1. Read PROJECT_CONTEXT.md from AI_DOC/
2. Verify current @CurrentState and @TechSpec
3. Check if operation aligns with current Focus

1. Update PROJECT_CONTEXT.md immediately
2. Update @History with new entry
3. Update @CurrentState if status changed

### @ProjectStructure
- `path/file.py`: [Core responsibility] -> [Outputs to/depends on]
- `config.yaml`: [Configuration] -> [Loaded by main.py]

### @CurrentState
- **Status**: [Planning | Coding | Debugging | Refactoring]
- **Focus**: The ONE core problem being solved now
- **Blockers**: Specific errors or dependencies blocking progress

### @TechSpec
- **Data Schemas**: Tensor shapes, API formats, DB schemas
- **Constraints**: Memory limits, hardware specs, performance targets
- **Environment**: OS, CUDA version, language version

### @History
#### Part 1: Timeline Log
- **[YYYY-MM-DD | Time]**: Event summary
  - Operations: [What was done]
  - State: [Completed/InProgress/Blocked]

#### Part 2: Evolution Tree
**[Feature Category]**
**1. [Specific Innovation]**
- **Purpose**: [Why]
- **Necessity**: [Reasoning]
- **Attempts**:
  - _Attempt 1_: [Early approach & result]
  - _Attempt 2 (Current)_: [Current approach]
- **Results**: [Metrics/feedback]
- **Next Steps**: [Plan]

# Task: [Brief Description]
Date: [YYYY-MM-DD HH:MM]

## A. Cognitive Anchors
- Current State: [From PROJECT_CONTEXT.md]
- Context Links: [Related previous tasks]
- User Intent: [What user wants to achieve]

## B. Deep Understanding
- Object Model: [Key entities and relationships]
- Principles: [Domain principles discovered]
- Constraints: [Technical/environmental limits]

## C. Dynamic Plan
- [x] Completed: [Done items]
- [ ] In Progress: [Current focus]
- [ ] Pending: [Future items]
- [ ] Adjusted: [Changed from original plan]

## D. Learning & Discovery
- Aha! Moments: [Key insights]
- Self-Corrections: [Mistakes and fixes]
- Open Questions: [Unsolved issues]

# AI Feedback Log

## [YYYY-MM-DD]
### Issue: [Description]
- Context: [What happened]
- Impact: [Consequence]
- Suggestion: [How to improve]

□ Read PROJECT_CONTEXT.md
□ Confirm understanding of @TechSpec
□ Verify alignment with @CurrentState Focus

□ Sketch logic in pseudocode
□ Write mathematical formulas if applicable
□ Validate logic BEFORE generating actual code

□ Will this modify/delete existing data?
□ Are there irreversible file operations?
□ What happens with empty/abnormal inputs?
□ Is there a rollback/undo strategy?

□ Generate code
□ Update PROJECT_CONTEXT.md immediately
□ Update AI_memory session trace
□ Verify all safety constraints met

# BAD: Bare try-catch
try:
    result = risky_operation()
except:
    pass

# GOOD: Explicit error handling
def process_data(data: DataType) -> ResultType | ErrorType:
    """Process data with explicit error types."""
    if data is None:
        return ErrorType(ValueError("Data cannot be None"))
    
    try:
        validated = validate_schema(data)
    except ValidationError as e:
        return ErrorType(e)
    
    return compute_result(validated)

def function(input_data: Any) -> Result:
    # Entry validation
    assert input_data is not None, "Precondition failed: input_data is None"
    assert len(input_data) > 0, "Precondition failed: empty input"
    
    # Boundary checks
    for item in input_data:
        assert 0 <= item.index < MAX_SIZE, f"Index {item.index} out of bounds"
    
    # Main logic
    ...

# BAD: What (obvious from code)
# Increment counter
counter += 1

# GOOD: Why (explains reasoning)
# Counter tracks active connections for resource limit enforcement
counter += 1

# BAD: Commented-out code
# old_function()
# new_function()

# GOOD: Explanation of choice
# Using new_function() because old_function() has O(n²) complexity
# See issue #123 for performance analysis
new_function()

# NEVER execute these:
rm -rf /
mkfs.*
fdisk
format
dd if=/dev/zero

1. Display full command to be executed
2. Explain what it does
3. Provide undo/reversal strategy
4. Confirm with user if destructive

I need to modify the database schema. Here's my plan:

Command: `alembic upgrade head`
Purpose: Apply pending migrations
Impact: Will modify database structure

Undo strategy: `alembic downgrade -1` to revert

Proceed? [Yes/No/Show migrations first]

# 1. Create project structure
mkdir -p AI_DOC/AI_memory

# 2. Initialize PROJECT_CONTEXT.md
cat > AI_DOC/PROJECT_CONTEXT.md << 'EOF'
### @ProjectStructure
- Root directory initialized, structure TBD

### @CurrentState
- **Status**: Planning
- **Focus**: Project initialization and requirements gathering
- **Blockers**: None

### @TechSpec
- **Environment**: TBD
- **Constraints**: TBD
- **Data Schemas**: TBD

### @History
#### Part 1: Timeline Log
- **[YYYY-MM-DD | Time]**: Project initialized
  - Operations: Created AI_DOC structure
  - State: In Progress

#### Part 2: Evolution Tree
**[Project Foundation]**
**1. Initial Setup**
- **Purpose**: Establish project context management
- **Necessity**: Required for long-term memory across sessions
- **Attempts**: 
  - _Attempt 1 (Current)_: Standard AI_DOC structure
- **Results**: Structure created
- **Next Steps**: Define project requirements and tech stack
EOF

# 3. Create initial AI_FEEDBACK.md
cat > AI_DOC/AI_FEEDBACK.md << 'EOF'
# AI Feedback Log
# Record collaboration improvements here
EOF

def load_project_context(project_path: str) -> Context:
    """Load existing project context."""
    context_file = Path(project_path) / "AI_DOC" / "PROJECT_CONTEXT.md"
    
    if not context_file.exists():
        raise FileNotFoundError(
            "No PROJECT_CONTEXT.md found. "
            "Is this a project-context-managed project?"
        )
    
    # Read and parse context
    content = context_file.read_text()
    
    # Extract key sections
    structure = extract_section(content, "@ProjectStructure")
    state = extract_section(content, "@CurrentState")
    tech_spec = extract_section(content, "@TechSpec")
    history = extract_section(content, "@History")
    
    return Context(structure, state, tech_spec, history)