reasonix 0.4.6 → 0.4.12

package/dist/index.js

@@ -47,8 +47,8 @@ function computeWait(attempt, initial, cap, retryAfter) {
 }
 function sleep(ms, signal) {
   if (ms <= 0) return Promise.resolve();
-  return new Promise((resolve3, reject) => {
-    const timer = setTimeout(resolve3, ms);
+  return new Promise((resolve4, reject) => {
+    const timer = setTimeout(resolve4, ms);
     if (signal) {
       const onAbort = () => {
         clearTimeout(timer);
@@ -133,6 +133,27 @@ var DeepSeekClient = class {
     if (opts.responseFormat) payload.response_format = opts.responseFormat;
     return payload;
   }
+  /**
+   * Fetch the current DeepSeek account balance. Separate endpoint
+   * from chat completions, no billing impact. Returns null on any
+   * network/auth failure so callers can gate the balance display
+   * without a hard error — the rest of the session works regardless.
+   */
+  async getBalance(opts = {}) {
+    try {
+      const resp = await this._fetch(`${this.baseUrl}/user/balance`, {
+        method: "GET",
+        headers: { Authorization: `Bearer ${this.apiKey}` },
+        signal: opts.signal
+      });
+      if (!resp.ok) return null;
+      const data = await resp.json();
+      if (!data || !Array.isArray(data.balance_infos)) return null;
+      return data;
+    } catch {
+      return null;
+    }
+  }
   async chat(opts) {
     const ctrl = new AbortController();
     const timer = setTimeout(() => ctrl.abort(), this.timeoutMs);
@@ -279,8 +300,9 @@ Constraints:
 - Each item is plain text, at most {maxItemLen} characters, no markdown.
 - Write in the same language as the trace (Chinese in \u2192 Chinese out, etc.).
 - Do not quote back the trace; write short, specific phrases.`;
-async function harvest(reasoningContent, client, options = {}) {
+async function harvest(reasoningContent, client, options = {}, signal) {
   if (!client || !reasoningContent) return emptyPlanState();
+  if (signal?.aborted) return emptyPlanState();
   const minLen = options.minReasoningLen ?? 40;
   const trimmed = reasoningContent.trim();
   if (trimmed.length < minLen) return emptyPlanState();
@@ -300,7 +322,8 @@ async function harvest(reasoningContent, client, options = {}) {
       ],
       responseFormat: { type: "json_object" },
       temperature: 0,
-      maxTokens: 600
+      maxTokens: 600,
+      signal
     });
     return parsePlanState(resp.content, maxItems, maxItemLen);
   } catch {
@@ -514,7 +537,7 @@ var ToolRegistry = class {
       }
     }));
   }
-  async dispatch(name, argumentsRaw) {
+  async dispatch(name, argumentsRaw, opts = {}) {
     const tool = this._tools.get(name);
     if (!tool) {
       return JSON.stringify({ error: `unknown tool: ${name}` });
@@ -531,7 +554,7 @@ var ToolRegistry = class {
       args = nestArguments(args);
     }
     try {
-      const result = await tool.fn(args);
+      const result = await tool.fn(args, { signal: opts.signal });
       return typeof result === "string" ? result : JSON.stringify(result);
     } catch (err) {
       return JSON.stringify({
@@ -565,8 +588,20 @@ async function bridgeMcpTools(client, opts = {}) {
       name: registeredName,
       description: mcpTool.description ?? "",
       parameters: mcpTool.inputSchema,
-      fn: async (args) => {
-        const toolResult = await client.callTool(mcpTool.name, args);
+      fn: async (args, ctx) => {
+        const toolResult = await client.callTool(mcpTool.name, args, {
+          // Forward server-side progress frames to the bridge caller,
+          // tagged with the registered name so multi-server UIs can
+          // disambiguate. No-op when `onProgress` isn't configured —
+          // the client then also omits the _meta.progressToken and
+          // the server won't emit progress.
+          onProgress: opts.onProgress ? (info) => opts.onProgress({ toolName: registeredName, ...info }) : void 0,
+          // Thread the tool-dispatch AbortSignal all the way down to
+          // the MCP request so Esc truly cancels in flight — the
+          // client will emit notifications/cancelled AND reject the
+          // pending promise immediately, no "wait for subprocess".
+          signal: ctx?.signal
+        });
         return flattenMcpResult(toolResult, { maxChars: maxResultChars });
       }
     });
@@ -1077,6 +1112,16 @@ function costUsd(model, usage) {
   if (!p) return 0;
   return (usage.promptCacheHitTokens * p.inputCacheHit + usage.promptCacheMissTokens * p.inputCacheMiss + usage.completionTokens * p.output) / 1e6;
 }
+function inputCostUsd(model, usage) {
+  const p = DEEPSEEK_PRICING[model];
+  if (!p) return 0;
+  return (usage.promptCacheHitTokens * p.inputCacheHit + usage.promptCacheMissTokens * p.inputCacheMiss) / 1e6;
+}
+function outputCostUsd(model, usage) {
+  const p = DEEPSEEK_PRICING[model];
+  if (!p) return 0;
+  return usage.completionTokens * p.output / 1e6;
+}
 function claudeEquivalentCost(usage) {
   return (usage.promptTokens * CLAUDE_SONNET_PRICING.input + usage.completionTokens * CLAUDE_SONNET_PRICING.output) / 1e6;
 }
@@ -1104,6 +1149,12 @@ var SessionStats = class {
     const c = this.totalClaudeEquivalent;
     return c > 0 ? 1 - this.totalCost / c : 0;
   }
+  get totalInputCost() {
+    return this.turns.reduce((sum, t) => sum + inputCostUsd(t.model, t.usage), 0);
+  }
+  get totalOutputCost() {
+    return this.turns.reduce((sum, t) => sum + outputCostUsd(t.model, t.usage), 0);
+  }
   get aggregateCacheHitRatio() {
     let hit = 0;
     let miss = 0;
@@ -1119,6 +1170,8 @@ var SessionStats = class {
     return {
       turns: this.turns.length,
       totalCostUsd: round(this.totalCost, 6),
+      totalInputCostUsd: round(this.totalInputCost, 6),
+      totalOutputCostUsd: round(this.totalOutputCost, 6),
       claudeEquivalentUsd: round(this.totalClaudeEquivalent, 6),
       savingsVsClaudePct: round(this.savingsVsClaude * 100, 2),
       cacheHitRatio: round(this.aggregateCacheHitRatio, 4),
@@ -1155,11 +1208,13 @@ var CacheFirstLoop = class {
   _turn = 0;
   _streamPreference;
   /**
-   * Set by {@link abort} to short-circuit the tool-call loop after the
-   * current iteration. Reset at the start of each `step()` so an Esc
-   * during one turn doesn't poison the next.
+   * AbortController per active turn. Threaded through the DeepSeek
+   * HTTP calls AND every tool dispatch so Esc actually cancels the
+   * in-flight network/subprocess work — not "we'll get to it after
+   * the current call finishes." Re-created at the start of each
+   * `step()` (the prior turn's signal has already fired).
    */
-  _aborted = false;
+  _turnAbort = new AbortController();
   constructor(opts) {
     this.client = opts.client;
     this.prefix = opts.prefix;
@@ -1191,8 +1246,12 @@ var CacheFirstLoop = class {
       for (const msg of messages) this.log.append(msg);
       this.resumedMessageCount = messages.length;
       if (healedCount > 0) {
+        try {
+          rewriteSession(this.sessionName, messages);
+        } catch {
+        }
         process.stderr.write(
-          `\u25B8 session "${this.sessionName}": healed ${healedCount} oversized tool result(s) (was ${healedFrom.toLocaleString()} chars total). Old payloads were truncated to fit DeepSeek's context window; the conversation is preserved.
+          `\u25B8 session "${this.sessionName}": healed ${healedCount} entr${healedCount === 1 ? "y" : "ies"}${healedFrom > 0 ? ` (was ${healedFrom.toLocaleString()} chars oversized)` : " (dropped dangling tool_calls tail)"}. Rewrote session file.
 `
         );
       }
@@ -1213,7 +1272,7 @@ var CacheFirstLoop = class {
    */
   compact(tightCapChars = 4e3) {
     const before = this.log.toMessages();
-    const { messages, healedCount, healedFrom } = healLoadedMessages(before, tightCapChars);
+    const { messages, healedCount, healedFrom } = shrinkOversizedToolResults(before, tightCapChars);
     const afterBytes = messages.filter((m) => m.role === "tool").reduce((s, m) => s + (typeof m.content === "string" ? m.content.length : 0), 0);
     const charsSaved = healedFrom - afterBytes;
     if (healedCount > 0) {
@@ -1236,6 +1295,29 @@ var CacheFirstLoop = class {
       }
     }
   }
+  /**
+   * Start a fresh conversation WITHOUT exiting. Drops every message
+   * in the in-memory log AND rewrites the session file to empty so
+   * a resume won't re-hydrate the old turns. Unlike `/forget`, which
+   * deletes the session entirely, this keeps the session name and
+   * config intact — it's the "new chat" button.
+   *
+   * The immutable prefix (system prompt + tool specs) is preserved
+   * — that's the cache-first invariant, not part of the conversation.
+   * Returns the number of messages dropped so the UI can show it.
+   */
+  clearLog() {
+    const dropped = this.log.length;
+    this.log.compactInPlace([]);
+    if (this.sessionName) {
+      try {
+        rewriteSession(this.sessionName, []);
+      } catch {
+      }
+    }
+    this.scratch.reset();
+    return { dropped };
+  }
   /**
    * Reconfigure model/harvest/branch/stream mid-session. The loop's log,
    * scratch, and stats are preserved — only the per-turn behavior changes.
@@ -1267,19 +1349,21 @@ var CacheFirstLoop = class {
     this.stream = this.branchEnabled ? false : this._streamPreference;
   }
   buildMessages(pendingUser) {
-    const msgs = [...this.prefix.toMessages(), ...this.log.toMessages()];
+    const healed = healLoadedMessages(this.log.toMessages(), DEFAULT_MAX_RESULT_CHARS);
+    const msgs = [...this.prefix.toMessages(), ...healed.messages];
     if (pendingUser !== null) msgs.push({ role: "user", content: pendingUser });
     return msgs;
   }
   /**
-   * Signal the currently-running {@link step} that the user wants to
-   * stop exploring. Takes effect at the next iteration boundary — if a
-   * tool call is mid-flight it will be allowed to finish, then the
-   * loop diverts to the forced-summary path so the user gets an
-   * answer instead of a cliff. Called by the TUI on Esc.
+   * Signal the currently-running {@link step} to stop **now**. Cancels
+   * the in-flight network request (DeepSeek HTTP/SSE) AND any tool call
+   * currently dispatching (MCP `notifications/cancelled` + promise
+   * reject). The loop itself also sees `signal.aborted` at each
+   * iteration boundary and exits quickly instead of looping again.
+   * Called by the TUI on Esc.
    */
   abort() {
-    this._aborted = true;
+    this._turnAbort.abort();
   }
   /**
    * Drop everything in the log after (and including) the most recent
@@ -1317,13 +1401,14 @@ var CacheFirstLoop = class {
   async *step(userInput) {
     this._turn++;
     this.scratch.reset();
-    this._aborted = false;
+    this._turnAbort = new AbortController();
+    const signal = this._turnAbort.signal;
     let pendingUser = userInput;
     const toolSpecs = this.prefix.tools();
     const warnAt = Math.max(1, Math.floor(this.maxToolIters * 0.7));
     let warnedForIterBudget = false;
     for (let iter = 0; iter < this.maxToolIters; iter++) {
-      if (this._aborted) {
+      if (signal.aborted) {
         yield {
           turn: this._turn,
           role: "warning",
@@ -1340,6 +1425,13 @@ var CacheFirstLoop = class {
         yield { turn: this._turn, role: "done", content: stoppedMsg };
         return;
       }
+      if (iter > 0) {
+        yield {
+          turn: this._turn,
+          role: "status",
+          content: "tool result uploaded \xB7 model thinking before next response\u2026"
+        };
+      }
       if (!warnedForIterBudget && iter >= warnAt) {
         warnedForIterBudget = true;
         yield {
@@ -1386,7 +1478,8 @@ var CacheFirstLoop = class {
             {
               model: this.model,
               messages,
-              tools: toolSpecs.length ? toolSpecs : void 0
+              tools: toolSpecs.length ? toolSpecs : void 0,
+              signal
             },
             {
               ...this.branchOptions,
@@ -1395,8 +1488,8 @@ var CacheFirstLoop = class {
             }
           );
           for (let k = 0; k < budget; k++) {
-            const sample = queue.shift() ?? await new Promise((resolve3) => {
-              waiter = resolve3;
+            const sample = queue.shift() ?? await new Promise((resolve4) => {
+              waiter = resolve4;
             });
             yield {
               turn: this._turn,
@@ -1436,7 +1529,8 @@ var CacheFirstLoop = class {
           for await (const chunk of this.client.stream({
             model: this.model,
             messages,
-            tools: toolSpecs.length ? toolSpecs : void 0
+            tools: toolSpecs.length ? toolSpecs : void 0,
+            signal
           })) {
             if (chunk.contentDelta) {
               assistantContent += chunk.contentDelta;
@@ -1475,7 +1569,8 @@ var CacheFirstLoop = class {
           const resp = await this.client.chat({
             model: this.model,
             messages,
-            tools: toolSpecs.length ? toolSpecs : void 0
+            tools: toolSpecs.length ? toolSpecs : void 0,
+            signal
           });
           assistantContent = resp.content;
           reasoningContent = resp.reasoningContent ?? "";
@@ -1497,7 +1592,14 @@ var CacheFirstLoop = class {
         pendingUser = null;
       }
       this.scratch.reasoning = reasoningContent || null;
-      const planState = preHarvestedPlanState ? preHarvestedPlanState : this.harvestEnabled ? await harvest(reasoningContent || null, this.client, this.harvestOptions) : emptyPlanState();
+      if (!preHarvestedPlanState && this.harvestEnabled && (reasoningContent?.trim().length ?? 0) >= 40) {
+        yield {
+          turn: this._turn,
+          role: "status",
+          content: "extracting plan state from reasoning\u2026"
+        };
+      }
+      const planState = preHarvestedPlanState ? preHarvestedPlanState : this.harvestEnabled ? await harvest(reasoningContent || null, this.client, this.harvestOptions, signal) : emptyPlanState();
       const { calls: repairedCalls, report } = this.repair.process(
         toolCalls,
         reasoningContent || null,
@@ -1519,15 +1621,38 @@ var CacheFirstLoop = class {
       }
       const ctxMax = DEEPSEEK_CONTEXT_TOKENS[this.model] ?? DEFAULT_CONTEXT_TOKENS;
       if (usage && usage.promptTokens / ctxMax > 0.8) {
-        yield {
-          turn: this._turn,
-          role: "warning",
-          content: `context ${usage.promptTokens}/${ctxMax} (${Math.round(
-            usage.promptTokens / ctxMax * 100
-          )}%) \u2014 more tools would overflow. Forcing summary from what was gathered.`
-        };
-        yield* this.forceSummaryAfterIterLimit({ reason: "context-guard" });
-        return;
+        const before = usage.promptTokens;
+        const compactResult = this.compact(4e3);
+        if (compactResult.healedCount > 0) {
+          const approxSaved = Math.round(compactResult.charsSaved / 4);
+          const after = before - approxSaved;
+          yield {
+            turn: this._turn,
+            role: "warning",
+            content: `context ${before.toLocaleString()}/${ctxMax.toLocaleString()} \u2014 auto-compacted ${compactResult.healedCount} oversized tool result(s), saved ~${approxSaved.toLocaleString()} tokens (now ~${after.toLocaleString()}). Continuing.`
+          };
+        } else {
+          yield {
+            turn: this._turn,
+            role: "warning",
+            content: `context ${before.toLocaleString()}/${ctxMax.toLocaleString()} (${Math.round(
+              before / ctxMax * 100
+            )}%) \u2014 nothing to auto-compact. Forcing summary from what was gathered.`
+          };
+          const tail = this.log.entries[this.log.entries.length - 1];
+          if (tail && tail.role === "assistant" && Array.isArray(tail.tool_calls) && tail.tool_calls.length > 0) {
+            const kept = this.log.entries.slice(0, -1);
+            this.log.compactInPlace([...kept]);
+            if (this.sessionName) {
+              try {
+                rewriteSession(this.sessionName, kept);
+              } catch {
+              }
+            }
+          }
+          yield* this.forceSummaryAfterIterLimit({ reason: "context-guard" });
+          return;
+        }
       }
       for (const call of repairedCalls) {
         const name = call.function?.name ?? "";
@@ -1539,7 +1664,7 @@ var CacheFirstLoop = class {
           toolName: name,
           toolArgs: args
         };
-        const result = await this.tools.dispatch(name, args);
+        const result = await this.tools.dispatch(name, args, { signal });
         this.appendAndPersist({
           role: "tool",
           tool_call_id: call.id ?? "",
@@ -1559,6 +1684,11 @@ var CacheFirstLoop = class {
   }
   async *forceSummaryAfterIterLimit(opts = { reason: "budget" }) {
     try {
+      yield {
+        turn: this._turn,
+        role: "status",
+        content: "summarizing what was gathered\u2026"
+      };
       const messages = this.buildMessages(null);
       messages.push({
         role: "user",
@@ -1566,8 +1696,9 @@ var CacheFirstLoop = class {
       });
       const resp = await this.client.chat({
         model: this.model,
-        messages
+        messages,
         // no tools → model is forced to answer in text
+        signal: this._turnAbort.signal
       });
       const rawContent = resp.content?.trim() ?? "";
       const cleaned = stripHallucinatedToolMarkup(rawContent);
@@ -1640,7 +1771,7 @@ function summarizeBranch(chosen, samples) {
     temperatures: samples.map((s) => s.temperature)
   };
 }
-function healLoadedMessages(messages, maxChars) {
+function shrinkOversizedToolResults(messages, maxChars) {
   let healedCount = 0;
   let healedFrom = 0;
   const out = messages.map((msg) => {
@@ -1653,6 +1784,51 @@ function healLoadedMessages(messages, maxChars) {
   });
   return { messages: out, healedCount, healedFrom };
 }
+function healLoadedMessages(messages, maxChars) {
+  const shrunk = shrinkOversizedToolResults(messages, maxChars);
+  let healedCount = shrunk.healedCount;
+  const out = [];
+  const openCallIds = /* @__PURE__ */ new Set();
+  let droppedAssistantCalls = 0;
+  let droppedStrayTools = 0;
+  for (let i = 0; i < shrunk.messages.length; i++) {
+    const msg = shrunk.messages[i];
+    if (msg.role === "assistant" && Array.isArray(msg.tool_calls) && msg.tool_calls.length > 0) {
+      const needed = /* @__PURE__ */ new Set();
+      for (const call of msg.tool_calls) {
+        if (call?.id) needed.add(call.id);
+      }
+      const candidates = [];
+      let j = i + 1;
+      while (j < shrunk.messages.length && needed.size > 0) {
+        const nxt = shrunk.messages[j];
+        if (nxt.role !== "tool") break;
+        const id = nxt.tool_call_id ?? "";
+        if (!needed.has(id)) break;
+        needed.delete(id);
+        candidates.push(nxt);
+        j++;
+      }
+      if (needed.size === 0) {
+        out.push(msg);
+        for (const r of candidates) out.push(r);
+        i = j - 1;
+      } else {
+        droppedAssistantCalls += 1;
+        droppedStrayTools += candidates.length;
+        i = j - 1;
+      }
+      continue;
+    }
+    if (msg.role === "tool") {
+      droppedStrayTools += 1;
+      continue;
+    }
+    out.push(msg);
+  }
+  healedCount += droppedAssistantCalls + droppedStrayTools;
+  return { messages: out, healedCount, healedFrom: shrunk.healedFrom };
+}
 function formatLoopError(err) {
   const msg = err.message ?? "";
   if (msg.includes("maximum context length")) {
@@ -1663,13 +1839,348 @@ function formatLoopError(err) {
   return msg;
 }
 
+// src/tools/filesystem.ts
+import { promises as fs } from "fs";
+import * as pathMod from "path";
+var DEFAULT_MAX_READ_BYTES = 2 * 1024 * 1024;
+var DEFAULT_MAX_LIST_BYTES = 256 * 1024;
+function registerFilesystemTools(registry, opts) {
+  const rootDir = pathMod.resolve(opts.rootDir);
+  const allowWriting = opts.allowWriting !== false;
+  const maxReadBytes = opts.maxReadBytes ?? DEFAULT_MAX_READ_BYTES;
+  const maxListBytes = opts.maxListBytes ?? DEFAULT_MAX_LIST_BYTES;
+  const safePath = (raw) => {
+    if (typeof raw !== "string" || raw.length === 0) {
+      throw new Error("path must be a non-empty string");
+    }
+    const resolved = pathMod.resolve(rootDir, raw);
+    const normRoot = pathMod.resolve(rootDir);
+    const rel = pathMod.relative(normRoot, resolved);
+    if (rel.startsWith("..") || pathMod.isAbsolute(rel)) {
+      throw new Error(`path escapes sandbox root (${normRoot}): ${raw}`);
+    }
+    return resolved;
+  };
+  registry.register({
+    name: "read_file",
+    description: "Read a file under the sandbox root. Returns the full contents (truncated with a notice if larger than the per-call cap). Paths may be relative to the root or absolute-under-root.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string", description: "Path to read (relative to rootDir or absolute)." },
+        head: { type: "integer", description: "If set, return only the first N lines." },
+        tail: { type: "integer", description: "If set, return only the last N lines." }
+      },
+      required: ["path"]
+    },
+    fn: async (args) => {
+      const abs = safePath(args.path);
+      const stat = await fs.stat(abs);
+      if (stat.isDirectory()) {
+        throw new Error(`not a file: ${args.path} (it's a directory)`);
+      }
+      const raw = await fs.readFile(abs);
+      if (raw.length > maxReadBytes) {
+        const head = raw.slice(0, maxReadBytes).toString("utf8");
+        return `${head}
+
+[\u2026truncated ${raw.length - maxReadBytes} bytes \u2014 file is ${raw.length} B, cap ${maxReadBytes} B. Retry with head/tail for targeted view.]`;
+      }
+      const text = raw.toString("utf8");
+      if (typeof args.head === "number" && args.head > 0) {
+        return text.split(/\r?\n/).slice(0, args.head).join("\n");
+      }
+      if (typeof args.tail === "number" && args.tail > 0) {
+        let lines = text.split(/\r?\n/);
+        if (lines.length > 0 && lines[lines.length - 1] === "") lines = lines.slice(0, -1);
+        return lines.slice(Math.max(0, lines.length - args.tail)).join("\n");
+      }
+      return text;
+    }
+  });
+  registry.register({
+    name: "list_directory",
+    description: "List entries in a directory under the sandbox root. Returns one line per entry, marking directories with a trailing slash. Not recursive \u2014 use directory_tree for that.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string", description: "Directory to list (default: root)." }
+      }
+    },
+    fn: async (args) => {
+      const abs = safePath(args.path ?? ".");
+      const entries = await fs.readdir(abs, { withFileTypes: true });
+      const lines = [];
+      for (const e of entries.sort((a, b) => a.name.localeCompare(b.name))) {
+        lines.push(e.isDirectory() ? `${e.name}/` : e.name);
+      }
+      return lines.join("\n") || "(empty directory)";
+    }
+  });
+  registry.register({
+    name: "directory_tree",
+    description: "Recursively list entries in a directory. Shows indented tree structure with directories marked '/'. Caps output so a huge tree doesn't drown the context.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string", description: "Root of the tree (default: sandbox root)." },
+        maxDepth: { type: "integer", description: "Max recursion depth (default 4)." }
+      }
+    },
+    fn: async (args) => {
+      const startAbs = safePath(args.path ?? ".");
+      const maxDepth = typeof args.maxDepth === "number" ? args.maxDepth : 4;
+      const lines = [];
+      let totalBytes = 0;
+      let truncated = false;
+      const walk2 = async (dir, depth) => {
+        if (truncated) return;
+        if (depth > maxDepth) return;
+        let entries;
+        try {
+          entries = await fs.readdir(dir, { withFileTypes: true });
+        } catch {
+          return;
+        }
+        entries.sort((a, b) => a.name.localeCompare(b.name));
+        for (const e of entries) {
+          if (truncated) return;
+          const indent = "  ".repeat(depth);
+          const line = e.isDirectory() ? `${indent}${e.name}/` : `${indent}${e.name}`;
+          totalBytes += line.length + 1;
+          if (totalBytes > maxListBytes) {
+            lines.push(`  [\u2026 tree truncated at ${maxListBytes} bytes \u2026]`);
+            truncated = true;
+            return;
+          }
+          lines.push(line);
+          if (e.isDirectory()) {
+            await walk2(pathMod.join(dir, e.name), depth + 1);
+          }
+        }
+      };
+      await walk2(startAbs, 0);
+      return lines.join("\n") || "(empty tree)";
+    }
+  });
+  registry.register({
+    name: "search_files",
+    description: "Find files whose NAME matches a substring or regex. Case-insensitive. Walks the directory recursively under the sandbox root. Returns one path per line.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string", description: "Directory to start the search at (default: root)." },
+        pattern: {
+          type: "string",
+          description: "Substring (or regex) to match against filenames."
+        }
+      },
+      required: ["pattern"]
+    },
+    fn: async (args) => {
+      const startAbs = safePath(args.path ?? ".");
+      const needle = args.pattern.toLowerCase();
+      let re = null;
+      try {
+        re = new RegExp(args.pattern, "i");
+      } catch {
+        re = null;
+      }
+      const matches = [];
+      let totalBytes = 0;
+      const walk2 = async (dir) => {
+        let entries;
+        try {
+          entries = await fs.readdir(dir, { withFileTypes: true });
+        } catch {
+          return;
+        }
+        for (const e of entries) {
+          const full = pathMod.join(dir, e.name);
+          const lower = e.name.toLowerCase();
+          const hit = re ? re.test(e.name) : lower.includes(needle);
+          if (hit) {
+            const rel = pathMod.relative(rootDir, full);
+            if (totalBytes + rel.length + 1 > maxListBytes) {
+              matches.push("[\u2026 search truncated \u2014 refine pattern \u2026]");
+              return;
+            }
+            matches.push(rel);
+            totalBytes += rel.length + 1;
+          }
+          if (e.isDirectory()) await walk2(full);
+        }
+      };
+      await walk2(startAbs);
+      return matches.length === 0 ? "(no matches)" : matches.join("\n");
+    }
+  });
+  registry.register({
+    name: "get_file_info",
+    description: "Stat a path under the sandbox root. Returns type (file|directory|symlink), size in bytes, mtime in ISO-8601.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string" }
+      },
+      required: ["path"]
+    },
+    fn: async (args) => {
+      const abs = safePath(args.path);
+      const st = await fs.lstat(abs);
+      const type = st.isDirectory() ? "directory" : st.isSymbolicLink() ? "symlink" : "file";
+      return JSON.stringify({
+        type,
+        size: st.size,
+        mtime: st.mtime.toISOString()
+      });
+    }
+  });
+  if (!allowWriting) return registry;
+  registry.register({
+    name: "write_file",
+    description: "Create or overwrite a file under the sandbox root with the given content. Parent directories are created as needed.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string" },
+        content: { type: "string" }
+      },
+      required: ["path", "content"]
+    },
+    fn: async (args) => {
+      const abs = safePath(args.path);
+      await fs.mkdir(pathMod.dirname(abs), { recursive: true });
+      await fs.writeFile(abs, args.content, "utf8");
+      return `wrote ${args.content.length} chars to ${pathMod.relative(rootDir, abs)}`;
+    }
+  });
+  registry.register({
+    name: "edit_file",
+    description: "Apply a SEARCH/REPLACE edit to an existing file. `search` must match exactly (whitespace sensitive) \u2014 no regex. The match must be unique in the file; otherwise the edit is refused to avoid surprise rewrites. This flat-string shape replaces the `{oldText, newText}[]` JSON array form that previously triggered R1 DSML hallucinations.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string" },
+        search: { type: "string", description: "Exact text to find (must be unique)." },
+        replace: { type: "string", description: "Text to substitute in place of `search`." }
+      },
+      required: ["path", "search", "replace"]
+    },
+    fn: async (args) => {
+      const abs = safePath(args.path);
+      const before = await fs.readFile(abs, "utf8");
+      if (args.search.length === 0) {
+        throw new Error("edit_file: search cannot be empty");
+      }
+      const firstIdx = before.indexOf(args.search);
+      if (firstIdx < 0) {
+        throw new Error(`edit_file: search text not found in ${pathMod.relative(rootDir, abs)}`);
+      }
+      const nextIdx = before.indexOf(args.search, firstIdx + 1);
+      if (nextIdx >= 0) {
+        throw new Error(
+          `edit_file: search text appears multiple times in ${pathMod.relative(rootDir, abs)} \u2014 include more context to disambiguate`
+        );
+      }
+      const after = before.slice(0, firstIdx) + args.replace + before.slice(firstIdx + args.search.length);
+      await fs.writeFile(abs, after, "utf8");
+      const rel = pathMod.relative(rootDir, abs);
+      const header = `edited ${rel} (${args.search.length}\u2192${args.replace.length} chars)`;
+      const startLine = before.slice(0, firstIdx).split(/\r?\n/).length;
+      const diff = renderEditDiff(args.search, args.replace, startLine);
+      return `${header}
+${diff}`;
+    }
+  });
+  registry.register({
+    name: "create_directory",
+    description: "Create a directory (and any missing parents) under the sandbox root.",
+    parameters: {
+      type: "object",
+      properties: { path: { type: "string" } },
+      required: ["path"]
+    },
+    fn: async (args) => {
+      const abs = safePath(args.path);
+      await fs.mkdir(abs, { recursive: true });
+      return `created ${pathMod.relative(rootDir, abs)}/`;
+    }
+  });
+  registry.register({
+    name: "move_file",
+    description: "Rename/move a file or directory under the sandbox root.",
+    parameters: {
+      type: "object",
+      properties: {
+        source: { type: "string" },
+        destination: { type: "string" }
+      },
+      required: ["source", "destination"]
+    },
+    fn: async (args) => {
+      const src = safePath(args.source);
+      const dst = safePath(args.destination);
+      await fs.mkdir(pathMod.dirname(dst), { recursive: true });
+      await fs.rename(src, dst);
+      return `moved ${pathMod.relative(rootDir, src)} \u2192 ${pathMod.relative(rootDir, dst)}`;
+    }
+  });
+  return registry;
+}
+function renderEditDiff(search, replace, startLine) {
+  const a = search.split(/\r?\n/);
+  const b = replace.split(/\r?\n/);
+  const diff = lineDiff(a, b);
+  const hunk = `@@ -${startLine},${a.length} +${startLine},${b.length} @@`;
+  const body = diff.map((d) => `${d.op === " " ? " " : d.op} ${d.line}`).join("\n");
+  return `${hunk}
+${body}`;
+}
+function lineDiff(a, b) {
+  const n = a.length;
+  const m = b.length;
+  const dp = Array.from({ length: n + 1 }, () => new Array(m + 1).fill(0));
+  for (let i2 = 1; i2 <= n; i2++) {
+    for (let j2 = 1; j2 <= m; j2++) {
+      if (a[i2 - 1] === b[j2 - 1]) dp[i2][j2] = dp[i2 - 1][j2 - 1] + 1;
+      else dp[i2][j2] = Math.max(dp[i2 - 1][j2], dp[i2][j2 - 1]);
+    }
+  }
+  const out = [];
+  let i = n;
+  let j = m;
+  while (i > 0 && j > 0) {
+    if (a[i - 1] === b[j - 1]) {
+      out.unshift({ op: " ", line: a[i - 1] });
+      i--;
+      j--;
+    } else if ((dp[i - 1][j] ?? 0) > (dp[i][j - 1] ?? 0)) {
+      out.unshift({ op: "-", line: a[i - 1] });
+      i--;
+    } else {
+      out.unshift({ op: "+", line: b[j - 1] });
+      j--;
+    }
+  }
+  while (i > 0) {
+    out.unshift({ op: "-", line: a[i - 1] });
+    i--;
+  }
+  while (j > 0) {
+    out.unshift({ op: "+", line: b[j - 1] });
+    j--;
+  }
+  return out;
+}
+
 // src/env.ts
 import { readFileSync as readFileSync2 } from "fs";
-import { resolve } from "path";
+import { resolve as resolve2 } from "path";
 function loadDotenv(path = ".env") {
   let raw;
   try {
-    raw = readFileSync2(resolve(process.cwd(), path), "utf8");
+    raw = readFileSync2(resolve2(process.cwd(), path), "utf8");
   } catch {
     return;
   }
@@ -1829,6 +2340,8 @@ function computeReplayStats(records) {
 }
 function summarizeTurns(turns) {
   const totalCost = turns.reduce((s, t) => s + t.cost, 0);
+  const totalInput = turns.reduce((s, t) => s + inputCostUsd(t.model, t.usage), 0);
+  const totalOutput = turns.reduce((s, t) => s + outputCostUsd(t.model, t.usage), 0);
   const totalClaude = turns.reduce((s, t) => s + claudeEquivalentCost(t.usage), 0);
   let hit = 0;
   let miss = 0;
@@ -1842,6 +2355,8 @@ function summarizeTurns(turns) {
   return {
     turns: turns.length,
     totalCostUsd: round2(totalCost, 6),
+    totalInputCostUsd: round2(totalInput, 6),
+    totalOutputCostUsd: round2(totalOutput, 6),
     claudeEquivalentUsd: round2(totalClaude, 6),
     savingsVsClaudePct: round2(savingsVsClaude * 100, 2),
     cacheHitRatio: round2(cacheHitRatio, 4),
@@ -2199,6 +2714,13 @@ var McpClient = class {
   _serverInfo = { name: "", version: "" };
   _protocolVersion = "";
   _instructions;
+  // Progress-token → handler for notifications/progress routing. Tokens
+  // are minted per call when the caller supplies an onProgress
+  // callback; cleared when the final response lands (or the pending
+  // request rejects). No leaks — the `try/finally` in callTool
+  // guarantees cleanup even on timeout.
+  progressHandlers = /* @__PURE__ */ new Map();
+  nextProgressToken = 1;
   constructor(opts) {
     this.transport = opts.transport;
     this.clientInfo = opts.clientInfo ?? { name: "reasonix", version: "0.3.0-dev" };
@@ -2253,13 +2775,36 @@ var McpClient = class {
     this.assertInitialized();
     return this.request("tools/list", {});
   }
-  /** Invoke a tool by name. Returns the raw MCP result (caller unwraps content). */
-  async callTool(name, args) {
+  /**
+   * Invoke a tool by name. When `onProgress` is supplied, attaches a
+   * fresh progress token so the server can send incremental updates
+   * via `notifications/progress`; they're routed to the callback until
+   * the final response arrives (or the request times out, in which
+   * case the handler is simply dropped — no extra notification).
+   *
+   * When `signal` is supplied, aborting it:
+   *   1) fires `notifications/cancelled` to the server (MCP 2024-11-05
+   *      way of saying "forget this request, I no longer care"), and
+   *   2) rejects the pending promise immediately with an AbortError,
+   *      so the caller doesn't have to wait for the subprocess to
+   *      finish its in-flight file write or network request.
+   * The server MAY still emit a late response; we drop it in dispatch
+   * since the request id is gone from `pending`.
+   */
+  async callTool(name, args, opts = {}) {
     this.assertInitialized();
-    return this.request("tools/call", {
-      name,
-      arguments: args ?? {}
-    });
+    const params = { name, arguments: args ?? {} };
+    let token;
+    if (opts.onProgress) {
+      token = this.nextProgressToken++;
+      this.progressHandlers.set(token, opts.onProgress);
+      params._meta = { progressToken: token };
+    }
+    try {
+      return await this.request("tools/call", params, opts.signal);
+    } finally {
+      if (token !== void 0) this.progressHandlers.delete(token);
+    }
   }
   /**
    * List resources the server exposes. Supports a pagination cursor;
@@ -2313,24 +2858,56 @@ var McpClient = class {
   assertInitialized() {
     if (!this.initialized) throw new Error("MCP client not initialized \u2014 call initialize() first");
   }
-  async request(method, params) {
+  async request(method, params, signal) {
     const id = this.nextId++;
     const frame = { jsonrpc: "2.0", id, method, params };
-    const promise = new Promise((resolve3, reject) => {
+    let abortHandler = null;
+    const promise = new Promise((resolve4, reject) => {
       const timeout = setTimeout(() => {
         this.pending.delete(id);
+        if (abortHandler && signal) signal.removeEventListener("abort", abortHandler);
         reject(
           new Error(`MCP request ${method} (id=${id}) timed out after ${this.requestTimeoutMs}ms`)
         );
       }, this.requestTimeoutMs);
       this.pending.set(id, {
-        resolve: resolve3,
+        resolve: resolve4,
         reject,
         timeout
       });
+      if (signal) {
+        if (signal.aborted) {
+          this.pending.delete(id);
+          clearTimeout(timeout);
+          reject(new Error(`MCP request ${method} (id=${id}) aborted before send`));
+          return;
+        }
+        abortHandler = () => {
+          this.pending.delete(id);
+          clearTimeout(timeout);
+          void this.transport.send({
+            jsonrpc: "2.0",
+            method: "notifications/cancelled",
+            params: { requestId: id, reason: "aborted by user" }
+          }).catch(() => {
+          });
+          reject(new Error(`MCP request ${method} (id=${id}) aborted by user`));
+        };
+        signal.addEventListener("abort", abortHandler, { once: true });
+      }
     });
-    await this.transport.send(frame);
-    return promise;
+    try {
+      await this.transport.send(frame);
+    } catch (err) {
+      this.pending.delete(id);
+      if (abortHandler && signal) signal.removeEventListener("abort", abortHandler);
+      throw err;
+    }
+    try {
+      return await promise;
+    } finally {
+      if (abortHandler && signal) signal.removeEventListener("abort", abortHandler);
+    }
   }
   startReaderIfNeeded() {
     if (this.readerStarted) return;
@@ -2351,7 +2928,16 @@ var McpClient = class {
     }
   }
   dispatch(msg) {
-    if (!("id" in msg) || msg.id === null || msg.id === void 0) return;
+    if (!("id" in msg) || msg.id === null || msg.id === void 0) {
+      if ("method" in msg && msg.method === "notifications/progress") {
+        const p = msg.params;
+        if (!p || p.progressToken === void 0) return;
+        const handler = this.progressHandlers.get(p.progressToken);
+        if (!handler) return;
+        handler({ progress: p.progress, total: p.total, message: p.message });
+      }
+      return;
+    }
     if (!("result" in msg) && !("error" in msg)) return;
     const pending = this.pending.get(msg.id);
     if (!pending) return;
@@ -2408,12 +2994,12 @@ var StdioTransport = class {
   }
   async send(message) {
     if (this.closed) throw new Error("MCP transport is closed");
-    return new Promise((resolve3, reject) => {
+    return new Promise((resolve4, reject) => {
       const line = `${JSON.stringify(message)}
 `;
       this.child.stdin.write(line, "utf8", (err) => {
         if (err) reject(err);
-        else resolve3();
+        else resolve4();
       });
     });
   }
@@ -2424,8 +3010,8 @@ var StdioTransport = class {
         continue;
       }
       if (this.closed) return;
-      const next = await new Promise((resolve3) => {
-        this.waiters.push(resolve3);
+      const next = await new Promise((resolve4) => {
+        this.waiters.push(resolve4);
       });
       if (next === null) return;
       yield next;
@@ -2491,8 +3077,8 @@ var SseTransport = class {
   constructor(opts) {
     this.url = opts.url;
     this.headers = opts.headers ?? {};
-    this.endpointReady = new Promise((resolve3, reject) => {
-      this.resolveEndpoint = resolve3;
+    this.endpointReady = new Promise((resolve4, reject) => {
+      this.resolveEndpoint = resolve4;
       this.rejectEndpoint = reject;
     });
     this.endpointReady.catch(() => void 0);
@@ -2519,8 +3105,8 @@ var SseTransport = class {
         continue;
       }
       if (this.closed) return;
-      const next = await new Promise((resolve3) => {
-        this.waiters.push(resolve3);
+      const next = await new Promise((resolve4) => {
+        this.waiters.push(resolve4);
       });
       if (next === null) return;
       yield next;
@@ -2720,7 +3306,7 @@ async function trySection(load) {
 
 // src/code/edit-blocks.ts
 import { existsSync as existsSync2, mkdirSync as mkdirSync2, readFileSync as readFileSync4, unlinkSync as unlinkSync2, writeFileSync as writeFileSync2 } from "fs";
-import { dirname as dirname2, resolve as resolve2 } from "path";
+import { dirname as dirname3, resolve as resolve3 } from "path";
 var BLOCK_RE = /^(\S[^\n]*)\n<{7} SEARCH\n([\s\S]*?)\n?={7}\n([\s\S]*?)\n?>{7} REPLACE/gm;
 function parseEditBlocks(text) {
   const out = [];
@@ -2738,8 +3324,8 @@ function parseEditBlocks(text) {
   return out;
 }
 function applyEditBlock(block, rootDir) {
-  const absRoot = resolve2(rootDir);
-  const absTarget = resolve2(absRoot, block.path);
+  const absRoot = resolve3(rootDir);
+  const absTarget = resolve3(absRoot, block.path);
   if (absTarget !== absRoot && !absTarget.startsWith(`${absRoot}${sep()}`)) {
     return {
       path: block.path,
@@ -2758,7 +3344,7 @@ function applyEditBlock(block, rootDir) {
           message: "file does not exist; to create it, use an empty SEARCH block"
         };
       }
-      mkdirSync2(dirname2(absTarget), { recursive: true });
+      mkdirSync2(dirname3(absTarget), { recursive: true });
       writeFileSync2(absTarget, block.replace, "utf8");
       return { path: block.path, status: "created" };
     }
@@ -2789,13 +3375,13 @@ function applyEditBlocks(blocks, rootDir) {
   return blocks.map((b) => applyEditBlock(b, rootDir));
 }
 function snapshotBeforeEdits(blocks, rootDir) {
-  const absRoot = resolve2(rootDir);
+  const absRoot = resolve3(rootDir);
   const seen = /* @__PURE__ */ new Set();
   const snapshots = [];
   for (const b of blocks) {
     if (seen.has(b.path)) continue;
     seen.add(b.path);
-    const abs = resolve2(absRoot, b.path);
+    const abs = resolve3(absRoot, b.path);
     if (!existsSync2(abs)) {
       snapshots.push({ path: b.path, prevContent: null });
       continue;
@@ -2809,9 +3395,9 @@ function snapshotBeforeEdits(blocks, rootDir) {
   return snapshots;
 }
 function restoreSnapshots(snapshots, rootDir) {
-  const absRoot = resolve2(rootDir);
+  const absRoot = resolve3(rootDir);
   return snapshots.map((snap) => {
-    const abs = resolve2(absRoot, snap.path);
+    const abs = resolve3(absRoot, snap.path);
     if (abs !== absRoot && !abs.startsWith(`${absRoot}${sep()}`)) {
       return {
         path: snap.path,
@@ -2845,7 +3431,7 @@ function sep() {
 
 // src/code/prompt.ts
 import { existsSync as existsSync3, readFileSync as readFileSync5 } from "fs";
-import { join as join2 } from "path";
+import { join as join3 } from "path";
 var CODE_SYSTEM_PROMPT = `You are Reasonix Code, a coding assistant. You have filesystem tools (read_file, write_file, list_directory, search_files, etc.) rooted at the user's working directory.
 
 # When to edit vs. when to explore
@@ -2894,7 +3480,7 @@ Rules:
 - If you need to explore first (list / grep / read), do it with tool calls before writing any prose \u2014 silence while exploring is fine.
 `;
 function codeSystemPrompt(rootDir) {
-  const gitignorePath = join2(rootDir, ".gitignore");
+  const gitignorePath = join3(rootDir, ".gitignore");
   if (!existsSync3(gitignorePath)) return CODE_SYSTEM_PROMPT;
   let content;
   try {
@@ -2920,9 +3506,9 @@ ${truncated}
 // src/config.ts
 import { chmodSync as chmodSync2, mkdirSync as mkdirSync3, readFileSync as readFileSync6, writeFileSync as writeFileSync3 } from "fs";
 import { homedir as homedir2 } from "os";
-import { dirname as dirname3, join as join3 } from "path";
+import { dirname as dirname4, join as join4 } from "path";
 function defaultConfigPath() {
-  return join3(homedir2(), ".reasonix", "config.json");
+  return join4(homedir2(), ".reasonix", "config.json");
 }
 function readConfig(path = defaultConfigPath()) {
   try {
@@ -2934,7 +3520,7 @@ function readConfig(path = defaultConfigPath()) {
   return {};
 }
 function writeConfig(cfg, path = defaultConfigPath()) {
-  mkdirSync3(dirname3(path), { recursive: true });
+  mkdirSync3(dirname4(path), { recursive: true });
   writeFileSync3(path, JSON.stringify(cfg, null, 2), "utf8");
   try {
     chmodSync2(path, 384);
@@ -3001,6 +3587,7 @@ export {
   formatLoopError,
   harvest,
   healLoadedMessages,
+  inputCostUsd,
   inspectMcpServer,
   isJsonRpcError,
   isPlanStateEmpty,
@@ -3011,6 +3598,7 @@ export {
   loadSessionMessages,
   nestArguments,
   openTranscriptFile,
+  outputCostUsd,
   parseEditBlocks,
   parseMcpSpec,
   parseTranscript,
@@ -3018,6 +3606,7 @@ export {
   readTranscript,
   recordFromLoopEvent,
   redactKey,
+  registerFilesystemTools,
   renderMarkdown as renderDiffMarkdown,
   renderSummaryTable as renderDiffSummary,
   repairTruncatedJson,