---

name: project-tracking description: Establish the files used to track and manage progress. allowed-tools: Read, Write, Bash, Glob

Project Tracking Setup

This skill establishes a token-efficient project tracking system for software projects.

System Overview

Token-Efficient Design: Separates hot state (current session) from cold storage (history).

Core Files:

• CONTEXT.md - Current session state (30-50 lines, read every session)
• IMPLEMENTATION.md - Phase progress tracker (400-600 lines)
• DECISIONS.md - Architectural decisions (heading-based, grep-friendly)
• chronicles/phase-N.md - Detailed session history

Key Benefits:

• 75% reduction in session pickup tokens (~200 lines → ~50 lines)
• 45% reduction in chronicle entry size (36 lines → 15-20 lines)
• Single file read for session pickup (CONTEXT.md only)

When to Use This Skill

• New codebases: Set up tracking from day one
• Existing codebases: Retroactively document project history and establish tracking
• Legacy projects: Migrate from old documentation systems to token-efficient format

Quick Start

PROJECT-TRACKING.md describes the complete token-efficient documentation system.

For New Projects

1. Create base structure (no git history to analyze)
2. Initialize Phase 0 in IMPLEMENTATION.md
3. Ready to document first session

For Existing Projects (Retroactive)

1. Use TodoWrite to track setup progress
2. Analyze project history (see commands below)
3. Identify phase boundaries based on major milestones
4. Create documentation structure in order
5. Document retroactively with appropriate detail level

Step-by-Step Process for Existing Projects

Step 1: Analyze Project History

Run these commands to understand the codebase:

Copy# Full git history
git log --all --oneline --decorate

# Identify major refactors/features
git log --all --oneline --grep="refactor\|feat\|major"

# Check existing documentation
find . -name "*.md" -type f | grep -v node_modules

# Review changelog if exists
cat CHANGELOG.md

# Get commit count and timeline
git log --oneline | wc -l
git log --reverse --format="%ai %s" | head -5
git log --format="%ai %s" | head -5

Step 2: Identify Phase Boundaries

Look for natural breakpoints:

Good phase boundaries:

• Major architectural refactors ("legendary refactor", "restructure")
• Feature milestones (v1.0.0, v2.0.0 releases)
• Significant pivots (monolith → microservices)
• Technology changes (rewrite in new language)

Aim for 3-6 phases for most projects:

• Too few: Loses meaningful structure
• Too many: Fragmented, hard to navigate

Example phase progressions:

• Foundation → Architecture → Features → Production
• MVP → Scale → Polish → Maintenance
• Prototype → Refactor → Features → Optimization

Step 3: Create Files in This Order

Use TodoWrite to track:

Copy1. Create docs/IMPLEMENTATION.md
2. Create docs/DECISIONS.md
3. Create docs/chronicles/ directory
4. Create chronicle files for each phase
5. Create docs/CONTEXT.md (from current phase)
6. Update docs/README.md (if exists)

Note: CHRONICLES.md is eliminated in token-efficient system. Use ls docs/chronicles/ for navigation instead.

Step 4: File Creation Details

IMPLEMENTATION.md

For existing projects (token-efficient format):

• Phase Overview table (all phases)
• Completed phases: 3-5 bullets each + link to chronicles
  • High-level achievements only
  • Link to chronicles/phase-N.md for details
• Current phase: Detailed (200-300 lines)
  • Active tasks with checkboxes
  • What's completed so far
  • What's next
• Future phases: High-level plans (5-10 bullets)

Target size: 400-600 lines (down from 800-1000)

Key change from legacy: Completed phases are compressed to 3-5 bullets, not 80-120 line summaries.

DECISIONS.md

For existing projects (heading-based format):

• Use heading-based format (NOT table format)
• Each decision is a ### DEC-XXX: Title (YYYY-MM-DD) heading
• Include: Status, Context, Decision, Alternatives, Consequences
• Aim for 5-10 decisions for retroactive docs
  • Focus on decisions still relevant today
  • Don't document every minor choice

Extract decisions from:

• Commit messages mentioning "decision", "choose", "vs"
• Architecture changes (modular refactor, tech choices)
• Configuration format choices
• UI/UX pattern choices

Decision Categories to look for:

• Architecture (structure, modules, patterns)
• Configuration (formats, storage)
• Display/UX (interaction patterns, visual design)
• Technology (language, framework, library choices)

Key change from legacy: Heading-based format is grep-friendly and eliminates duplication between table and chronicle entries.

CONTEXT.md

For existing projects:

• Create from current phase section in IMPLEMENTATION.md
• Extract: current focus, active tasks, blockers
• Get latest commit hash and entry number
• Keep under 50 lines total

Sections:

• Frontmatter (phase, updated, last_commit, last_entry)
• Current Focus (1-2 sentences)
• Active Tasks (checkboxes)
• Blockers (if any)
• Context (3-5 bullets)
• Next Session (where to pick up)

This file replaces reading ~200 lines from IMPLEMENTATION.md for session pickup.

chronicles/phase-X-name.md

For existing projects (retroactive entries):

Balance detail vs. effort:

• Older phases (completed long ago):

  • One summary entry per phase
  • High-level "what and why"
  • Light on implementation details
  • Focus on decisions and outcomes

• Recent phases (last 3-6 months):

  • More detailed entries
  • Can document specific features
  • Include code examples if relevant

• Current phase:

  • Full detail from this point forward
  • Document each session

Retroactive entry format (slim template):

Copy## Entry N: [Phase Name] (YYYY-MM-DD)

**What**: [1-2 sentences - what was accomplished in this phase]

**Why**: [1-2 sentences - context/motivation]

**How**: [Bullet points - key implementation details, max 5-7 bullets]

- Key change 1
- Key change 2
- Key change 3

**Decisions**: [Optional - only if DEC-XXX made]

- DEC-XXX: [one-line summary, detail in DECISIONS.md]

**Files**: [key files changed, or "see commits abc-def"]

Target: 15-20 lines per entry (down from 36 lines)

Full template available: Use chronicle-entry-full.md for complex entries if needed.

Step 5: Extract Decisions from Code/History

Commands to find decisions:

Copy# Search commit messages
git log --all --grep="decide\|choose\|vs\|instead\|alternative" --oneline

# Find architectural changes
git log --all --grep="refactor\|architecture\|structure" --oneline

# Check for design docs
find . -name "*DESIGN*" -o -name "*ARCHITECTURE*" -o -name "*ADR*"

# Look for configuration choices
git log --all --grep="config\|yaml\|json\|toml" --oneline

In code, look for:

• Major abstraction choices (OOP vs functional, monolith vs modular)
• Technology selections (which library, which format)
• Pattern implementations (how errors handled, how config loaded)
• Interface designs (API structure, CLI commands)

Step 6: Update Existing Documentation

If project has docs/README.md or similar:

• Add "Development Documentation" section
• Link to CONTEXT.md, IMPLEMENTATION.md, DECISIONS.md
• Add quick paths for contributors
• Explain session pickup workflow (read CONTEXT.md first)

File Size Guidelines (Token-Efficient System)

Key Changes from Legacy:

• CONTEXT.md replaces reading IMPLEMENTATION.md for pickup (200 lines → 50 lines)
• IMPLEMENTATION.md compressed by 40% (completed phases now 3-5 bullets)
• CHRONICLES.md eliminated (use ls chronicles/ instead)
• Chronicle entries 45% shorter (36 lines → 15-20 lines)

Common Pitfalls for Existing Projects

❌ Don't: Try to document every commit ✅ Do: Document major milestones and decisions

❌ Don't: Create 20+ phases for a 2-year project ✅ Do: Aim for 3-6 meaningful phases

❌ Don't: Write novel-length retroactive entries ✅ Do: Write summaries with links to code/commits

❌ Don't: Document every decision ever made ✅ Do: Focus on architectural decisions still relevant

❌ Don't: Spend weeks on retroactive docs ✅ Do: Spend 2-4 hours, then track going forward

Success Criteria

After setup, you should have:

• ✅ CONTEXT.md with current session state (< 50 lines)
• ✅ Clear phase progression in IMPLEMENTATION.md (< 600 lines)
• ✅ Heading-based DECISIONS.md (not table format)
• ✅ 5-10 key decisions documented
• ✅ One chronicle file per phase
• ✅ 1-2 entries per phase (more detail for recent/current)
• ✅ All files cross-referenced
• ✅ Current phase ready for ongoing tracking

What to Tell the User

After setup, summarize:

1. Number of phases identified
2. Number of decisions documented
3. Number of chronicle entries created
4. Where to start for next session: Read CONTEXT.md (30-50 lines)
5. How to use the system going forward
6. Token savings achieved (session pickup now < 50 lines vs ~200 lines)

Migrating Existing Documentation

If project already has legacy documentation (CHRONICLES.md, table-based DECISIONS.md):

Use the /project-tracking:migrate-to-token-efficient command to migrate:

1. Creates CONTEXT.md from current phase
2. Compresses completed phases in IMPLEMENTATION.md
3. Converts DECISIONS.md to heading-based format
4. Eliminates CHRONICLES.md
5. Updates to slim templates

See: commands/migrate-to-token-efficient.md

Templates

See template files for detailed formats:

• CONTEXT.md - Hot state template (NEW)
• chronicle-entry-template.md - Slim entry (15-20 lines)
• chronicle-entry-full.md - Full entry (36 lines, optional)
• decision-entry-template.md - Heading-based decision (NEW)
• DECISIONS.md - Full decisions file template (NEW)

Legacy templates (kept for backward compatibility):

• decision-template.md - Old decision format
• decision-table-row-template.md - Old table row format

References

• See PROJECT-TRACKING.md for complete token-efficient system explanation
• See PLAN-token-efficient-tracking.md for system design rationale