@tuan_son.dinh/gsd 2.6.0

package/src/resources/extensions/voice/index.ts

@@ -0,0 +1,195 @@
+import type { ExtensionAPI, ExtensionContext } from "@mariozechner/pi-coding-agent";
+import { shortcutDesc } from "../shared/terminal.js";
+import type { AssistantMessage } from "@mariozechner/pi-ai";
+import { isKeyRelease, Key, matchesKey, truncateToWidth, visibleWidth } from "@mariozechner/pi-tui";
+import { spawn, execSync, type ChildProcess } from "node:child_process";
+import * as fs from "node:fs";
+import * as path from "node:path";
+import * as readline from "node:readline";
+
+const SWIFT_SRC = path.join(__dirname, "speech-recognizer.swift");
+const RECOGNIZER_BIN = path.join(__dirname, "speech-recognizer");
+
+function ensureBinary(): boolean {
+	if (fs.existsSync(RECOGNIZER_BIN)) return true;
+	try {
+		execSync(`swiftc "${SWIFT_SRC}" -o "${RECOGNIZER_BIN}" -framework Speech -framework AVFoundation`, {
+			timeout: 60000,
+		});
+		return true;
+	} catch {
+		return false;
+	}
+}
+
+export default function (pi: ExtensionAPI) {
+	if (process.platform !== "darwin") return;
+
+	let active = false;
+	let recognizerProcess: ChildProcess | null = null;
+	let flashOn = true;
+	let flashTimer: ReturnType<typeof setInterval> | null = null;
+	let footerTui: { requestRender: () => void } | null = null;
+
+	function setVoiceFooter(ctx: ExtensionContext, on: boolean) {
+		if (!on) {
+			stopFlash();
+			ctx.ui.setFooter(undefined);
+			return;
+		}
+
+		flashOn = true;
+		flashTimer = setInterval(() => {
+			flashOn = !flashOn;
+			footerTui?.requestRender();
+		}, 500);
+
+		ctx.ui.setFooter((tui, theme, footerData) => {
+			footerTui = tui;
+			const branchUnsub = footerData.onBranchChange(() => tui.requestRender());
+
+			return {
+				dispose: branchUnsub,
+				invalidate() {},
+				render(width: number): string[] {
+					// Row 1: pwd (branch) ... ● transcribing
+					let pwd = process.cwd();
+					const home = process.env.HOME || process.env.USERPROFILE;
+					if (home && pwd.startsWith(home)) pwd = `~${pwd.slice(home.length)}`;
+					const branch = footerData.getGitBranch();
+					if (branch) pwd = `${pwd} (${branch})`;
+
+					const dot = flashOn ? theme.fg("error", "●") : theme.fg("dim", "●");
+					const voiceTag = `${dot} ${theme.fg("error", "transcribing")}`;
+					const voiceTagWidth = visibleWidth(voiceTag);
+
+					const maxPwdWidth = width - voiceTagWidth - 2;
+					const pwdStr = truncateToWidth(theme.fg("dim", pwd), maxPwdWidth, theme.fg("dim", "..."));
+					const pad1 = " ".repeat(Math.max(1, width - visibleWidth(pwdStr) - voiceTagWidth));
+					const row1 = truncateToWidth(pwdStr + pad1 + voiceTag, width);
+
+					// Row 2: stats ... model
+					let totalInput = 0, totalOutput = 0, totalCost = 0;
+					for (const entry of ctx.sessionManager.getEntries()) {
+						if (entry.type === "message" && entry.message.role === "assistant") {
+							const m = entry.message as AssistantMessage;
+							totalInput += m.usage.input;
+							totalOutput += m.usage.output;
+							totalCost += m.usage.cost.total;
+						}
+					}
+
+					const fmt = (n: number) => n < 1000 ? `${n}` : n < 10000 ? `${(n / 1000).toFixed(1)}k` : `${Math.round(n / 1000)}k`;
+					const parts: string[] = [];
+					if (totalInput) parts.push(`↑${fmt(totalInput)}`);
+					if (totalOutput) parts.push(`↓${fmt(totalOutput)}`);
+					if (totalCost) parts.push(`$${totalCost.toFixed(3)}`);
+
+					const usage = ctx.getContextUsage();
+					const ctxPct = usage?.percent !== null && usage?.percent !== undefined ? `${usage.percent.toFixed(1)}%` : "?";
+					const ctxWin = usage?.contextWindow ?? ctx.model?.contextWindow ?? 0;
+					parts.push(`${ctxPct}/${fmt(ctxWin)}`);
+
+					const statsLeft = theme.fg("dim", parts.join(" "));
+					const modelRight = theme.fg("dim", ctx.model?.id || "no-model");
+					const statsLeftW = visibleWidth(statsLeft);
+					const modelRightW = visibleWidth(modelRight);
+					const pad2 = " ".repeat(Math.max(2, width - statsLeftW - modelRightW));
+					const row2 = truncateToWidth(statsLeft + pad2 + modelRight, width);
+
+					return [row1, row2];
+				},
+			};
+		});
+	}
+
+	function stopFlash() {
+		if (flashTimer) { clearInterval(flashTimer); flashTimer = null; }
+		footerTui = null;
+	}
+
+	async function toggleVoice(ctx: ExtensionContext) {
+		if (active) {
+			killRecognizer();
+			active = false;
+			setVoiceFooter(ctx, false);
+			return;
+		}
+
+		if (!ensureBinary()) {
+			ctx.ui.notify("Voice: failed to compile speech recognizer (need Xcode CLI tools)", "error");
+			return;
+		}
+
+		active = true;
+		setVoiceFooter(ctx, true);
+		await runVoiceSession(ctx);
+	}
+
+	pi.registerCommand("voice", {
+		description: "Toggle voice mode",
+		handler: async (_args, ctx) => toggleVoice(ctx),
+	});
+
+	pi.registerShortcut("ctrl+alt+v", {
+		description: shortcutDesc("Toggle voice mode", "/voice"),
+		handler: async (ctx) => toggleVoice(ctx),
+	});
+
+	function killRecognizer() {
+		if (recognizerProcess) { recognizerProcess.kill("SIGTERM"); recognizerProcess = null; }
+	}
+
+	function startRecognizer(
+		onPartial: (text: string) => void,
+		onFinal: (text: string) => void,
+		onError: (msg: string) => void,
+		onReady: () => void,
+	) {
+		recognizerProcess = spawn(RECOGNIZER_BIN, [], { stdio: ["pipe", "pipe", "pipe"] });
+		const rl = readline.createInterface({ input: recognizerProcess.stdout! });
+		rl.on("line", (line: string) => {
+			if (line === "READY") { onReady(); return; }
+			if (line.startsWith("PARTIAL:")) onPartial(line.slice(8));
+			else if (line.startsWith("FINAL:")) onFinal(line.slice(6));
+			else if (line.startsWith("ERROR:")) onError(line.slice(6));
+		});
+		recognizerProcess.on("error", (err) => onError(err.message));
+		recognizerProcess.on("exit", () => { recognizerProcess = null; });
+	}
+
+	async function runVoiceSession(ctx: ExtensionContext): Promise<void> {
+		return new Promise<void>((resolve) => {
+			// The Swift recognizer handles accumulation across pause-induced
+			// transcription resets. Both PARTIAL and FINAL messages contain
+			// the full accumulated text, so we just pass them through.
+			startRecognizer(
+				(text) => {
+					ctx.ui.setEditorText(text);
+				},
+				(text) => {
+					ctx.ui.setEditorText(text);
+				},
+				(msg) => ctx.ui.notify(`Voice: ${msg}`, "error"),
+				() => {},
+			);
+
+			ctx.ui.custom<void>(
+				(_tui, _theme, _kb, done) => ({
+					render(): string[] { return []; },
+					handleInput(data: string) {
+						if (isKeyRelease(data)) return;
+						if (matchesKey(data, Key.escape) || matchesKey(data, Key.enter)) {
+							killRecognizer();
+							active = false;
+							setVoiceFooter(ctx, false);
+							done();
+						}
+					},
+					invalidate() {},
+				}),
+				{ overlay: true, overlayOptions: { anchor: "bottom-center", width: "100%" } },
+			).then(() => resolve());
+		});
+	}
+}

package/src/resources/extensions/voice/speech-recognizer.swift

@@ -0,0 +1,154 @@
+import Foundation
+import Speech
+import AVFoundation
+
+// Unbuffered stdout
+setbuf(stdout, nil)
+
+guard SFSpeechRecognizer.authorizationStatus() == .authorized ||
+      SFSpeechRecognizer.authorizationStatus() == .notDetermined else {
+    print("ERROR:Speech recognition not authorized")
+    exit(1)
+}
+
+SFSpeechRecognizer.requestAuthorization { status in
+    guard status == .authorized else {
+        print("ERROR:Speech recognition denied")
+        exit(1)
+    }
+}
+
+let recognizer = SFSpeechRecognizer(locale: Locale(identifier: "en-US"))!
+guard recognizer.isAvailable else {
+    print("ERROR:Speech recognizer not available")
+    exit(1)
+}
+
+let audioEngine = AVAudioEngine()
+let request = SFSpeechAudioBufferRecognitionRequest()
+request.shouldReportPartialResults = true
+request.requiresOnDeviceRecognition = true
+
+let node = audioEngine.inputNode
+let format = node.outputFormat(forBus: 0)
+
+node.installTap(onBus: 0, bufferSize: 1024, format: format) { buffer, _ in
+    request.append(buffer)
+}
+
+audioEngine.prepare()
+do {
+    try audioEngine.start()
+    print("READY")
+} catch {
+    print("ERROR:Failed to start audio engine: \(error.localizedDescription)")
+    exit(1)
+}
+
+// Accumulated finalized text from previous recognition segments.
+// On-device recognition (especially macOS/iOS 18+) can reset
+// bestTranscription.formattedString after a pause, discarding
+// previous text. We detect this by tracking the last known good
+// text and noticing when the new text is shorter / doesn't start
+// with the previous text. When that happens we treat the previous
+// text as finalized and start accumulating the new segment on top.
+var accumulated = ""
+var lastPartialText = ""
+var lastEmitted = ""
+
+recognizer.recognitionTask(with: request) { result, error in
+    if let result = result {
+        let text = result.bestTranscription.formattedString
+
+        if result.isFinal {
+            // True final from the recognizer — commit everything
+            let full: String
+            // Check if the final text already includes accumulated content
+            // (some OS versions give cumulative finals, others reset)
+            if !accumulated.isEmpty && !text.lowercased().hasPrefix(accumulated.lowercased()) {
+                full = accumulated + " " + text
+            } else if !accumulated.isEmpty && text.count < accumulated.count {
+                // Final is shorter than what we accumulated — use accumulated + new
+                full = accumulated + " " + text
+            } else {
+                full = text
+            }
+            accumulated = ""
+            lastPartialText = ""
+            if full != lastEmitted {
+                lastEmitted = full
+                print("FINAL:\(full)")
+            }
+            return
+        }
+
+        // Detect transcription reset: if the new partial text is significantly
+        // shorter than what we had, or doesn't start with the previous text,
+        // the recognizer has reset after a pause. Finalize what we had.
+        let prevText = lastPartialText
+        if !prevText.isEmpty && !text.isEmpty {
+            let prevWords = prevText.split(separator: " ")
+            let newWords = text.split(separator: " ")
+
+            // Reset detection: new text has fewer words than previous AND
+            // the first few words don't match (i.e. it's truly new speech,
+            // not just the recognizer revising the last word)
+            let looksLikeReset: Bool
+            if newWords.count < prevWords.count / 2 {
+                // Significant drop in word count — likely a reset
+                looksLikeReset = true
+            } else if newWords.count < prevWords.count &&
+                      !prevWords.isEmpty && !newWords.isEmpty &&
+                      newWords[0] != prevWords[0] {
+                // Different starting word + fewer words — reset
+                looksLikeReset = true
+            } else {
+                looksLikeReset = false
+            }
+
+            if looksLikeReset {
+                // Commit the previous partial text to accumulated
+                if accumulated.isEmpty {
+                    accumulated = prevText
+                } else {
+                    accumulated = accumulated + " " + prevText
+                }
+                // Emit a FINAL for the committed text so the TS side updates
+                print("FINAL:\(accumulated)")
+                lastEmitted = accumulated
+            }
+        }
+
+        lastPartialText = text
+
+        // Build the full display text
+        let displayText: String
+        if accumulated.isEmpty {
+            displayText = text
+        } else {
+            displayText = accumulated + " " + text
+        }
+
+        if displayText != lastEmitted {
+            lastEmitted = displayText
+            print("PARTIAL:\(displayText)")
+        }
+    }
+    if let error = error {
+        // Task finished errors are normal on kill
+        let nsError = error as NSError
+        if nsError.code != 216 { // kAFAssistantErrorDomain code for cancelled
+            print("ERROR:\(error.localizedDescription)")
+        }
+    }
+}
+
+// Handle SIGTERM/SIGINT gracefully
+signal(SIGTERM) { _ in
+    exit(0)
+}
+signal(SIGINT) { _ in
+    exit(0)
+}
+
+RunLoop.current.run()

package/src/resources/skills/debug-like-expert/SKILL.md

@@ -0,0 +1,231 @@
+---
+name: debug-like-expert
+description: Deep analysis debugging mode for complex issues. Activates methodical investigation protocol with evidence gathering, hypothesis testing, and rigorous verification. Use when standard troubleshooting fails or when issues require systematic root cause analysis.
+---
+
+<objective>
+Deep analysis debugging mode for complex issues. This skill activates methodical investigation protocols with evidence gathering, hypothesis testing, and rigorous verification when standard troubleshooting has failed.
+
+The skill emphasizes treating code you wrote with MORE skepticism than unfamiliar code, as cognitive biases about "how it should work" can blind you to actual implementation errors. Use scientific method to systematically identify root causes rather than applying quick fixes.
+</objective>
+
+<context>
+This skill activates when standard troubleshooting has failed. The issue requires methodical investigation, not quick fixes. You are entering the mindset of a senior engineer who debugs with scientific rigor.
+
+**Important**: If you wrote or modified any of the code being debugged, you have cognitive biases about how it works. Your mental model of "how it should work" may be wrong. Treat code you wrote with MORE skepticism than unfamiliar code - you're blind to your own assumptions.
+</context>
+
+<core_principle>
+**VERIFY, DON'T ASSUME.** Every hypothesis must be tested. Every "fix" must be validated. No solutions without evidence.
+
+**ESPECIALLY**: Code you designed or implemented is guilty until proven innocent. Your intent doesn't matter - only the code's actual behavior matters. Question your own design decisions as rigorously as you'd question anyone else's.
+</core_principle>
+
+<analysis_only_rule>
+**THIS SKILL IS READ-ONLY. DO NOT MODIFY CODE.**
+
+The entire purpose is deep analysis and diagnosis. Making changes during investigation:
+- Pollutes the evidence
+- Introduces new variables
+- Makes root cause harder to isolate
+
+You are a diagnostician, not a surgeon. Present findings, then let the user decide.
+</analysis_only_rule>
+
+<quick_start>
+
+<evidence_gathering>
+
+Before proposing any solution:
+
+**A. Document Current State**
+- What is the EXACT error message or unexpected behavior?
+- What are the EXACT steps to reproduce?
+- What is the ACTUAL output vs EXPECTED output?
+- When did this start working incorrectly (if known)?
+
+**B. Map the System**
+- Trace the execution path from entry point to failure point
+- Identify all components involved
+- Read relevant source files completely, not just scanning
+- Note dependencies, imports, configurations affecting this area
+
+**C. Gather External Knowledge (when needed)**
+- Use MCP servers for API documentation, library details, or domain knowledge
+- Use web search for error messages, framework-specific behaviors, or recent changes
+- Check official docs for intended behavior vs what you observe
+- Look for known issues, breaking changes, or version-specific quirks
+
+See [references/when-to-research.md](references/when-to-research.md) for detailed guidance on research strategy.
+
+</evidence_gathering>
+
+<root_cause_analysis>
+
+**A. Form Hypotheses**
+
+Based on evidence, list possible causes:
+1. [Hypothesis 1] - because [specific evidence]
+2. [Hypothesis 2] - because [specific evidence]
+3. [Hypothesis 3] - because [specific evidence]
+
+**B. Test Each Hypothesis**
+
+For each hypothesis:
+- What would prove this true?
+- What would prove this false?
+- Design a minimal test
+- Execute and document results
+
+See [references/hypothesis-testing.md](references/hypothesis-testing.md) for scientific method application.
+
+**C. Eliminate or Confirm**
+
+Don't move forward until you can answer:
+- Which hypothesis is supported by evidence?
+- What evidence contradicts other hypotheses?
+- What additional information is needed?
+
+</root_cause_analysis>
+
+<solution_proposal>
+
+**Only after confirming root cause:**
+
+**A. Design Recommended Fix**
+- What is the MINIMAL change that would address the root cause?
+- What are potential side effects?
+- What could this break?
+- What tests should run after implementation?
+
+**B. Document, Don't Implement**
+- Describe the fix with enough detail for implementation
+- Include specific file paths, line numbers, and code snippets
+- Explain WHY this addresses the root cause
+- Note any prerequisites or dependencies
+
+**DO NOT make any code changes. Present your recommendations only.**
+
+See [references/verification-patterns.md](references/verification-patterns.md) for verification approaches to use after implementation.
+
+</solution_proposal>
+
+</quick_start>
+
+<critical_rules>
+
+1. **NO DRIVE-BY FIXES**: If you can't explain WHY a change works, don't make it
+2. **VERIFY EVERYTHING**: Test your assumptions. Read the actual code. Check the actual behavior
+3. **USE ALL TOOLS**:
+   - MCP servers for external knowledge
+   - Web search for error messages, docs, known issues
+   - Extended thinking ("think deeply") for complex reasoning
+   - File reading for complete context
+4. **THINK OUT LOUD**: Document your reasoning at each step
+5. **ONE VARIABLE**: Change one thing at a time, verify, then proceed
+6. **COMPLETE READS**: Don't skim code. Read entire relevant files
+7. **CHASE DEPENDENCIES**: If the issue involves libraries, configs, or external systems, investigate those too
+8. **QUESTION PREVIOUS WORK**: Maybe the earlier "fix" was wrong. Re-examine with fresh eyes
+
+</critical_rules>
+
+<success_criteria>
+
+Before completing:
+- [ ] Do you understand WHY the issue occurred?
+- [ ] Have you identified a root cause with evidence?
+- [ ] Have you documented your reasoning?
+- [ ] Can you explain the issue to someone else?
+- [ ] Is your recommended fix specific and actionable?
+
+If you can't answer "yes" to all of these, keep investigating.
+
+**CRITICAL**: Present findings via decision gate. Do NOT implement changes.
+
+</success_criteria>
+
+<output_format>
+
+```markdown
+## Issue: [Problem Description]
+
+### Evidence
+[What you observed - exact errors, behaviors, outputs]
+
+### Investigation
+[What you checked, what you found, what you ruled out]
+
+### Root Cause
+[The actual underlying problem with evidence]
+
+### Recommended Fix
+[What SHOULD be changed and WHY - specific files, lines, code]
+
+### Verification Plan
+[How to confirm the fix works after implementation]
+
+### Risk Assessment
+[Potential side effects, what could break, confidence level]
+```
+
+</output_format>
+
+<advanced_topics>
+
+For deeper topics, see reference files:
+
+**Debugging mindset**: [references/debugging-mindset.md](references/debugging-mindset.md)
+- First principles thinking applied to debugging
+- Cognitive biases that lead to bad fixes
+- The discipline of systematic investigation
+- When to stop and restart with fresh assumptions
+
+**Investigation techniques**: [references/investigation-techniques.md](references/investigation-techniques.md)
+- Binary search / divide and conquer
+- Rubber duck debugging
+- Minimal reproduction
+- Working backwards from desired state
+- Adding observability before changing code
+
+**Hypothesis testing**: [references/hypothesis-testing.md](references/hypothesis-testing.md)
+- Forming falsifiable hypotheses
+- Designing experiments that prove/disprove
+- What makes evidence strong vs weak
+- Recovering from wrong hypotheses gracefully
+
+**Verification patterns**: [references/verification-patterns.md](references/verification-patterns.md)
+- Definition of "verified" (not just "it ran")
+- Testing reproduction steps
+- Regression testing adjacent functionality
+- When to write tests before fixing
+
+**Research strategy**: [references/when-to-research.md](references/when-to-research.md)
+- Signals that you need external knowledge
+- What to search for vs what to reason about
+- Balancing research time vs experimentation
+
+</advanced_topics>
+
+<decision_gate>
+
+**After presenting findings, ALWAYS offer these options:**
+
+```
+─────────────────────────────────────────
+ANALYSIS COMPLETE
+
+What would you like to do?
+
+1. **Fix it now** - I'll implement the recommended changes
+2. **Create findings document** - Save analysis to a markdown file
+3. **Explore further** - Investigate additional hypotheses
+4. **Get second opinion** - Review with different assumptions
+5. **Other** - Tell me what you need
+─────────────────────────────────────────
+```
+
+**Wait for user response before taking any action.**
+
+This gate is MANDATORY. Never skip it. Never auto-implement.
+
+</decision_gate>