+  const base = { headless: cfg.headless, timeout: cfg.timeoutMs, ...overrides };
+  try {
+    return await launch(base);
+  } catch (err) {
+    const explicit = overrides.executablePath || overrides.channel;
+    if (explicit || !isMissingBrowserError(err))
+      throw err;
+    let lastErr = err;
+    for (const channel of SYSTEM_CHANNEL_FALLBACKS) {
+      try {
+        return await launch({ ...base, channel });
+      } catch (e) {
+        lastErr = e;
+      }
+    }
+    throw lastErr;
+  }
+}
+function launchChromium(cfg, overrides) {
+  return launchChromiumWithFallback((opts) => chromium.launch(opts), cfg, overrides);
+}
 
 class PlaywrightDriver {
   browser = null;
@@ -19675,10 +19760,7 @@ class PlaywrightDriver {
   async launch(cfg) {
     if (this.browser)
       return;
-    this.browser = await chromium.launch({
-      headless: cfg.headless,
-      timeout: cfg.timeoutMs
-    });
+    this.browser = await launchChromium(cfg, cfg.overrides ?? {});
     this.context = await this.browser.newContext({ viewport: cfg.viewport });
     this.page = await this.context.newPage();
     this.page.setDefaultTimeout(cfg.timeoutMs);
@@ -19767,7 +19849,8 @@ function createBrowserDriver(config) {
   const common = {
     headless: config.headless,
     viewport: { width: config.viewportWidth, height: config.viewportHeight },
-    timeoutMs: config.navigationTimeout
+    timeoutMs: config.navigationTimeout,
+    overrides: resolveBrowserLaunchOverrides()
   };
   return (async () => {
     if (process.versions.bun) {
@@ -19789,9 +19872,10 @@ async function launchBridge(config, common) {
   await bridge.launch(common);
   return bridge;
 }
-var DIRECT_LAUNCH_TIMEOUT_MS = 8000;
+var DIRECT_LAUNCH_TIMEOUT_MS = 8000, SYSTEM_CHANNEL_FALLBACKS;
 var init_driver = __esm(() => {
   init_session();
+  SYSTEM_CHANNEL_FALLBACKS = ["chrome", "msedge"];
 });
 
 // src/modules/browser/types.ts
@@ -22985,6 +23069,9 @@ var init_plan_tool = __esm(() => {
 // src/modules/execution/module.ts
 import { existsSync as existsSync39, readFileSync as readFileSync23 } from "fs";
 import { resolve as resolve18 } from "path";
+function planPathToken(p) {
+  return toForwardSlash(p).toLowerCase();
+}
 
 class ExecutionModule {
   name = "execution";
@@ -23263,38 +23350,28 @@ class ExecutionModule {
     const step = this.tracker.getCurrentStep();
     if (!step)
       return null;
-    const allowedAlways = [
-      "plan",
-      "todo",
-      "verify",
-      "list_dir",
-      "read_file",
-      "glob",
-      "grep",
-      "file_info",
-      "load_skill"
-    ];
-    if (allowedAlways.includes(call.name))
+    if (PLAN_ALIGNMENT_EXEMPT_TOOLS.has(call.name))
       return null;
-    const stepPaths = extractFileLikeTokens(stripUrls(step.description)).map((p) => p.toLowerCase());
+    const stepPaths = extractFileLikeTokens(stripUrls(step.description)).map(planPathToken);
     if (stepPaths.length === 0)
       return null;
     const argStr = this.pathArgStrings(call.arguments).join(" ");
-    const callPaths = extractFileLikeTokens(stripUrls(argStr)).map((p) => p.toLowerCase());
+    const callPaths = extractFileLikeTokens(stripUrls(argStr)).map(planPathToken);
     if (callPaths.length === 0)
       return null;
     const plan = this.tracker.getPlan();
     const finishedPaths = new Set;
     for (const s of plan.steps) {
       if (s.status === "done" || s.status === "skipped") {
-        extractFileLikeTokens(stripUrls(s.description)).map((p) => p.toLowerCase()).forEach((p) => finishedPaths.add(p));
+        extractFileLikeTokens(stripUrls(s.description)).map(planPathToken).forEach((p) => finishedPaths.add(p));
       }
     }
+    const matchesAny = (p, tokens) => tokens.some((t2) => t2 === p || t2.includes(p) || p.includes(t2));
+    const finished = [...finishedPaths];
     const offPath = callPaths.some((p) => {
-      if (finishedPaths.has(p) || [...finishedPaths].some((s) => s.includes(p))) {
+      if (matchesAny(p, finished))
         return false;
-      }
-      return !stepPaths.some((s) => p.includes(s) || s.includes(p));
+      return !matchesAny(p, stepPaths);
     });
     if (!offPath)
       return null;
@@ -23466,7 +23543,7 @@ class ExecutionModule {
     return getMessageText(firstUser.content).trim();
   }
 }
-var ERROR_SEARCH_MIN_INTERVAL_MS = 30000;
+var ERROR_SEARCH_MIN_INTERVAL_MS = 30000, PLAN_ALIGNMENT_EXEMPT_TOOLS;
 var init_module = __esm(() => {
   init_i18n();
   init_tracker();
@@ -23475,10 +23552,36 @@ var init_module = __esm(() => {
   init_auditor();
   init_plan_store();
   init_js_identifiers();
+  init_path_utils();
   init_web_search();
   init_session_isolation();
   init_plan_tool();
   init_execution_plugin();
+  PLAN_ALIGNMENT_EXEMPT_TOOLS = new Set([
+    "plan",
+    "todo",
+    "verify",
+    "enable_tools",
+    "list_dir",
+    "read_file",
+    "glob",
+    "grep",
+    "file_info",
+    "lsp_check",
+    "project_map",
+    "chunk_query",
+    "web_search",
+    "web_fetch",
+    "web_browse",
+    "recall",
+    "search_history",
+    "session_info",
+    "process_list",
+    "process_log",
+    "question",
+    "approve",
+    "load_skill"
+  ]);
 });
 
 // src/modules/security/session-encryption.ts
@@ -39921,7 +40024,7 @@ function toDisplayPath(baseDir, p) {
 }
 var GUTTER = "  ";
 var MAX_OUTPUT_LINES = 50;
-var BUSY_TOOLS = new Set(["lsp_check"]);
+var SPINNERLESS_TOOLS = new Set(["question", "approve"]);
 function toolMarker(tool) {
   switch (tool) {
     case "write_file":
@@ -39995,7 +40098,7 @@ class Renderer {
 `), this.width, (text) => this.out.write(text));
   }
   showLoader() {
-    this.spinner.start("thinking");
+    this.spinner.start(t("ui.thinking"));
   }
   text(chunk) {
     this.endCard();
@@ -40049,6 +40152,7 @@ ${pc2.dim("→")} Thought: `);
     this.thoughtStarted = true;
     this.thoughtStartMs = Date.now();
     this.thoughtHeaderPrinted = false;
+    this.spinner.start(t("ui.thinking"));
   }
   thinkingEnd() {
     this.spinner.stop();
@@ -40068,14 +40172,18 @@ ${pc2.dim("→")} Thought: `);
     const displayArgs = PATH_TOOLS.has(tool) && typeof args.path === "string" ? { ...args, path: toDisplayPath(this.baseDir, args.path) } : args;
     const summary = summarizeArgs2(displayArgs);
     const marker = icon || toolMarker(tool);
+    const label = friendlyTool(tool);
     this.card = { tool, args, start: Date.now() };
     if (stepContext) {
       this.out.write(`
 ${pc2.dim("↓")} ${pc2.cyan(stepContext)}
 `);
     }
-    this.out.write(`${pc2.dim(marker)} ${friendlyTool(tool)}${summary ? ` ${pc2.dim(summary)}` : ""}
+    this.out.write(`${pc2.dim(marker)} ${label}${summary ? ` ${pc2.dim(summary)}` : ""}
 `);
+    if (!SPINNERLESS_TOOLS.has(tool)) {
+      this.spinner.start(`${label}${summary ? ` ${summary}` : ""}`);
+    }
   }
   planBlock(lines) {
     this.endCard();

package/dist/modules/browser/bridge-server.mjs

@@ -41,13 +41,46 @@ function readInt(v, fallback) {
   return Number.isFinite(n) && n > 0 ? n : fallback;
 }
 
+// System browsers to try when the bundled Chromium revision is absent (CDN
+// blocked / offline). Mirrors driver.ts so both transports behave the same.
+const SYSTEM_CHANNEL_FALLBACKS = ["chrome", "msedge"];
+
+function isMissingBrowserError(err) {
+  return /Executable doesn't exist|is not found|looks like Playwright was just installed/i.test(
+    String(err && err.message ? err.message : err),
+  );
+}
+
+async function launchChromium(params, extra) {
+  const base = {
+    headless: params?.headless !== false,
+    timeout: readInt(params?.timeoutMs, 30000),
+    ...extra,
+  };
+  try {
+    return await chromium.launch(base);
+  } catch (err) {
+    // An explicit executable/channel is the user's choice — fail loudly.
+    if (base.executablePath || base.channel || !isMissingBrowserError(err)) throw err;
+    let lastErr = err;
+    for (const channel of SYSTEM_CHANNEL_FALLBACKS) {
+      try {
+        return await chromium.launch({ ...base, channel });
+      } catch (e) {
+        lastErr = e;
+      }
+    }
+    throw lastErr;
+  }
+}
+
 async function handleLaunch(params) {
   if (browser) return { ok: true };
   maxLineChars = readInt(params?.maxConsoleLineChars, 400);
-  browser = await chromium.launch({
-    headless: params?.headless !== false,
-    timeout: readInt(params?.timeoutMs, 30000),
-  });
+  const extra = {};
+  if (params?.executablePath) extra.executablePath = String(params.executablePath);
+  if (params?.channel) extra.channel = String(params.channel);
+  browser = await launchChromium(params, extra);
   context = await browser.newContext({
     viewport: params?.viewport || { width: 1280, height: 720 },
   });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "micro-models-agent",
-  "version": "0.63.0",
+  "version": "0.63.3",
   "description": "Micro Models Agent (MMA) — LLM agent harness for small models (Qwen3.5-9B, 32K-64K context)",
   "type": "module",
   "bin": {