feat: add command palette with autocomplete and fuzzy matching

.grok/PR-COMMAND-PALETTE.md

@@ -0,0 +1,154 @@
+## Summary
+
+Adds interactive command palette with autocomplete, keyboard navigation, fuzzy matching, and "did you mean" suggestions.
+
+**Problem**: Typing `/` in Grok TUI shows nothing - users must already know command names. No autocomplete, no command discovery, no typo assistance.
+
+**Solution**: Interactive command palette that appears when typing `/`, with smart filtering and keyboard navigation.
+
+## Features
+
+### 1. Live Command Filtering
+
+Type `/` to see all commands, then filter as you type:
+
+```
+> /
+┌─ Commands (7 matches) ─┐
+▶ /auth - Authentication management
+  /clear - Clear conversation
+  /exit - Exit application
+  /help - Show help
+  /history - Show conversation history
+  /model - Set AI model
+  /prompt - Load prompt from file
+└─────────────────────────┘
+Tab: autocomplete • ↑/↓: navigate • Esc: close • Enter: submit
+```
+
+### 2. Smart Matching
+
+**Prefix match**:
+```
+> /pr
+┌─ Commands (1 match) ─┐
+▶ /prompt - Load prompt from file
+└───────────────────────┘
+```
+
+**Alias match**:
+```
+> /pf
+┌─ Commands (1 match) ─┐
+▶ /prompt (via pf) - Load prompt from file
+└────────────────────────────────────────┘
+```
+
+**Fuzzy match** (typo tolerance):
+```
+> /promt
+┌─ Commands (1 match) ─┐
+▶ /prompt - Load prompt from file
+└───────────────────────┘
+```
+
+### 3. Keyboard Navigation
+
+- **↑/↓**: Navigate suggestions (only when palette visible)
+- **Tab**: Autocomplete to selected command + add space
+- **Esc**: Close palette
+- **Enter**: Submit command as usual
+
+### 4. "Did You Mean" for Unknown Commands
+
+```
+> /promt myfile.txt
+[Enter]
+
+Error: Unknown command: /promt
+
+Did you mean: /prompt?
+
+Use /help to see all available commands.
+```
+
+## Implementation
+
+### Phase 1: Suggestion Algorithm (`src/ui/utils/command-suggestions.ts`)
+
+**Levenshtein Distance**:
+- Calculates edit distance for fuzzy matching
+- Tolerates up to 2 character edits
+- No external dependencies (30 lines, pure algorithm)
+
+**Matching Types** (ranked):
+1. **Exact** (1000 points): `/prompt` → `prompt`
+2. **Prefix** (900 points): `/pr` → `prompt`
+3. **Alias Exact** (800 points): `/pf` → `prompt`
+4. **Alias Prefix** (700 points): `/promptf` → `prompt` (via `promptfile`)
+5. **Substring** (600 points): `/odel` → `model`
+6. **Fuzzy** (500-distance*100): `/promt` → `prompt` (1 edit = 400 points)
+
+### Phase 2: UI Component (`src/ui/components/command-palette.tsx`)
+
+- Cyan bordered box (consistent with tool selection)
+- Selected item: inverse text + bold + arrow indicator
+- Shows matched alias when applicable
+- Keyboard hint at bottom
+
+### Phase 3: Input Integration (`src/ui/components/input.tsx`)
+
+- Computes suggestions in real-time (`useMemo`)
+- Shows palette when `value.startsWith('/')` and not pasting
+- Arrow keys captured when palette visible (priority over tool navigation)
+- Tab autocompletes and preserves any args already typed
+- Escape closes palette without clearing input
+
+### Phase 4: Error Enhancement (`src/commands/index.ts`)
+
+- Unknown command errors now include fuzzy suggestion
+- Uses `getDidYouMeanSuggestion()` for best match
+- Shows alias if matched via alias
+
+## Compatibility
+
+**No conflicts with existing features**:
+- ✅ Bracketed paste: Palette hidden during `isPasting`
+- ✅ Tool navigation: Only captures arrows when palette visible (tool nav when `isActive=false`)
+- ✅ Multi-line input: Palette respects multi-line display
+- ✅ Command execution: Unchanged behavior
+
+## Test Coverage
+
+**New Test File**: `tests/unit/command-suggestions.test.ts` (25 tests)
+
+- Levenshtein distance correctness (7 tests)
+- Suggestion matching (exact, prefix, alias, substring, fuzzy) (12 tests)
+- Ranking and sorting (2 tests)
+- "Did you mean" logic (4 tests)
+
+**Test Results**: 238/238 pass (+25 from 213)
+
+## Verification
+
+```bash
+✅ npm run build - TypeScript compilation clean
+✅ npm test - 238/238 tests pass (+25 new tests)
+✅ No conflicts with paste/navigation modes
+```
+
+## Manual Testing
+
+After merge, verify in TUI:
+
+1. **Discovery**: Type `/` → see all 7 commands
+2. **Filter**: Type `/pr` → see `/prompt`
+3. **Alias**: Type `/pf` → matches `/prompt (via pf)`
+4. **Fuzzy**: Type `/promt` → matches `/prompt`
+5. **Navigate**: Press ↓/↑ → selection moves
+6. **Autocomplete**: Press Tab → completes to `/prompt `
+7. **Typo Help**: Submit `/promt` → error shows "Did you mean: /prompt?"
+
+---
+
+🤖 Generated with [Claude Code](https://claude.com/claude-code)

.grok/PR-EXPLORE-QUALITY.md

@@ -0,0 +1,98 @@
+## Summary
+
+Fixes explore subagent output quality - it now reliably finds code evidence using token-based searches instead of failing with phrase-based searches.
+
+**Problem**: When asked to find code like "where taskSuccesses increments", the explore subagent searched for English phrases ("successful Task", "count successful") instead of code tokens (`taskSuccesses`, `call.tool === 'Task'`), resulting in "no matches found."
+
+**Solution**: Enhanced explore.md system prompt with explicit operating rules for code evidence gathering.
+
+## Root Cause
+
+**Current behavior** (broken):
+```
+User: "Find where taskSuccesses increments"
+Subagent: Grep: "count successful" → no matches
+```
+
+**Expected behavior** (fixed):
+```
+User: "Find where taskSuccesses increments"
+Subagent: Grep: "taskSuccesses" → finds definition + uses
+Subagent: Grep: "taskSuccesses.*\+=" → finds increment line
+Subagent: Quotes file:line:content
+```
+
+## Changes
+
+**File**: `.grok/agents/explore.md`
+
+### Added Operating Rules
+
+**1. Token-Based Code Searches** (CRITICAL)
+- Use code tokens/identifiers: `taskSuccesses`, `call.tool`, `runSubagent`
+- NOT English phrases: "successful Task", "count successful"
+- Includes examples: variable names, function names, class names, operators
+
+**2. File:Line Evidence Requirements**
+- Use Grep with identifiers
+- Include actual grep output (file:line:content)
+- Provide exact line numbers and exact code
+- Read file if grep output is long
+
+**3. Search Scope Preferences**
+- **Prefer `src/**` over `tests/**`** unless tests explicitly requested
+- Use patterns: `Glob: "src/**/*.ts"`, `Grep: "pattern" --glob "src/**"`
+
+**4. Follow-Up Strategy**
+- When finding a function: do second Grep for key identifiers
+- Search same file for related tokens
+- Read for context if needed
+
+**5. Quality Checklist**
+- Verify token/identifier searches used
+- Verify file:line:content included
+- Verify src/** searched first
+- Verify exact matching code quoted
+
+## Regression Tests
+
+**File**: `tests/unit/truthfulness.test.ts` (+5 tests)
+
+Tests verify explore.md system prompt contains:
+- ✅ Token/identifier search guidance
+- ✅ File:line evidence requirements
+- ✅ `src/**` preference over `tests/**`
+- ✅ Follow-up strategy for function findings
+- ✅ Quality checklist
+
+## Impact
+
+**Before** (phrase search fails):
+```
+User: "Find where taskSuccesses increments"
+=== SUBAGENT TRACE ===
+[1] ✓ Grep: "successful Task"
+    No matches found
+```
+
+**After** (token search succeeds):
+```
+User: "Find where taskSuccesses increments"
+=== SUBAGENT TRACE ===
+[1] ✓ Grep: "taskSuccesses"
+    src/agent/grok-agent.ts:106: taskSuccesses += 1;
+[2] ✓ Read: src/agent/grok-agent.ts
+    [context around line 106]
+```
+
+## Verification
+
+```bash
+✅ npm run build - TypeScript compilation clean
+✅ npm test - 209/209 tests pass (+5 new tests)
+✅ explore.md contains all required guidance strings
+```
+
+---
+
+🤖 Generated with [Claude Code](https://claude.com/claude-code)

.grok/PR-EXPLORE-TOKEN-RELIABILITY.md

@@ -0,0 +1,98 @@
+## Summary
+
+Fixes explore subagent token search reliability by eliminating fragile operator regex patterns that miss code with whitespace.
+
+**Problem**: explore subagent searches for `taskSuccesses++` or `taskSuccesses+=` (no whitespace), but actual code is `taskSuccesses += 1;` (with spaces), causing "no matches found."
+
+**Root Cause**:
+```
+Grep: "taskSuccesses\+\+|taskSuccesses\+="  ← Misses whitespace
+Actual code: taskSuccesses += 1;             ← Has space before =
+Result: No matches found ❌
+```
+
+**Solution**: Use simple token-only searches that find ALL uses (including increments/assignments), then quote the relevant lines.
+
+## Changes
+
+**File**: `.grok/agents/explore.md`
+
+### New Critical Guidance (added to Section 1)
+
+**For increment/assignment questions**:
+- ✅ **ALWAYS start with token ONLY**: `Grep: "taskSuccesses"`
+- ❌ **DO NOT use operator patterns**: `taskSuccesses++`, `taskSuccesses+=` (fragile)
+- ❌ **DO NOT use regex alternation**: `\+\+|\+=` (fragile)
+- ❌ **DO NOT use word boundaries**: `\b`, `\s` (fragile)
+
+**Why this works**:
+- Simple `Grep: "taskSuccesses"` finds **ALL** uses including:
+  - `taskSuccesses += 1;` (increment with spaces)
+  - `const x = taskSuccesses;` (assignment)
+  - `if (taskSuccesses > 0)` (condition)
+- Then quote the relevant lines from grep output
+
+## Before vs After
+
+### Before (broken)
+
+**Subagent**:
+```
+=== SUBAGENT TRACE ===
+[1] ✓ Grep: "taskSuccesses\+\+|taskSuccesses\+="
+    No matches found
+```
+
+**User sees**: "no matches found" and thinks subagent failed
+
+### After (fixed)
+
+**Subagent**:
+```
+=== SUBAGENT TRACE ===
+[1] ✓ Grep: "taskSuccesses"
+    src/agent/grok-agent.ts:106: taskSuccesses += 1;
+    src/agent/grok-agent.ts:112: const subagentsSpawned = taskSuccesses;
+
+[2] ✓ Read: src/agent/grok-agent.ts
+    [context around increment...]
+```
+
+**User sees**: Actual file:line:content evidence ✅
+
+## Regression Tests
+
+**File**: `tests/unit/truthfulness.test.ts` (+3 tests in "Explore Subagent Prompt Quality")
+
+Tests enforce explore.md contains:
+1. ✅ "ALWAYS start with" + "token ONLY" for increments
+2. ✅ "DO NOT use operator patterns" warning
+3. ✅ "DO NOT use regex alternation" warning
+
+## Impact
+
+**Search Reliability**:
+- Before: Fragile patterns miss whitespace variations
+- After: Token-only search finds all uses reliably
+
+**User Experience**:
+- Before: "no matches found" (confusing)
+- After: Actual `file:line:content` evidence (clear)
+
+## Verification
+
+```bash
+✅ npm run build - TypeScript compilation clean
+✅ npm test - 212/212 tests pass (+3 new tests)
+✅ explore.md contains all anti-fragile guidance
+```
+
+**Manual smoke test** (after merge):
+```
+User: "Find where taskSuccesses increments"
+Expected: Grep returns src/agent/grok-agent.ts:106: taskSuccesses += 1;
+```
+
+---
+
+🤖 Generated with [Claude Code](https://claude.com/claude-code)