wogiflow 1.0.24 → 1.0.25

package/.claude/commands/wogi-compact.md

@@ -15,21 +15,6 @@ Root (overview)
     └── File changes (expandable)
 ```
 
-## CRITICAL: Task Queue Check
-
-**Before ANY compaction**, you MUST:
-
-1. Read `.workflow/state/ready.json`
-2. Note pending tasks in your summary
-3. **NEVER claim "nothing pending" without checking ready.json**
-
-If tasks exist:
-- **In Progress**: List task IDs currently being worked on
-- **Ready**: Count + list task IDs awaiting work
-- **Blocked**: Count blocked tasks
-
-This ensures task awareness survives context compaction.
-
 ## Before Compacting
 
 1. **Update progress.md** with:
@@ -63,12 +48,7 @@ Provide this information for the compaction system:
 - [Task/change 1]
 - [Task/change 2]
 
-**Pending Tasks** (REQUIRED - from ready.json):
-- In Progress: [task IDs or "none"]
-- Ready: [count] tasks - [first 5 task IDs]
-- Blocked: [count] tasks
-
-**In Progress (Current)**:
+**In Progress**:
 - TASK-XXX: [description] - [current state, what's left]
 
 **Key Decisions**:
@@ -85,8 +65,6 @@ Provide this information for the compaction system:
 
 **Context to Preserve**:
 - [Important context that should survive compaction]
-
-**ON RESUME**: Check `.workflow/state/ready.json` for pending work.
 ```
 
 ## Context Pressure Monitoring

package/.claude/commands/wogi-extract-review.md

@@ -0,0 +1,175 @@
+# /wogi-extract-review - Zero-Loss Task Extraction with Mandatory Review
+
+Extract tasks from long input with 100% capture rate and mandatory human review.
+
+## Purpose
+
+When processing transcripts, meeting notes, or long prompts, this command ensures NOTHING is missed by:
+1. Capturing EVERY distinct statement (no filtering)
+2. Deduplicating similar items
+3. Requiring explicit human review and confirmation
+4. Only proceeding when user confirms the list is complete
+
+## Philosophy
+
+**OLD approach (lossy):** Input → Filter → Filter → Output (70-80% lost)
+**NEW approach (zero-loss):** Input → Capture All → Dedupe → Review → Confirm → Output (100% captured)
+
+## Usage
+
+```bash
+# Start zero-loss extraction
+flow extract-zero-loss start
+
+# Or pipe content
+cat transcript.txt | flow extract-zero-loss start
+
+# Check review status
+flow extract-zero-loss status
+
+# View items by category
+flow extract-zero-loss show pending
+flow extract-zero-loss show high      # High confidence items
+flow extract-zero-loss show medium    # Medium confidence
+flow extract-zero-loss show low       # Low confidence
+flow extract-zero-loss show filler    # Potential filler
+
+# Review actions
+flow extract-zero-loss confirm <id>           # Confirm as task
+flow extract-zero-loss remove <id> "<reason>" # Remove (reason required!)
+flow extract-zero-loss merge <src> <target>   # Merge duplicate
+
+# Bulk actions
+flow extract-zero-loss confirm-high           # Confirm all high-confidence
+flow extract-zero-loss dismiss-filler         # Dismiss filler items
+
+# Complete review (MANDATORY before proceeding)
+flow extract-zero-loss complete
+
+# Get confirmed tasks
+flow extract-zero-loss tasks
+```
+
+## Review Workflow
+
+### Step 1: Start Extraction
+```
+flow extract-zero-loss start < transcript.txt
+```
+
+This extracts EVERYTHING from the input using multiple strategies:
+- Sentence boundaries
+- Line breaks
+- Speaker changes
+- List items (bullets, numbers)
+- Comma-separated items with action verbs
+
+### Step 2: Quick Review High-Confidence Items
+```
+flow extract-zero-loss show high
+```
+
+These items almost certainly contain tasks. Review and confirm or adjust.
+
+### Step 3: Review Medium-Confidence Items
+```
+flow extract-zero-loss show medium
+```
+
+Many valid tasks here. Review carefully.
+
+### Step 4: Review Low-Confidence Items
+```
+flow extract-zero-loss show low
+```
+
+Some may be tasks phrased informally. Don't skip!
+
+### Step 5: Handle Filler (Optional)
+```
+flow extract-zero-loss dismiss-filler
+```
+
+Dismiss conversational filler like "um", "okay", "thanks".
+
+### Step 6: Confirm Completeness (MANDATORY)
+```
+flow extract-zero-loss complete
+```
+
+User must explicitly confirm the task list is complete before proceeding.
+
+## Integration with Long-Input Processing
+
+After zero-loss extraction and review:
+1. Confirmed tasks become the input for topic extraction
+2. Topics are generated from the confirmed task list
+3. Standard 4-pass processing continues
+
+```bash
+# Full flow
+cat transcript.txt | flow extract-zero-loss start
+flow extract-zero-loss confirm-high
+# ... manual review ...
+flow extract-zero-loss complete
+flow long-input topics    # Now uses confirmed tasks
+flow long-input pass2
+flow long-input pass3
+flow long-input pass4
+```
+
+## Why This Matters
+
+> "When I work with my employees, when we have a meeting, even if it takes an hour or two, when I give a task, an employee will write it down, add a comment on Figma, add it to Jira, write it down in his notebook, but nothing is getting missed."
+
+This system ensures:
+- **100% capture rate** - Nothing is auto-filtered
+- **Explicit confirmation** - User reviews everything
+- **Audit trail** - Track what was confirmed vs removed
+- **Reason required** - Can't remove without explanation
+
+## Confidence Levels
+
+Items are scored (not filtered!) by confidence:
+
+**High Confidence** - Contains explicit requirement signals:
+- "We need to add..."
+- "Should display..."
+- "Must have..."
+- "I would like..."
+- "Change X to Y"
+
+**Medium Confidence** - Contains softer signals:
+- "Maybe we could..."
+- "What if we..."
+- "Going to need..."
+
+**Low Confidence** - No clear signals but may be tasks:
+- Short statements
+- Questions
+- Partial sentences
+
+**Filler** - Conversational noise (still captured!):
+- "Um", "Okay", "Thanks"
+- "Can you hear me?"
+- "Makes sense"
+
+## Files
+
+| File | Location |
+|------|----------|
+| Extraction module | `scripts/flow-zero-loss-extraction.js` |
+| Review module | `scripts/flow-extraction-review.js` |
+| Review session | `.workflow/tmp/long-input/review-session.json` |
+
+## For Claude
+
+When processing long transcripts:
+
+1. **Always use zero-loss extraction** for meeting transcripts and reviews
+2. **Present items by confidence level** to the user
+3. **Never skip low-confidence items** - user must explicitly dismiss
+4. **Require completeness confirmation** before proceeding to topic extraction
+5. **Log all confirmed tasks** for audit trail
+
+The goal is **100% task capture rate**, not 90%.

package/.claude/commands/wogi-init.md

@@ -509,6 +509,9 @@ This file tracks all changes made to the project.
 **Files**: .workflow/*, .claude/*
 ```
 
+**roadmap.md** (future work):
+Copy from `templates/roadmap.md` to `.workflow/roadmap.md`
+
 #### 4.4 Create Spec Files
 
 **stack.md** in `.workflow/specs/`:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "wogiflow",
-  "version": "1.0.24",
+  "version": "1.0.25",
   "description": "AI-powered development workflow management system with multi-model support",
   "main": "lib/index.js",
   "bin": {