docs: archive completed phases to COMPLETED.md

Move Phases 17.5-17.8 and 18.1-18.2 from TODO.md to docs/COMPLETED.md:
- Phase 17.5: Smart Directory Navigation (pcd command)
- Phase 17.6: Enhanced PCD Algorithm
- Phase 17.7: PCD Inline Predictions & Relative Paths
- Phase 17.8: Partial Command Predictions
- Phase 18.1: Dynamic Workflow Learning (already archived)

Updated TODO.md Current Status to show v0.3.0 features as archived.
Removed ~320 lines of completed implementation details.
TODO.md now focuses only on planned work (17.9, 17.10, 18.3+).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: lucaspimentel

TODO.md

@@ -9,21 +9,24 @@ This document tracks active and planned work for PSCue. For architectural detail
 ## Current Status
 
 **Latest Release**: v0.3.0 (2025-11-09)
-- ✅ Multi-word prediction suggestions (Phase 17.4)
-- ✅ Dynamic workflow learning (Phase 18.1)
-- ✅ Smart directory navigation with `pcd` command (Phases 17.5-17.7)
-- ✅ Enhanced path handling and fuzzy matching
+- ✅ Multi-word prediction suggestions (Phase 17.4) - archived to docs/COMPLETED.md
+- ✅ Dynamic workflow learning (Phase 18.1) - archived to docs/COMPLETED.md
+- ✅ Smart directory navigation with `pcd` command (Phases 17.5-17.8) - archived to docs/COMPLETED.md
+  - Phase 17.5: Basic `pcd` command with tab completion
+  - Phase 17.6: Enhanced algorithm (fuzzy matching, frecency, distance scoring)
+  - Phase 17.7: Inline predictions and relative paths
+  - Phase 17.8: Partial command predictions (typing "g" suggests "git")
 - ✅ Comprehensive test coverage (300+ tests)
 - 🎉 Available for installation via one-line command
 
 **Previous Release**: v0.2.0 (2025-11-05)
-- ML-based N-gram sequence prediction (Phase 17.1)
-- Privacy & security - sensitive data protection (Phase 17.2)
-- Partial word completion filtering (Phase 17.3)
+- ML-based N-gram sequence prediction (Phase 17.1) - archived
+- Privacy & security - sensitive data protection (Phase 17.2) - archived
+- Partial word completion filtering (Phase 17.3) - archived
 - Automated CI/CD with GitHub Actions
 
 **Recent Improvements** (not yet released):
-- None currently - all features shipped in v0.3.0
+- None currently - all v0.3.0 features archived to docs/COMPLETED.md
 
 **Installation**:
 ```powershell
@@ -34,329 +37,6 @@ irm https://raw.githubusercontent.com/lucaspimentel/PSCue/main/scripts/install-r
 
 ## Planned Work
 
-### Phase 17.5: Smart Directory Navigation (`pcd` command)
-**Status**: ✅ **COMPLETE** (2025-11-08)
-
-Add a PowerShell function with smart tab completion for directory navigation, leveraging PSCue's learned directory data without interfering with native `cd` completion.
-
-**Completed**:
-- ✅ Created `module/Functions/PCD.ps1` with `Invoke-PCD` function
-  - ✅ Function logic: calls `Set-Location` with argument
-  - ✅ Fallback to native behavior if no learned data available
-  - ✅ Handles empty argument (cd to home directory)
-  - ✅ Tab completion: `Register-ArgumentCompleter -CommandName` (in-process)
-    - ✅ Queries `PSCueModule.KnowledgeGraph.GetSuggestions("cd", @())`
-    - ✅ Filters by `$wordToComplete` (StartsWith matching)
-    - ✅ Returns `CompletionResult` objects with usage count and last used date tooltip
-  - ✅ Created alias: `pcd` (PowerShell Change Directory)
-- ✅ Updated `module/PSCue.psd1`:
-  - ✅ Added to `FunctionsToExport`: `'Invoke-PCD'`
-  - ✅ Added to `AliasesToExport`: `'pcd'`
-  - ✅ Also added missing workflow management functions to exports
-- ✅ Created comprehensive tests in `test/PSCue.Module.Tests/PCDTests.cs`:
-  - ✅ 19 tests covering all functionality
-  - ✅ ArgumentGraph integration tests
-  - ✅ Tab completion simulation tests
-  - ✅ Scoring and ranking tests
-  - ✅ Edge cases and performance tests
-- ✅ Documentation:
-  - ✅ Updated README.md with `pcd` command usage and examples
-  - ✅ Updated CLAUDE.md quick reference
-
-**Files Modified**:
-- `module/Functions/PCD.ps1` (new file)
-- `module/PSCue.psd1` (added function/alias exports)
-- `test/PSCue.Module.Tests/PCDTests.cs` (new file)
-- `README.md` (added pcd usage section)
-- `CLAUDE.md` (added pcd to function reference)
-
-**Design Decisions**:
-- **In-process completion**: Direct call to `PSCueModule.KnowledgeGraph` (<1ms, no IPC overhead)
-- **Non-invasive**: Separate command (`pcd`), doesn't interfere with native `cd` tab completion
-- **Reuses existing data**: ArgumentGraph already tracks `cd` command with directory normalization
-- **Name choice**: `pcd` = PowerShell/PSCue Change Directory (alternatives: `scd`, `lcd`, `qcd`, `cue`)
-- **Display**: PSReadLine handles menu display (`MenuComplete`, `ListView`) - we just return ordered results
-
-**Note**: See Phase 17.6 below for enhanced algorithm improvements (fuzzy matching, frecency scoring, distance awareness).
-
-### Phase 17.6: Enhanced PCD Algorithm
-**Status**: ✅ **COMPLETE** (2025-11-08)
-
-Enhance the `pcd` command with advanced matching, scoring, and navigation features.
-
-**Goals**:
-1. **Well-known shortcuts**: Fast handling of `~`, `..`, `.` with highest priority
-2. **Fuzzy matching**: Levenshtein distance + substring matching for flexible directory finding
-3. **Frecency scoring**: Configurable blend of frequency + recency + distance
-4. **Distance scoring**: Boost directories closer to current location (parent/child/sibling)
-5. **Recursive search**: Optional deep filesystem search for matching directories
-6. **Best-match navigation**: `pcd <name>` finds best match if no exact path exists
-
-**Implementation Plan**:
-
-**Core Engine** (✅ COMPLETE):
-- [x] Create `src/PSCue.Module/PCDCompletionEngine.cs` with multi-stage algorithm
-  - [x] Stage 1: Well-known shortcuts (~, .., .) with priority 1000/999/998
-  - [x] Stage 2: Learned directories with enhanced scoring
-    - [x] Match scoring: Exact (1.0) > Prefix (0.9) > Fuzzy (0.0-0.8)
-    - [x] Fuzzy matching: Substring + Levenshtein distance
-    - [x] Frecency: Configurable weights (default: 50% frequency, 30% recency, 20% distance)
-    - [x] Distance scoring: Same dir (1.0), Parent (0.9), Child (0.85-0.5), Sibling (0.7), Ancestor (0.6-0.1)
-  - [x] Stage 3: Recursive filesystem search (optional, configurable depth)
-
-**PowerShell Integration** (✅ COMPLETE):
-- [x] Update `module/Functions/PCD.ps1` (205 lines total)
-  - [x] Tab completion: Use `PcdCompletionEngine.GetSuggestions()`
-    - [x] Create engine instance with config from env vars
-    - [x] Call `GetSuggestions($wordToComplete, $currentDirectory, 20)`
-    - [x] Convert `PcdSuggestion` objects to `CompletionResult`
-    - [x] Show match type indicators in tooltips
-  - [x] `Invoke-PCD` function: Smart path resolution with best-match fallback
-    - [x] If path doesn't exist, use engine to find best match
-    - [x] Show message: "No exact match, navigating to: <best-match>"
-    - [x] Fall back to Set-Location error if no good matches
-  - [x] Read configuration from environment variables
-
-**Testing** (✅ COMPLETE):
-- [x] Create `test/PSCue.Module.Tests/PcdEnhancedTests.cs`
-  - [x] Well-known shortcuts tests
-  - [x] Fuzzy matching tests
-  - [x] Frecency scoring tests
-  - [x] Distance scoring tests
-  - [x] Recursive search tests
-  - [x] Best-match navigation tests
-  - [x] Performance tests
-  - [x] Edge cases and error handling
-  - [x] Tooltip and display tests
-  - [x] All tests passing ✅
-
-**Documentation** (✅ COMPLETE):
-- [x] Update `README.md`: Enhanced features section with full algorithm details
-- [x] Update `CLAUDE.md`: Algorithm details and file references
-- [x] Update `TODO.md`: Mark Phase 17.6 complete
-
-**Environment Variables**:
-```powershell
-# Scoring weights (must sum to ~1.0 with match weight)
-$env:PSCUE_PCD_FREQUENCY_WEIGHT = "0.5"  # Default: 0.5 (50%)
-$env:PSCUE_PCD_RECENCY_WEIGHT = "0.3"    # Default: 0.3 (30%)
-$env:PSCUE_PCD_DISTANCE_WEIGHT = "0.2"   # Default: 0.2 (20%)
-
-# Recursive search (disabled by default for performance)
-$env:PSCUE_PCD_RECURSIVE_SEARCH = "true"  # Default: false
-$env:PSCUE_PCD_MAX_DEPTH = "3"            # Default: 3 levels deep
-```
-
-**Files Modified**:
-- `src/PSCue.Module/PcdCompletionEngine.cs` (new file) ✅
-- `module/Functions/PCD.ps1` (enhanced) ✅
-- `test/PSCue.Module.Tests/PcdEnhancedTests.cs` (new file) ✅
-- `README.md` (enhanced pcd section) ✅
-- `CLAUDE.md` (algorithm details + file references) ✅
-- `TODO.md` (Phase 17.6 marked complete) ✅
-
-**Key Achievements**:
-- Multi-stage completion algorithm: Shortcuts → Learned → Recursive
-- Fuzzy matching with Levenshtein distance + substring matching
-- Configurable frecency scoring (frequency + recency + distance)
-- Distance-aware scoring for nearby directories
-- Best-match navigation without Tab completion
-- Comprehensive test coverage for all features
-- Full documentation in README and CLAUDE.md
-- Performance targets met: <10ms tab completion, <50ms best-match
-
-### Phase 17.7: PCD Inline Predictions & Relative Paths
-**Status**: ✅ **COMPLETE** (2025-11-09)
-
-Enhance `pcd` with inline predictions and relative path display.
-
-**Completed**:
-- [x] Inline predictor integration
-  - [x] Add pcd command detection to CommandPredictor
-  - [x] Create GetPcdSuggestions() method
-  - [x] Support both "pcd" and "Invoke-PCD" commands
-  - [x] Show suggestions even for "pcd<space>" (always show predictions)
-  - [x] Return multiple suggestions (up to 5 for predictor)
-  - [x] Add environment variable helper methods
-- [x] Relative path conversion
-  - [x] Add ToRelativePath() method to PcdCompletionEngine
-  - [x] Convert parent directory to ".."
-  - [x] Convert child directories to "./subdir" when shorter
-  - [x] Convert sibling directories to "../sibling" when shorter
-  - [x] Fall back to absolute path if relative isn't shorter
-  - [x] Apply to both tab completion and inline predictor
-  - [x] Keep full paths in DisplayPath and tooltips
-- [x] Filter unhelpful suggestions
-  - [x] Don't suggest current directory (".")
-  - [x] Don't suggest previous directory ("-")
-  - [x] Don't suggest path that matches current directory
-
-**Files Modified**:
-- `src/PSCue.Module/CommandPredictor.cs` (added GetPcdSuggestions, env helpers)
-- `src/PSCue.Module/PcdCompletionEngine.cs` (added ToRelativePath, updated suggestions)
-- `test/PSCue.Module.Tests/PcdEnhancedTests.cs` (updated for filtering behavior)
-
-**Key Achievements**:
-- Inline predictions work like other commands (git, gh, etc.)
-- Relative paths reduce visual noise significantly
-- Full context still available in tooltips
-- Performance maintained: <10ms for predictor responses
-
-### Phase 17.8: Partial Command Predictions
-**Status**: ✅ **COMPLETE** (2025-11-09)
-**Priority**: Medium
-**Estimated Effort**: 8-12 hours (Actual: ~2 hours)
-
-**Goal**: Show inline predictions for partial commands (e.g., typing "g" suggests "git", "gh", "gt").
-
-**Why This Matters**:
-- Saves keystrokes for frequently-used commands
-- Improves discoverability of available commands
-- Complements existing argument/workflow predictions
-
-**Current Behavior**: Inline predictions only show after command + space (e.g., "git "). Partial commands like "g" get no inline suggestions.
-
-**Architecture**:
-```csharp
-// Extend src/PSCue.Module/CommandPredictor.cs (~100 lines)
-public override SuggestionPackage GetSuggestion(
-    PredictionClient client,
-    PredictionContext context,
-    CancellationToken cancellationToken)
-{
-    var commandLine = context.InputAst.Extent.Text;
-
-    // NEW: If line contains no space, suggest full commands
-    if (!commandLine.Contains(' '))
-    {
-        var partial = commandLine.Trim();
-        if (string.IsNullOrEmpty(partial))
-            return null; // Empty prompt = workflow predictions only
-
-        return GetPartialCommandSuggestions(partial);
-    }
-
-    // Existing: Suggest arguments/next commands
-    // ...
-}
-
-private SuggestionPackage GetPartialCommandSuggestions(string partial)
-{
-    // 1. Query ArgumentGraph for commands starting with partial
-    // 2. Query WorkflowLearner for recently used commands
-    // 3. Filter by StartsWith(partial, StringComparison.OrdinalIgnoreCase)
-    // 4. Score by frequency + recency
-    // 5. Return top 5 suggestions
-}
-```
-
-**Data Sources**:
-1. **Learned Commands**: Query `ArgumentGraph` for commands matching partial
-   - Use existing frequency + recency scoring
-2. **Workflow History**: Query `WorkflowLearner` for recently executed commands
-   - Boost commands used in last 10 minutes (1.5× multiplier)
-3. **Known Commands**: Fallback to PSCue's hardcoded command list (git, gh, az, etc.)
-   - Lower priority (0.5× multiplier) - prefer learned commands
-
-**Tasks**:
-1. [ ] Add partial command detection to `CommandPredictor.GetSuggestion()` (~30 lines)
-2. [ ] Implement `GetPartialCommandSuggestions()` method (~70 lines)
-   - [ ] Query ArgumentGraph for matching commands
-   - [ ] Query WorkflowLearner for recent commands
-   - [ ] Apply filtering and scoring
-   - [ ] Return top 5 suggestions
-3. [ ] Add configuration environment variable
-   - [ ] `PSCUE_PARTIAL_COMMAND_PREDICTIONS` (default: true)
-4. [ ] Write tests (~10 test cases)
-   - [ ] Single-character partial ("g" → "git", "gh")
-   - [ ] Multi-character partial ("do" → "docker", "dotnet")
-   - [ ] Case-insensitive matching ("GI" → "git")
-   - [ ] Scoring accuracy (frequency + recency)
-   - [ ] Empty prompt (no suggestions)
-   - [ ] Workflow boost for recent commands
-5. [ ] Documentation updates
-   - [ ] README: Add partial command section
-   - [ ] CLAUDE.md: Update feature list
-
-**Example Behavior**:
-```powershell
-# User types "g" (no space, no Tab)
-PS> g█
-    # Inline predictions:
-    # → git     (most frequent)
-    # → gh      (second most frequent)
-    # → gt      (least frequent)
-
-# User types "doc" (no space, no Tab)
-PS> doc█
-    # → docker  (exact prefix match)
-
-# User types "GI" (case-insensitive)
-PS> GI█
-    # → git    (case-insensitive match)
-
-# Empty prompt - no command suggestions (only workflow predictions)
-PS> █
-    # → git status  (workflow prediction, not command prediction)
-```
-
-**Performance Targets**:
-- Query time: <5ms (same as existing inline predictions)
-- No noticeable typing lag
-- Efficient filtering with StartsWith() + case-insensitive comparison
-
-**Edge Cases**:
-- Empty prompt: No command suggestions (reserve for workflow predictions only)
-- Single space: No command suggestions (move to argument suggestions)
-- Very long partial (10+ chars): Likely not a command, no suggestions
-- No matching commands: Return empty (no fallback noise)
-
-**Configuration**:
-```powershell
-# Enable/disable partial command predictions (default: true)
-$env:PSCUE_PARTIAL_COMMAND_PREDICTIONS = "true"
-```
-
-**Dependencies**: None (uses existing ArgumentGraph + WorkflowLearner)
-
-**Success Criteria**:
-- ✅ Partial commands show relevant suggestions
-- ✅ Scoring prioritizes frequent/recent commands
-- ✅ Performance meets <5ms target
-- ✅ No interference with existing workflow predictions
-- ✅ All tests passing
-- ✅ User can disable feature via environment variable
-
-**Follow-up Questions** (to test after implementation):
-- Is this helpful or noisy? (May need user feedback to decide)
-- Should we show command suggestions at empty prompt? (Currently reserved for workflows)
-- Should we limit to commands with high confidence? (Avoid suggesting rarely-used commands)
-
-**Completed**:
-- [x] Added `GetPartialCommandSuggestions()` method to CommandPredictor
-- [x] Integrated with existing suggestion pipeline (merges with genericSuggestions)
-- [x] Environment variable configuration: `PSCUE_PARTIAL_COMMAND_PREDICTIONS` (default: true)
-- [x] Scoring algorithm: 60% frequency + 40% recency, with 1.5× boost for recently used commands
-- [x] Filters: Very long partials (>10 chars) ignored, case-insensitive matching
-- [x] Comprehensive unit tests (7 test cases covering core functionality)
-- [x] Performance: <5ms target met (reuses existing ArgumentGraph queries)
-
-**Files Modified**:
-- `src/PSCue.Module/CommandPredictor.cs` (~100 lines added)
-  - `GetPartialCommandSuggestions()` method
-  - `FormatLastUsed()` helper method
-  - Integration into GetSuggestion() workflow
-- `test/PSCue.Module.Tests/CommandPredictorTests.cs` (7 new tests)
-
-**Key Achievements**:
-- Frequency-based command suggestions from learned data
-- Works without workflow context (fresh sessions, unrelated commands)
-- Complements existing workflow predictions
-- User-friendly tooltips show usage stats ("Used X times, last: Xm ago")
-- Can be disabled via environment variable
-
----
-
 ### Phase 17.9: Tab Completion on Empty Input
 **Status**: Planned
 **Priority**: Medium-Low
@@ -445,27 +125,9 @@ PS> █  # Press Tab
 - [ ] Background pre-computation for heavy ML
 - [ ] Research ONNX Runtime + NativeAOT compatibility
 
-### Phase 18.1: Dynamic Workflow Learning
-**Status**: ✅ **COMPLETE** (2025-11-07)
-
-**Completed**:
-- ✅ WorkflowLearner.cs core infrastructure (~500 lines)
-- ✅ SQLite schema extension (workflow_transitions table - 8 tables total)
-- ✅ Module initialization and lifecycle integration
-- ✅ FeedbackProvider integration (automatic learning)
-- ✅ CommandPredictor integration (inline predictions at empty prompt)
-- ✅ Configuration via environment variables (4 new vars)
-- ✅ PersistenceManager save/load methods
-- ✅ Comprehensive testing (35+ unit tests)
-- ✅ PowerShell module functions (5 functions: Get/Stats/Clear/Export/Import)
-- ✅ Documentation updates (CLAUDE.md, README.md, TODO.md)
-
-**Phase 18.2 (Time-Based Detection)** also completed as part of 18.1:
-- ✅ Time-sensitive scoring with timing pattern awareness
-- ✅ Time delta tracking in database
-- ✅ Confidence adjustments based on timing (1.5× to 0.8× boost)
-
-**See**: [COMPLETED.md](docs/COMPLETED.md) for detailed implementation notes
+---
+
+## Phase 18: Workflow Improvements
 
 ### Phase 18.3: Workflow Chains (3+ Commands)
 **Status**: Planned