Iterative Development Skill

Supports adding features and analyzing impact without full upfront planning.

When to Load

Load this skill for:

• /katachi:add-feature - Add new feature on-the-go
• /katachi:analyze-impact - Analyze change impact

Philosophy

The framework should support "add as you go" not "define everything upfront":

• Features can be added mid-project
• Dependencies are analyzed dynamically
• Phases recalculate automatically
• Quick-start mode for MVPs

Add Feature Workflow

1. Capture Feature Description

Ask user to describe the feature:

• What does it do?
• Who uses it?
• Any known dependencies?

2. Determine Category

Categories follow the pattern: CATEGORY-NNN

Process:

1. Read existing categories from FEATURES.md
2. Infer category from description
3. Confirm with user

# Example categories (domain-oriented, organized by user capability area)
AUTH - Authentication flows (Login, Logout, Password Reset, Session Timeout)
USER - User management (Registration, Profile, Settings, Account Deletion)
ORDERS - Order management (Create Order, View Orders, Cancel Order)
PAYMENTS - Payment flows (Checkout, Refund, Payment Methods)
ADMIN - Admin capabilities (Manage Users, View Reports, System Settings)
CORE - Core infrastructure (when truly cross-cutting and not user-facing)

If new category needed, confirm with user before creating.

3. Assign ID

Find next available ID in category:

# Check existing IDs
python scripts/features.py status list --category CORE

# Result: CORE-001, CORE-002, CORE-003
# New ID: CORE-004

4. Capture Complexity

Ask user for complexity estimate:

• Easy: 1-2 hours, straightforward
• Medium: Half day, some complexity
• Hard: Full day+, significant complexity

5. Analyze Dependencies

Option A: User knows dependencies

• Ask: "Does this depend on any existing features?"
• Validate dependencies exist

Option B: Agent analysis

• Dispatch katachi:impact-analyzer with feature description
• Agent identifies likely dependencies based on description
• Present to user for confirmation

6. Update FEATURES.md

Add new feature entry:

| CORE-004 | New feature description | Medium | ✗ Defined |

7. Update DEPENDENCIES.md

Add to dependency matrix:

python scripts/features.py deps add-feature CORE-004
python scripts/features.py deps add-dep CORE-004 CORE-001  # If depends on CORE-001

8. Recalculate Phases

Phases recalculate based on dependencies:

python scripts/features.py deps recalculate-phases

Show user the new phase assignment:

• "CORE-004 assigned to Phase 2 (depends on CORE-001)"

9. Offer Next Step

After adding:

• "CORE-004 added. Create spec now? [Y/N]"
• If yes, transition to /katachi:spec-feature CORE-004

Impact Analysis Workflow

1. Capture Change Description

Ask user to describe the proposed change:

• What is being changed?
• Why is this change needed?
• What areas might be affected?

2. Dispatch Impact Analyzer

Task(
    subagent_type="katachi:impact-analyzer",
    prompt=f"""
Analyze the impact of this proposed change:

## Change Description
{change_description}

## FEATURES.md
{features_content}

## DEPENDENCIES.md
{dependencies_content}

## Existing Specs
{list_of_spec_paths}

Trace dependencies and report affected features.
"""
)

3. Present Findings

Show user:

• Directly affected features
• Transitively affected features (dependency chain)
• Documents needing updates
• Risk assessment

4. Ask Next Steps

Based on impact level:

Isolated:

• "This change is isolated to X. Proceed with implementation?"

Moderate:

• "This affects N features. Review affected specs before proceeding?"

Significant:

• "This is a significant change. Create an ADR to document this decision?"

Structural:

• "This affects core architecture. Recommend detailed analysis before proceeding."

Quick-Start Mode

For new projects, offer quick-start:

1. Minimal VISION.md

   • Problem statement
   • MVP scope (not full scope)
   • Key workflows (top 3)

2. MVP Features Only

   • Extract only features needed for MVP
   • Skip nice-to-haves
   • Aim for 5-10 features max

3. Simple Dependencies

   • Phase 1 = MVP
   • Linear dependencies where possible
   • Skip complex dependency analysis

4. First Feature Guidance

   • Guide through first spec
   • Establish patterns early
   • User learns workflow on real work