dev.plan

Execute the implementation planning workflow to generate design artifacts from a feature specification

When to Use

Use this skill when:

• •A spec has been created and clarified via /dev.specify and /dev.clarify
• •You need to resolve technical unknowns before implementation
• •Design artifacts (data model, contracts) are needed before task generation

Invocation

/dev.plan [optional context]

Prerequisites

• •Must be on a feature branch (e.g., ep01-feature-name)
• •spec.md must exist in the feature directory
• •Spec should be clarified (minimal [NEEDS CLARIFICATION] markers)

Workflow

Phase 0: Setup & Environment

Step 0.1: Initialize Environment

# Get feature paths
SCRIPT_DIR="$(dirname "$0")/scripts"
source "$SCRIPT_DIR/common.sh"
eval "$(get_feature_paths)"

# Validate prerequisites
if [[ -z "$FEATURE_SPEC" ]] || [[ ! -f "$FEATURE_SPEC" ]]; then
    echo "Error: No spec.md found. Run /dev.specify first."
    exit 1
fi

Step 0.2: Load Context Documents

Read these files in parallel:

1. •$FEATURE_SPEC - The feature specification
2. •$CONSTITUTION - Project principles (.specify/memory/constitution.md)
3. •Copy templates/plan-template.md → $FEATURE_DIR/plan.md (if not exists)

Step 0.3: Fill Technical Context

In plan.md, populate the Technical Context section with project-specific values:

Mark any unknown fields as [NEEDS CLARIFICATION].

Step 0.4: Constitution Check (Gate 1)

Read $CONSTITUTION and validate the plan against ALL principles.

## Constitution Compliance

| # | Principle | Status | Evidence |
|---|-----------|--------|----------|
| I | [Name] | ✓/✗ | [How this plan complies] |
| II | [Name] | ✓/✗ | [How this plan complies] |
...

Gate Criteria: All principles must pass OR violations must be justified in a "Complexity Tracking" table.

Phase 1: Research & Clarification

Purpose: Resolve all technical unknowns before design

Step 1.1: Extract Unknowns

Parse plan.md Technical Context for:

• •Fields marked [NEEDS CLARIFICATION]
• •Dependencies without version clarity
• •Integrations without defined contracts
• •Performance targets without metrics

Step 1.2: Research Tasks

For each unknown, create a research task:

• •"Research {technology} best practices for {project type}"
• •"Investigate {integration} patterns"
• •"Evaluate {options} for {requirement}"

Use codebase search to find existing patterns:

# Search for similar implementations
grep -r "pattern" src/
# Check existing architecture docs
cat docs/architecture/*.md

Step 1.3: Consolidate Findings

Create $FEATURE_DIR/research.md with format:

# Research Findings: {{FEATURE_NAME}}

## Decision Log

### {{Topic 1}}

**Decision**: [What was chosen]
**Rationale**: [Why chosen]
**Alternatives Considered**: [What else was evaluated]
**References**: [Links to docs, ADRs, etc.]

### {{Topic 2}}
...

Step 1.4: Resolve Remaining Unknowns

Use AskUserQuestion tool for any unresolved items:

• •Present options with recommendations
• •Document answers in research.md
• •Update plan.md Technical Context

Gate: All [NEEDS CLARIFICATION] must be resolved before Phase 2.

Phase 2: Design & Contracts

Prerequisites: research.md complete, no unresolved unknowns

Step 2.1: Data Model Design

Create $FEATURE_DIR/data-model.md:

# Data Model: {{FEATURE_NAME}}

## Entities

### {{Entity Name}}

| Field | Type | Required | Description |
|-------|------|----------|-------------|
| id | string | Yes | Unique identifier |
...

**Relationships**:
- `EntityA` --1:N--> `EntityB`

**Validation Rules**:
- [Rule from requirements]

**State Transitions** (if applicable):
- Initial → Processing → Complete

Extract entities from:

• •spec.md Key Entities section
• •Functional requirements
• •User story acceptance criteria

Step 2.2: API/Contract Design

Create $FEATURE_DIR/contracts/ directory with:

• •Interface definitions (TypeScript)
• •API schemas (OpenAPI/JSON Schema)
• •Event contracts (if applicable)

Format depends on project type:

• •CLI: Command interface definitions
• •Library: Public API types
• •Service: REST/GraphQL schemas

Step 2.3: Quickstart Guide

Create $FEATURE_DIR/quickstart.md:

# Quickstart: {{FEATURE_NAME}}

## Installation

[Steps to set up]

## Basic Usage

[Minimal example]

## Common Patterns

[Frequent use cases]

Step 2.4: Constitution Re-check (Gate 2)

Re-validate design against constitution:

• •ERROR if new violations introduced
• •Document justified violations in "Complexity Tracking"

Step 2.5: Update Plan Status

Update plan.md header:

> **Status**: Design Complete

For Agentic Applications

When planning features for agentic systems (like agentlint), apply these additional design principles:

Tool Design Principles

Per Constitution Principle VII and Anthropic's "Building Effective Agents" guidance:

Data Model for Agentic Features

When designing data models:

Include:

• •Entities that tools operate on (sessions, configs, invocations)
• •Indexes for efficient querying
• •Raw data structures with timestamps, IDs, context

Exclude:

• •Detection results (agent generates these)
• •Orchestration state (agent manages this)
• •Computed judgments ("is_low", "missed_opportunity")

Contract Design

Tool contracts should specify:

// GOOD: Returns data for agent to interpret
interface SkillInvocationResult {
  skillName: string;
  invocations: InvocationRecord[];  // Raw data
  sessionCount: number;              // Fact
  invocationCount: number;           // Fact
  // Agent decides if this is "low" or "high"
}

// BAD: Tool makes judgment
interface SkillAnalysisResult {
  skillName: string;
  status: 'underutilized' | 'healthy' | 'overused';  // Judgment!
  recommendation: string;  // Agent's job!
}

Context Window Economics

Design tools with context window limits in mind:

• •Summarize large results - Don't return full session logs; return summaries
• •Implement pagination - Allow limit and offset parameters
• •Provide filtering - Let agent request only what it needs
• •Use _rawData pattern - Structured data alongside human-readable summary

Constitution Principle VII Checklist

Before finalizing the plan, verify:

• • No tool returns judgments ("low", "bad", "missed")
• • No tool encodes thresholds (if X > 5 then Y)
• • No tool prescribes when to use other tools
• • No tool implements detection/matching logic that requires judgment
• • Tool descriptions explain capabilities, not orchestration

Testing Strategy for Agentic Features

Plan for two levels of testing:

1. •

   Tool-level tests (unit/integration)

   • •Tool returns correct data for given inputs
   • •Tool handles edge cases (empty data, large datasets)
   • •Tool filtering/pagination works correctly
2. •

   Agent-level tests (evaluations)

   • •Agent reasons correctly given tool outputs
   • •Agent makes appropriate tool selections
   • •Agent provides useful analysis (requires LLM-as-judge)

Output Artifacts

On completion, the following files exist:

specs/{epic-id}-{feature}/
├── spec.md           # Input (from /dev.specify)
├── plan.md           # Technical context & constitution check
├── research.md       # Phase 1 output (resolved unknowns)
├── data-model.md     # Phase 2 output (entity definitions)
├── quickstart.md     # Phase 2 output (usage guide)
├── contracts/        # Phase 2 output (API definitions)
│   └── interfaces.ts # Or openapi.yaml, schema.graphql
└── checklists/       # Validation checklists
    └── design.md

Completion Message

Implementation plan complete!

  Epic:     EP01
  Branch:   ep01-feature-name

  Artifacts Generated:
    ✓ plan.md          - Technical context defined
    ✓ research.md      - X decisions documented
    ✓ data-model.md    - X entities defined
    ✓ contracts/       - API interfaces created
    ✓ quickstart.md    - Usage guide ready

  Constitution: All principles pass
  Status: Ready for task generation

Next: Run /dev.tasks to generate implementation tasks

Constitution Alignment

This skill supports:

• •III. Causal-First: Design decisions trace to requirements
• •IV. Minimal: Focus on minimal viable design
• •VI. Traceable: ADR and Arc42 references
• •IX. Agent-Aware: Structured artifacts for agent consumption

Files

• •templates/plan-template.md - Technical context template
• •templates/research-template.md - Research findings template
• •templates/data-model-template.md - Entity model template
• •scripts/common.sh - Shared utilities
• •scripts/setup-plan.sh - Plan initialization

Handoff

After completing this skill, suggest:

• •/dev.tasks - Generate implementation tasks from plan
• •/dev.clarify - If design reveals new spec ambiguities
• •/dev.constitution validate - Optional principle compliance check
• •/dev.analyze plan - Optional quality validation before proceeding