claude-overnight 1.25.20 → 1.25.22

package/dist/_version.d.ts

@@ -1 +1 @@
-export declare const VERSION = "1.25.20";
+export declare const VERSION = "1.25.22";

package/dist/_version.js

@@ -1,2 +1,2 @@
 // Auto-generated by build — do not edit manually.
-export const VERSION = "1.25.20";
+export const VERSION = "1.25.22";

package/dist/index.js

@@ -822,14 +822,44 @@ async function main() {
         // proxy now uses an account pool (`CURSOR_CONFIG_DIRS`) — each parallel
         // cursor-agent subprocess gets its own config dir, eliminating the race.
         // See ensureCursorAccountPool() in providers.ts.
-        const progress = (msg) => process.stdout.write(chalk.dim(`    ${msg}\n`));
+        //
+        // Single in-place status line collapses N parallel progress streams (one
+        // per provider) into one tty line updated via `\r` + ANSI clear. Keeps the
+        // "window head" calm instead of appending 3 lines per 3s tick.
+        const statuses = new Map();
+        const isTTY = process.stdout.isTTY;
+        let statusLineActive = false;
+        const renderStatus = () => {
+            if (!isTTY)
+                return;
+            const parts = [...statuses.entries()].map(([r, s]) => `${r} ${chalk.dim(s)}`);
+            process.stdout.write(`\x1B[2K\r`);
+            if (parts.length === 0) {
+                statusLineActive = false;
+                return;
+            }
+            process.stdout.write(chalk.dim("    " + parts.join("  ·  ")));
+            statusLineActive = true;
+        };
+        const clearStatusLine = () => {
+            if (isTTY && statusLineActive) {
+                process.stdout.write(`\x1B[2K\r`);
+                statusLineActive = false;
+            }
+        };
         /** Cursor agent cold start + thinking-variant model latency can exceed 20s; API providers stay tight. */
         const preflightMs = (p) => isCursorProxyProvider(p) ? 60_000 : 20_000;
-        const results = await Promise.all(pending.map(async ([role, p]) => ({
-            role,
-            provider: p,
-            result: await preflightProvider(p, cwd, preflightMs(p), { onProgress: progress }),
-        })));
+        const results = await Promise.all(pending.map(async ([role, p]) => {
+            statuses.set(role, "connecting…");
+            renderStatus();
+            const result = await preflightProvider(p, cwd, preflightMs(p), {
+                onProgress: (msg) => { statuses.set(role, msg); renderStatus(); },
+            });
+            statuses.delete(role);
+            renderStatus();
+            return { role, provider: p, result };
+        }));
+        clearStatusLine();
         for (const { role, provider, result } of results) {
             if (!result.ok) {
                 console.error(chalk.red(`  ✗ ${role} preflight failed: ${chalk.dim(result.error)}`));

package/dist/planner-query.d.ts

@@ -25,6 +25,16 @@ export interface PlannerOpts {
     };
     /** When set, stream events are appended to <runDir>/transcripts/<name>.ndjson */
     transcriptName?: string;
+    /**
+     * Hard cap on conversation turns. Defaults to 20.
+     */
+    maxTurns?: number;
+    /**
+     * Tools the planner agent may use. Defaults to read-only + Write (for outFile
+     * resilience). Deliberately excludes Bash/Agent/TodoWrite/WebFetch to prevent
+     * the multi-turn tool loops that cause error_max_turns with thinking models.
+     */
+    tools?: string[];
 }
 export declare function setPlannerEnvResolver(fn: ((model?: string) => Record<string, string> | undefined) | undefined): void;
 export declare function getTotalPlannerCost(): number;

package/dist/planner-query.js

@@ -2,6 +2,7 @@ import { query } from "@anthropic-ai/claude-agent-sdk";
 import { readFileSync } from "fs";
 import { NudgeError } from "./types.js";
 import { writeTranscriptEvent } from "./transcripts.js";
+const DEFAULT_MAX_TURNS = 20;
 // ── Shared env resolver (set once at run start, used by every planner query) ──
 //
 // Swarm and planner calls share a model→env map so a custom provider configured
@@ -143,20 +144,32 @@ async function runPlannerQueryOnce(prompt, opts, onLog) {
         options: {
             cwd: opts.cwd,
             model: opts.model,
-            tools: ["Read", "Glob", "Grep", "Write", "Bash", "WebFetch", "WebSearch", "TodoWrite", "Agent"],
-            allowedTools: ["Read", "Glob", "Grep", "Write", "Bash", "WebFetch", "WebSearch", "TodoWrite", "Agent"],
+            tools: opts.tools ?? ["Read", "Glob", "Grep", "Write"],
+            allowedTools: opts.tools ?? ["Read", "Glob", "Grep", "Write"],
             permissionMode: opts.permissionMode,
             ...(opts.permissionMode === "bypassPermissions" && { allowDangerouslySkipPermissions: true }),
             persistSession: true,
             includePartialMessages: true,
+            maxTurns: opts.maxTurns ?? DEFAULT_MAX_TURNS,
             ...(isResume && { resume: opts.resumeSessionId }),
             ...(opts.outputFormat && { outputFormat: opts.outputFormat }),
             ...(envOverride && { env: envOverride }),
         },
     });
-    let lastLogText = "";
+    // Default to "thinking…" so the ticker conveys meaning during the pre-output
+    // reasoning phase. Thinking-variant models (e.g. claude-opus-4-7-thinking-*)
+    // can sit silent for 60-90s before emitting any tokens, and cursor-agent
+    // doesn't forward thinking deltas — without this, the ticker reads "4m 5s"
+    // with nothing else for a minute plus.
+    let lastLogText = "thinking…";
     let toolCount = 0;
     let costUsd = 0;
+    const jsonOutput = opts.outputFormat?.type === "json_schema";
+    let jsonCharCount = 0;
+    // Dedup identical text snippets: cursor-agent with json_schema-ignoring
+    // proxies causes the SDK to loop multiple turns, each re-emitting the same
+    // final JSON. We don't want to spam the ticker or transcript with it.
+    let lastTextSeen = "";
     const ticker = setInterval(() => {
         const elapsed = Math.round((Date.now() - startedAt) / 1000);
         const m = Math.floor(elapsed / 60);
@@ -202,6 +215,11 @@ async function runPlannerQueryOnce(prompt, opts, onLog) {
     // Track the open tool block so we can re-log with the enriched target once
     // the input arrives, and write a complete transcript entry on block stop.
     let pendingTool = null;
+    // Dedup tool_use logging between stream_event path and full-assistant-message
+    // path. The Cursor proxy doesn't always relay content_block_* frames through
+    // the Claude CLI's stream-json, so we also mine complete `SDKAssistantMessage`
+    // content for progress  -- without double-counting when both paths fire.
+    const seenToolIds = new Set();
     const logTool = (name, input) => {
         const target = extractToolTarget(input);
         lastLogText = target ? `${name} ${target}` : name;
@@ -220,6 +238,8 @@ async function runPlannerQueryOnce(prompt, opts, onLog) {
                         toolCount++;
                         const input = (cb.input ?? {});
                         const hasInput = Object.keys(input).length > 0;
+                        if (cb.id)
+                            seenToolIds.add(cb.id);
                         pendingTool = {
                             index: ev.index ?? 0,
                             name: cb.name,
@@ -252,9 +272,18 @@ async function runPlannerQueryOnce(prompt, opts, onLog) {
                         : delta?.type === "thinking_delta" ? delta.thinking
                             : undefined;
                     if (typeof raw === "string" && raw) {
-                        const snippet = raw.trim().replace(/[{}"\\,[\]]+/g, " ").replace(/\s+/g, " ").trim();
-                        if (snippet.length > 5)
-                            lastLogText = snippet.slice(-60);
+                        if (jsonOutput && delta.type === "text_delta") {
+                            // Don't surface tail-of-JSON as "progress" — it reads as noise
+                            // like `…ppression and optimistic-update rollback`. Show size
+                            // growing instead, which is a genuine signal.
+                            jsonCharCount += raw.length;
+                            lastLogText = `writing JSON (${jsonCharCount} chars)…`;
+                        }
+                        else {
+                            const snippet = raw.trim().replace(/[{}"\\,[\]]+/g, " ").replace(/\s+/g, " ").trim();
+                            if (snippet.length > 5)
+                                lastLogText = snippet.slice(-60);
+                        }
                         if (tname)
                             writeTranscriptEvent(tname, { kind: delta.type, text: raw });
                     }
@@ -274,6 +303,48 @@ async function runPlannerQueryOnce(prompt, opts, onLog) {
                     pendingTool = null;
                 }
             }
+            // Fallback progress surfacing: when stream events are sparse (e.g. the
+            // Cursor proxy's heartbeat thinking block doesn't always round-trip
+            // through the Claude CLI as partial messages), mine the full assistant
+            // turn message for tool_use / thinking / text so the ticker still moves
+            // every ~6-15s instead of sitting silent for minutes.
+            if (msg.type === "assistant") {
+                const content = msg.message?.content;
+                if (Array.isArray(content)) {
+                    for (const part of content) {
+                        if (part?.type === "tool_use" && part.id && !seenToolIds.has(part.id)) {
+                            seenToolIds.add(part.id);
+                            toolCount++;
+                            const input = (part.input ?? {});
+                            logTool(part.name, input);
+                            if (tname)
+                                writeTranscriptEvent(tname, { kind: "tool_use", tool: part.name, input });
+                        }
+                        else if (part?.type === "thinking" && typeof part.thinking === "string" && part.thinking) {
+                            const snippet = part.thinking.trim().replace(/\s+/g, " ").slice(-60);
+                            if (snippet)
+                                lastLogText = snippet;
+                            if (tname)
+                                writeTranscriptEvent(tname, { kind: "thinking", text: part.thinking });
+                        }
+                        else if (part?.type === "text" && typeof part.text === "string" && part.text) {
+                            if (part.text === lastTextSeen)
+                                continue; // dedup repeated turns
+                            lastTextSeen = part.text;
+                            if (jsonOutput) {
+                                lastLogText = `writing JSON (${part.text.length} chars)…`;
+                            }
+                            else {
+                                const snippet = part.text.trim().replace(/[{}"\\,[\]]+/g, " ").replace(/\s+/g, " ").slice(-60);
+                                if (snippet.length > 5)
+                                    lastLogText = snippet;
+                            }
+                            if (tname)
+                                writeTranscriptEvent(tname, { kind: "text", text: part.text });
+                        }
+                    }
+                }
+            }
             if (msg.type === "rate_limit_event") {
                 const info = msg.rate_limit_info;
                 if (info) {

package/dist/planner.js

@@ -158,7 +158,8 @@ export async function planTasks(objective, cwd, plannerModel, workerModel, permi
     const fileInstruction = outFile ? `\n\nAFTER generating the JSON, also write it to ${outFile} using the Write tool.` : "";
     let resultText;
     try {
-        resultText = await runPlannerQuery(prompt + fileInstruction, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName }, onLog);
+        resultText = await runPlannerQuery(prompt + fileInstruction, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName, maxTurns: 40,
+            tools: ["Read", "Glob", "Grep", "Write"] }, onLog);
     }
     catch (err) {
         const salvaged = salvageFromFile(outFile, budget, onLog, err?.message ?? String(err));
@@ -168,7 +169,7 @@ export async function planTasks(objective, cwd, plannerModel, workerModel, permi
     }
     const parsed = await extractTaskJson(resultText, async () => {
         onLog("Retrying...");
-        return runPlannerQuery(`Your previous response was not valid JSON. Respond with ONLY a JSON object {"tasks":[{"prompt":"..."}]}.\n\n${prompt}`, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName: `${transcriptName}-retry` }, onLog);
+        return runPlannerQuery(`Your previous response was not valid JSON. Respond with ONLY a JSON object {"tasks":[{"prompt":"..."}]}.\n\n${prompt}`, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName: `${transcriptName}-retry`, maxTurns: 15 }, onLog);
     }, onLog, outFile);
     let tasks = (parsed.tasks || []).map((t, i) => ({
         id: String(i), prompt: typeof t === "string" ? t : t.prompt,
@@ -188,7 +189,7 @@ Then pick ${count} angles that carve up THIS specific codebase orthogonally. Pre
 
 Objective: ${objective}
 
-Return ONLY a JSON object: {"themes": ["angle description", ...]}`, { cwd, model, permissionMode, outputFormat: THEMES_SCHEMA, transcriptName }, onLog);
+Return ONLY a JSON object: {"themes": ["angle description", ...]}`, { cwd, model, permissionMode, outputFormat: THEMES_SCHEMA, transcriptName, maxTurns: 12 }, onLog);
     const parsed = attemptJsonParse(resultText);
     if (parsed?.themes && Array.isArray(parsed.themes))
         return parsed.themes.slice(0, count);
@@ -259,7 +260,8 @@ Respond with ONLY a JSON object (no markdown fences):
     onLog("Synthesizing...");
     let resultText;
     try {
-        resultText = await runPlannerQuery(prompt, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName }, onLog);
+        resultText = await runPlannerQuery(prompt, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName, maxTurns: 25,
+            tools: ["Write"] }, onLog);
     }
     catch (err) {
         const salvaged = salvageFromFile(outFile, budget, onLog, err?.message ?? String(err));
@@ -269,7 +271,7 @@ Respond with ONLY a JSON object (no markdown fences):
     }
     const parsed = await extractTaskJson(resultText, async () => {
         onLog("Retrying...");
-        return runPlannerQuery(`Your previous response was not valid JSON. Respond with ONLY a JSON object {"tasks":[{"prompt":"..."}]}.\n\n${prompt}`, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName: `${transcriptName}-retry` }, onLog);
+        return runPlannerQuery(`Your previous response was not valid JSON. Respond with ONLY a JSON object {"tasks":[{"prompt":"..."}]}.\n\n${prompt}`, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName: `${transcriptName}-retry`, maxTurns: 10 }, onLog);
     }, onLog, outFile);
     let tasks = (parsed.tasks || []).map((t, i) => ({
         id: String(i), prompt: typeof t === "string" ? t : t.prompt,
@@ -303,10 +305,10 @@ ${scaleNote} ${concurrency} agents run in parallel. Update the plan accordingly.
 
 Respond with ONLY a JSON object (no markdown):
 {"tasks":[{"prompt":"..."}]}`;
-    const resultText = await runPlannerQuery(prompt, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName }, onLog);
+    const resultText = await runPlannerQuery(prompt, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName, maxTurns: 15 }, onLog);
     const parsed = await extractTaskJson(resultText, async () => {
         onLog("Retrying...");
-        return runPlannerQuery(`Your previous response was not valid JSON. Respond with ONLY a JSON object {"tasks":[{"prompt":"..."}]}.\n\n${prompt}`, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName: `${transcriptName}-retry` }, onLog);
+        return runPlannerQuery(`Your previous response was not valid JSON. Respond with ONLY a JSON object {"tasks":[{"prompt":"..."}]}.\n\n${prompt}`, { cwd, model: plannerModel, permissionMode, outputFormat: TASKS_SCHEMA, transcriptName: `${transcriptName}-retry`, maxTurns: 8 }, onLog);
     }, onLog);
     let tasks = (parsed.tasks || []).map((t, i) => ({
         id: String(i), prompt: typeof t === "string" ? t : t.prompt,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-overnight",
-  "version": "1.25.20",
+  "version": "1.25.22",
   "description": "Parallel Claude agents in git worktrees with a usage cap that reserves headroom for your interactive Claude Code. Crash-safe resume. Provider-agnostic model catalog (Anthropic, Cursor, OpenAI, Gemini, DeepSeek, Llama, Qwen) with capability-based task scoping.",
   "type": "module",
   "bin": {

package/plugins/claude-overnight/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-overnight",
-  "version": "1.25.20",
+  "version": "1.25.22",
   "description": "Claude Code skill for understanding, installing, and inspecting claude-overnight runs  -- parallel Claude agents in git worktrees with thinking waves, multi-wave steering, and crash-safe resume. Supports Cursor API Proxy, Qwen, OpenRouter.",
   "author": {
     "name": "Francesco Fornace"