reasonix 0.37.0 → 0.39.0

package/README.md

@@ -162,6 +162,7 @@ For live cache-hit rates, costs, and methodology, see [`benchmarks/`](./benchmar
 ## Documentation
 
 - [**Architecture**](./docs/ARCHITECTURE.md) — three pillars: cache-first loop, tool-call repair, cost control
+- [**CLI Reference**](./docs/CLI-REFERENCE.md) — every shell subcommand, every slash command, every keybinding
 - [**Benchmarks**](./benchmarks/) — τ-bench-lite harness, transcripts, cost methodology
 - [**Website**](https://esengine.github.io/DeepSeek-Reasonix/) — getting started, dashboard mockup, TUI mockup
 - [**Contributing**](./CONTRIBUTING.md) — comment policy, error-handling rules, library-over-hand-rolled

package/README.zh-CN.md

@@ -157,6 +157,7 @@ npx reasonix code --dir /path/to/project
 ## 文档
 
 - [**架构**](./docs/ARCHITECTURE.md) —— 四大支柱、缓存优先循环、思维提取、脚手架
+- [**CLI 参考**](./docs/CLI-REFERENCE.md) —— 每个 shell 子命令、每个 slash 命令、每个快捷键
 - [**基准测试**](./benchmarks/) —— τ-bench-lite harness、transcript、成本方法论
 - [**官方网站**](https://esengine.github.io/DeepSeek-Reasonix/) —— 入门、Dashboard 设计稿、TUI 设计稿
 - [**贡献指南**](./CONTRIBUTING.md) —— 注释规则、错误处理、用现成库不手写

package/dist/cli/{chat-7257YAPG.js → chat-QCY6CH7O.js}

@@ -1,42 +1,42 @@
 #!/usr/bin/env node
 import {
   chatCommand
-} from "./chunk-MSKUP6PD.js";
+} from "./chunk-TPK2CHWR.js";
 import "./chunk-BQNUJJN7.js";
-import "./chunk-RFX7TYVV.js";
+import "./chunk-7G3SESEU.js";
 import "./chunk-MRLXEMZ7.js";
 import "./chunk-CPOV2O73.js";
-import "./chunk-SEFXUF24.js";
-import "./chunk-UNMYFZPZ.js";
+import "./chunk-5ZCRXN7S.js";
+import "./chunk-SJNIIH5W.js";
 import "./chunk-XJLZ4HKU.js";
 import "./chunk-XHQIK7B6.js";
-import "./chunk-YER7WCHF.js";
-import "./chunk-T52GAWPP.js";
-import "./chunk-ZJR4QLXB.js";
+import "./chunk-DDA76P44.js";
+import "./chunk-NLV2YORE.js";
+import "./chunk-SUZRC4NC.js";
 import "./chunk-MHDNZXJJ.js";
-import "./chunk-JULZ7JTO.js";
+import "./chunk-TGO7X47P.js";
+import "./chunk-AFFZF3MW.js";
 import "./chunk-DAEAAVDF.js";
 import "./chunk-KMWKGPFZ.js";
 import "./chunk-3Q3C4W66.js";
 import "./chunk-4DCHFFEY.js";
 import "./chunk-WJ3YX4PZ.js";
-import "./chunk-VF57YX2M.js";
+import "./chunk-TPDWAMG6.js";
 import "./chunk-SOZE7V7V.js";
 import "./chunk-6NMWJSES.js";
-import "./chunk-S4GF3HPO.js";
-import "./chunk-XOIDSPMQ.js";
+import "./chunk-NTVW2TWO.js";
+import "./chunk-6DR4F3MC.js";
 import "./chunk-FM57FNPJ.js";
-import "./chunk-XQIFIB3U.js";
-import "./chunk-JGZKTAOH.js";
+import "./chunk-4D662BWT.js";
+import "./chunk-AKDDHHE6.js";
 import "./chunk-5X7LZJDE.js";
 import "./chunk-6CXT5JRM.js";
-import "./chunk-GKZJXYMY.js";
-import "./chunk-BHLHOS5Y.js";
-import "./chunk-WUI3P4RA.js";
+import "./chunk-AJGLCSZS.js";
+import "./chunk-SWLIVNTP.js";
 import "./chunk-ZTLZO42A.js";
 import "./chunk-ORM6PK57.js";
 import "./chunk-CRPQUBP6.js";
 export {
   chatCommand
 };
-//# sourceMappingURL=chat-7257YAPG.js.map
+//# sourceMappingURL=chat-QCY6CH7O.js.map

package/dist/cli/{chunk-XQIFIB3U.js → chunk-4D662BWT.js}

@@ -7,7 +7,7 @@ import {
   compileFilters,
   defaultIndexConfig,
   resolveSemanticEmbeddingConfig
-} from "./chunk-BHLHOS5Y.js";
+} from "./chunk-SWLIVNTP.js";
 
 // src/index/semantic/builder.ts
 import { promises as fs3 } from "fs";
@@ -948,4 +948,4 @@ export {
   startOllamaDaemon,
   pullOllamaModel
 };
-//# sourceMappingURL=chunk-XQIFIB3U.js.map
+//# sourceMappingURL=chunk-4D662BWT.js.map

package/dist/cli/{chunk-SEFXUF24.js → chunk-5ZCRXN7S.js}

@@ -2,7 +2,7 @@
 import {
   MemoryStore,
   sanitizeMemoryName
-} from "./chunk-YER7WCHF.js";
+} from "./chunk-DDA76P44.js";
 import {
   countTokens,
   estimateConversationTokens,
@@ -13,15 +13,15 @@ import {
 } from "./chunk-KMWKGPFZ.js";
 import {
   pauseGate
-} from "./chunk-S4GF3HPO.js";
+} from "./chunk-NTVW2TWO.js";
 import {
   NEGATIVE_CLAIM_RULE,
   TUI_FORMATTING_RULES
-} from "./chunk-XOIDSPMQ.js";
+} from "./chunk-6DR4F3MC.js";
 import {
   formatHookOutcomeMessage,
   runHooks
-} from "./chunk-JGZKTAOH.js";
+} from "./chunk-AKDDHHE6.js";
 import {
   ignoredByLayers,
   loadGitignoreAt,
@@ -36,12 +36,12 @@ import {
 } from "./chunk-6CXT5JRM.js";
 import {
   t
-} from "./chunk-GKZJXYMY.js";
+} from "./chunk-AJGLCSZS.js";
 import {
   DEFAULT_INDEX_EXCLUDES,
   webSearchEndpoint,
   webSearchEngine
-} from "./chunk-BHLHOS5Y.js";
+} from "./chunk-SWLIVNTP.js";
 import {
   DEEPSEEK_CONTEXT_TOKENS,
   DEFAULT_CONTEXT_TOKENS,
@@ -287,6 +287,8 @@ var ToolRegistry = class {
   _interceptor = null;
   _auditListener = null;
   _resultAugmenter = null;
+  /** Per-tool fingerprint of the last call that failed schema validation. Cleared by any successful validation for that tool. */
+  _lastMalformed = /* @__PURE__ */ new Map();
   constructor(opts = {}) {
     this._autoFlatten = opts.autoFlatten !== false;
   }
@@ -309,6 +311,10 @@ var ToolRegistry = class {
   setResultAugmenter(fn) {
     this._resultAugmenter = fn;
   }
+  /** True when an augmenter is already wired — lets late-installing callers skip clobbering an earlier one. */
+  get hasResultAugmenter() {
+    return this._resultAugmenter !== null;
+  }
   register(def) {
     if (!def.name) throw new Error("tool requires a name");
     const internal = { ...def };
@@ -357,17 +363,29 @@ var ToolRegistry = class {
     if (!tool) {
       return JSON.stringify({ error: `unknown tool: ${name}` });
     }
+    const fingerprint = fingerprintArgs(argumentsRaw);
     let args;
     try {
       args = typeof argumentsRaw === "string" ? argumentsRaw.trim() ? JSON.parse(argumentsRaw) ?? {} : {} : argumentsRaw ?? {};
     } catch (err) {
-      return JSON.stringify({
-        error: `invalid tool arguments JSON: ${err.message}`
-      });
+      return this._noteMalformed(
+        name,
+        fingerprint,
+        `invalid tool arguments JSON: ${err.message}`
+      );
     }
     if (tool.flatSchema && args && typeof args === "object" && hasDotKey(args)) {
       args = nestArguments(args);
     }
+    const missing = tool.parameters ? missingRequiredParam(tool.parameters, args) : null;
+    if (missing) {
+      return this._noteMalformed(
+        name,
+        fingerprint,
+        `missing required parameter "${missing}". Retry with all required parameters filled.`
+      );
+    }
+    this._lastMalformed.delete(name);
     if (this._planMode && !isReadOnlyCall(tool, args)) {
       return JSON.stringify({
         error: `${name}: unavailable in plan mode \u2014 this is a read-only exploration phase. Use read_file / list_directory / search_files / directory_tree / web_search / allowlisted shell commands to investigate. Call submit_plan with your proposed plan when you're ready for the user's review.`,
@@ -423,6 +441,18 @@ var ToolRegistry = class {
     }
     return finalResult;
   }
+  /** Records the failed call's fingerprint; on the 2nd consecutive identical malformed call to the same tool, returns a sharper error that tells the model to stop retrying. */
+  _noteMalformed(name, fingerprint, detail) {
+    const prev = this._lastMalformed.get(name);
+    this._lastMalformed.set(name, fingerprint);
+    if (prev === fingerprint) {
+      return JSON.stringify({
+        error: `${name}: same call just failed validation (${detail}) \u2014 DO NOT retry with identical args. Either fix the call (read the schema in the tool spec) or pick a different tool.`,
+        consecutiveMalformed: true
+      });
+    }
+    return JSON.stringify({ error: `${name}: ${detail}` });
+  }
 };
 function isReadOnlyCall(tool, args) {
   if (tool.readOnlyCheck) {
@@ -440,6 +470,22 @@ function hasDotKey(obj) {
   }
   return false;
 }
+function fingerprintArgs(argumentsRaw) {
+  if (typeof argumentsRaw === "string") return argumentsRaw;
+  try {
+    return JSON.stringify(argumentsRaw);
+  } catch {
+    return "";
+  }
+}
+function missingRequiredParam(schema, args) {
+  const required = schema.required;
+  if (!required || required.length === 0) return null;
+  for (const key of required) {
+    if (args[key] === void 0) return key;
+  }
+  return null;
+}
 
 // src/memory/runtime.ts
 import { createHash } from "crypto";
@@ -1375,6 +1421,7 @@ function signature(call) {
 
 // src/loop.ts
 var ESCALATION_MODEL = "deepseek-v4-pro";
+var PARENT_BUDGET_WARN_THRESHOLD = 5;
 var CacheFirstLoop = class {
   client;
   prefix;
@@ -1411,6 +1458,7 @@ var CacheFirstLoop = class {
   _turnFailures = new TurnFailureTracker();
   _turnSelfCorrected = false;
   _foldedThisTurn = false;
+  _toolDispatchesThisStep = 0;
   context;
   /** Subscribe API so UI hooks can derive `running` from finally-guaranteed insertions. */
   get inflight() {
@@ -1465,6 +1513,23 @@ var CacheFirstLoop = class {
       stormThreshold: parsePositiveIntEnv(process.env.REASONIX_STORM_THRESHOLD),
       stormWindow: parsePositiveIntEnv(process.env.REASONIX_STORM_WINDOW)
     });
+    if (!this.tools.hasResultAugmenter) {
+      this.tools.setResultAugmenter((_name, _args, result) => {
+        this._toolDispatchesThisStep++;
+        const remaining = this.maxToolIters - this._toolDispatchesThisStep;
+        if (remaining <= 0) {
+          return `${result}
+
+[budget: 0 of ${this.maxToolIters} tool calls left this turn \u2014 finalize NOW; the next iter forces a summary]`;
+        }
+        if (remaining <= PARENT_BUDGET_WARN_THRESHOLD) {
+          return `${result}
+
+[budget: ${remaining} of ${this.maxToolIters} tool calls left this turn \u2014 wrap up soon]`;
+        }
+        return result;
+      });
+    }
     this.sessionName = opts.session ?? null;
     if (this.sessionName) {
       const prior = loadSessionMessages(this.sessionName);
@@ -1718,6 +1783,7 @@ ${reason}`
     this._turnSelfCorrected = false;
     this._escalateThisTurn = false;
     this._foldedThisTurn = false;
+    this._toolDispatchesThisStep = 0;
     let armedConsumed = false;
     if (this._proArmedForNextTurn) {
       this._escalateThisTurn = true;
@@ -3129,6 +3195,74 @@ Prefer \`list_directory\` for a single-level view, \`search_files\` to find spec
       return `moved ${displayRel4(rootDir, src)} \u2192 ${displayRel4(rootDir, dst)}`;
     }
   });
+  registry.register({
+    name: "delete_file",
+    description: "Delete one file under the sandbox root. Refuses directories \u2014 use delete_directory for those. Errors if the path doesn't exist.",
+    parameters: {
+      type: "object",
+      properties: { path: { type: "string" } },
+      required: ["path"]
+    },
+    fn: async (args) => {
+      const abs = safePath(args.path);
+      const st = await fs4.lstat(abs);
+      if (st.isDirectory()) {
+        throw new Error(
+          `delete_file: ${args.path} is a directory \u2014 use delete_directory to remove it`
+        );
+      }
+      await fs4.unlink(abs);
+      return `deleted ${displayRel4(rootDir, abs)}`;
+    }
+  });
+  registry.register({
+    name: "delete_directory",
+    description: "Recursively delete a directory under the sandbox root. Pass `recursive:false` to refuse non-empty directories. Errors if the path doesn't exist.",
+    parameters: {
+      type: "object",
+      properties: {
+        path: { type: "string" },
+        recursive: {
+          type: "boolean",
+          description: "When true (default) deletes the directory and all its contents. When false, only removes empty directories \u2014 non-empty refuses with an error."
+        }
+      },
+      required: ["path"]
+    },
+    fn: async (args) => {
+      const abs = safePath(args.path);
+      const st = await fs4.lstat(abs);
+      if (!st.isDirectory()) {
+        throw new Error(`delete_directory: ${args.path} is a file \u2014 use delete_file to remove it`);
+      }
+      const recursive = args.recursive !== false;
+      if (recursive) {
+        await fs4.rm(abs, { recursive: true, force: false });
+      } else {
+        await fs4.rmdir(abs);
+      }
+      return `deleted ${displayRel4(rootDir, abs)}/${recursive ? " (recursive)" : ""}`;
+    }
+  });
+  registry.register({
+    name: "copy_file",
+    description: "Copy a file or directory under the sandbox root. Both source and destination resolve under the sandbox. Parent directories of the destination are created as needed. Refuses to overwrite an existing destination \u2014 delete it first if you want to replace it.",
+    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 fs4.mkdir(pathMod4.dirname(dst), { recursive: true });
+      await fs4.cp(src, dst, { recursive: true, force: false, errorOnExist: true });
+      return `copied ${displayRel4(rootDir, src)} \u2192 ${displayRel4(rootDir, dst)}`;
+    }
+  });
   return registry;
 }
 
@@ -3343,7 +3477,7 @@ function registerChoiceTool(registry, opts = {}) {
 
 // src/tools/plan-core.ts
 var SUBMIT_PLAN_DESCRIPTION = "Submit ONE concrete plan you've already decided on. Use this for tasks that warrant a review gate \u2014 multi-file refactors, architecture changes, anything that would be expensive or confusing to undo. Skip it for small fixes (one-line typo, obvious bug with a clear fix) \u2014 just make the change. The user will either approve (you then implement it), ask for refinement, or cancel. If the user has already enabled /plan mode, writes are blocked at dispatch and you MUST use this. CRITICAL: do NOT use submit_plan to present alternative routes (A/B/C, option 1/2/3) for the user to pick from \u2014 the picker only exposes approve/refine/cancel, so a menu plan strands the user with no way to choose. For branching decisions, call `ask_choice` instead; only call submit_plan once the user has picked a direction and you have a single actionable plan. Write the plan as markdown with a one-line summary, a bulleted list of files to touch and what will change, and any risks or open questions. STRONGLY PREFERRED: pass `steps` \u2014 an array of {id, title, action, risk?} \u2014 so the UI renders a structured step list above the approval picker and tracks per-step progress. Use risk='high' for steps that touch prod data / break public APIs / are hard to undo; 'med' for non-trivial but reversible (multi-file edits, schema tweaks); 'low' for safe local work. After each step, call `mark_step_complete` so the user sees progress ticks.";
-var MARK_STEP_COMPLETE_DESCRIPTION = "Mark one step of the approved plan as done. Call this after finishing each step, then immediately continue with the NEXT step \u2014 do not stop or wait for the user. The TUI updates the plan card's progress in place. After the FINAL step, write a brief reply summarizing what was done and end the turn. Pass the `stepId` from the plan's steps array, a short `result` (what you did), and optional `notes` for anything surprising (errors, scope changes, follow-ups). This tool doesn't change any files. Don't call it if the plan didn't include structured steps, and don't invent ids that weren't in the original plan.";
+var MARK_STEP_COMPLETE_DESCRIPTION = "Mark one step of the approved plan as done. MANDATORY: call this exactly once after finishing each step, before starting the next one \u2014 skipping it leaves the user staring at `0/N done` on the resume banner even when the work is finished, and they have no way to know which steps actually ran. The TUI updates the plan card's progress in place; the count is persisted to disk so it survives session resume. After the FINAL step, write a brief reply summarizing what was done and end the turn. Pass the `stepId` from the plan's steps array, a short `result` (what you did), and optional `notes` for anything surprising (errors, scope changes, follow-ups). This tool doesn't change any files. Don't call it if the plan didn't include structured steps, and don't invent ids that weren't in the original plan. If you only realized at the end that you skipped marking steps, mark them then \u2014 late is still better than never.";
 var REVISE_PLAN_DESCRIPTION = "Surgically replace the REMAINING steps of an in-flight plan. Call this when the user has given feedback at a checkpoint that warrants a structured plan change \u2014 skip a step, swap two steps, add a new step, change risk, etc. Pass: `reason` (one sentence why), `remainingSteps` (the new tail of the plan, replacing whatever steps haven't been done yet), and optional `summary` (updated one-line plan summary). Done steps are NEVER touched \u2014 keep them out of `remainingSteps`. The TUI shows a diff (removed in red, kept in gray, added in green) and the user accepts or rejects. Don't call this for trivial mid-step adjustments \u2014 just keep executing. Don't call submit_plan for revisions either \u2014 that resets the whole plan including completed steps. Use submit_plan only when the entire approach has changed; use revise_plan when the tail needs editing.";
 var STEP_ITEM_SCHEMA = {
   type: "object",
@@ -3667,16 +3801,19 @@ async function searchMojeek(query, opts = {}) {
     signal: opts.signal,
     redirect: "follow"
   });
-  if (!resp.ok) throw new Error(`web_search ${resp.status}`);
+  if (!resp.ok) throw new Error(t("webErrors.status", { status: resp.status }));
   const html = await resp.text();
   const results = parseMojeekResults(html).slice(0, topK);
   if (results.length === 0) {
     if (/no results found|did not match any documents/i.test(html)) return [];
     if (/captcha|verify you are human|access denied|forbidden/i.test(html)) {
-      throw new Error("web_search: Mojeek anti-bot page \u2014 rate-limited or blocked");
+      throw new Error(t("webErrors.mojeekBlocked"));
     }
     throw new Error(
-      `web_search: 0 results but response doesn't look like a real empty page (${html.length} chars, first 120: ${html.slice(0, 120).replace(/\s+/g, " ")})`
+      t("webErrors.mojeekNoResults", {
+        chars: html.length,
+        preview: html.slice(0, 120).replace(/\s+/g, " ")
+      })
     );
   }
   return results;
@@ -3686,10 +3823,10 @@ function normalizeSearxngEndpoint(raw) {
   try {
     url = new URL(raw.includes("://") ? raw : `http://${raw}`);
   } catch {
-    throw new Error(`web_search: invalid SearXNG endpoint "${raw}"`);
+    throw new Error(t("webErrors.invalidEndpoint", { endpoint: raw }));
   }
   if (url.protocol !== "http:" && url.protocol !== "https:") {
-    throw new Error(`web_search: SearXNG endpoint must be http(s), got ${url.protocol}`);
+    throw new Error(t("webErrors.endpointMustBeHttp", { protocol: url.protocol }));
   }
   return url.origin;
 }
@@ -3709,19 +3846,17 @@ async function searchSearxng(query, opts = {}) {
   } catch (err) {
     if (err instanceof TypeError && err.message.includes("fetch")) {
       throw new Error(
-        `web_search: Cannot reach SearXNG server at ${opts.endpoint ?? "http://localhost:8080"}. Please install SearXNG (https://github.com/searxng/searxng) and start it (e.g. \`docker run -d -p 8080:8080 searxng/searxng\`), or switch to the default engine with /search-engine mojeek.`
+        t("webErrors.cannotReach", { endpoint: opts.endpoint ?? "http://localhost:8080" })
       );
     }
     throw err;
   }
-  if (!resp.ok) throw new Error(`web_search ${resp.status}`);
+  if (!resp.ok) throw new Error(t("webErrors.status", { status: resp.status }));
   const html = await resp.text();
   const results = parseSearxngHtmlResults(html).slice(0, topK);
   if (results.length === 0) {
     if (/no results found|did not match any documents/i.test(html)) return [];
-    throw new Error(
-      `web_search: 0 results but SearXNG response doesn't look like an empty results page (${html.length} chars)`
-    );
+    throw new Error(t("webErrors.searxngNoResults", { chars: html.length }));
   }
   return results;
 }
@@ -3815,13 +3950,11 @@ async function webFetch(url, opts = {}) {
     clearTimeout(timer);
     opts.signal?.removeEventListener("abort", cancel);
   }
-  if (!resp.ok) throw new Error(`web_fetch ${resp.status} for ${url}`);
+  if (!resp.ok) throw new Error(t("webErrors.fetchStatus", { status: resp.status, url }));
   const contentType = resp.headers.get("content-type") ?? "";
   const declaredLen = Number(resp.headers.get("content-length") ?? "");
   if (Number.isFinite(declaredLen) && declaredLen > FETCH_MAX_BYTES) {
-    throw new Error(
-      `web_fetch refused: content-length ${declaredLen} bytes exceeds ${FETCH_MAX_BYTES}-byte cap (${url})`
-    );
+    throw new Error(t("webErrors.fetchTooLarge", { len: declaredLen, cap: FETCH_MAX_BYTES, url }));
   }
   const raw = await readBodyCapped(resp, FETCH_MAX_BYTES);
   const title = extractTitle(raw);
@@ -3848,9 +3981,7 @@ async function readBodyCapped(resp, maxBytes) {
           await reader.cancel();
         } catch {
         }
-        throw new Error(
-          `web_fetch refused: response body exceeded ${maxBytes}-byte cap (${total} bytes seen)`
-        );
+        throw new Error(t("webErrors.fetchBodyTooLarge", { cap: maxBytes, seen: total }));
       }
       out += decoder.decode(value, { stream: true });
     }
@@ -3978,7 +4109,7 @@ function registerWebTools(registry, opts = {}) {
     },
     fn: async (args, ctx) => {
       if (!/^https?:\/\//i.test(args.url)) {
-        throw new Error("web_fetch: url must start with http:// or https://");
+        throw new Error(t("webErrors.fetchInvalidUrl"));
       }
       const page = await webFetch(args.url, { maxChars: maxFetchChars, signal: ctx?.signal });
       const header = page.title ? `${page.title}
@@ -5077,4 +5208,4 @@ export {
   snapshotBeforeEdits,
   restoreSnapshots
 };
-//# sourceMappingURL=chunk-SEFXUF24.js.map
+//# sourceMappingURL=chunk-5ZCRXN7S.js.map