SuperAPI AI Directory

blueprint

General & other

View repository

Updated 10-16 00:03

About this skill

Create a Blueprint

CRITICAL: This is a WORKFLOW SKILL, not Claude's built-in plan mode. Do NOT use EnterPlanMode or ExitPlanMode tools.

Task Tracking Setup

TASK_TRACKING = config_read("task_tracking.enabled", "false")

If TASK_TRACKING:
  BLUEPRINT_WORKFLOW_ID = "blueprint-{timestamp}"
  PHASE_TASKS = {}

PHASES = [ {num: 1, name: "Discovery", active: "Running discovery"}, {num: 2, name: "Research", active: "Researching codebase"}, {num: 3, name: "Research Validation", active: "Validating research"}, {num: 4, name: "Architecture", active: "Designing architecture"}, {num: 5, name: "Capture Learnings", active: "Capturing learnings"}, {num: 6, name: "Write Plan", active: "Writing plan"}, {num: 7, name: "Review Plan", active: "Reviewing plan"}, {num: 8, name: "User Decision", active: "Awaiting user decision"}, {num: 9, name: "Execution", active: "Executing plan"} ]

For each P in PHASES: PHASE_TASKS[P.num] = TaskCreate( subject: "Phase {P.num}: {P.name}", activeForm: P.active, metadata: {workflow: BLUEPRINT_WORKFLOW_ID, phase: P.num} )

Dependencies: phase 4 blocked by 1-3, then sequential 5→6→7→8→9

TaskUpdate(PHASE_TASKS[4], addBlockedBy: [PHASE_TASKS[1], PHASE_TASKS[2], PHASE_TASKS[3]]) TaskUpdate(PHASE_TASKS[5], addBlockedBy: [PHASE_TASKS[4]]) TaskUpdate(PHASE_TASKS[6], addBlockedBy: [PHASE_TASKS[5]]) TaskUpdate(PHASE_TASKS[7], addBlockedBy: [PHASE_TASKS[6]]) TaskUpdate(PHASE_TASKS[8], addBlockedBy: [PHASE_TASKS[7]]) TaskUpdate(PHASE_TASKS[9], addBlockedBy: [PHASE_TASKS[8]])


## Feature Description

<feature_description> $ARGUMENTS </feature_description>

**If empty:** Ask user to describe the feature, bug fix, or improvement.

## Workflow

### Phase 1: Discovery

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[1], status: "in_progress")

Skill("blueprint-discovery")


Handles: Idea Refinement, Interview decision, Acceptance Criteria, Feature Classification.

**Wait for output:** `discovery_result`

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[1], status: "completed")

### Phase 2: Research

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[2], status: "in_progress")

Skill("blueprint-research") feature_description, discovery_result


Handles: Toolbox resolution, lessons discovery, parallel research agents, spec review.

**Wait for output:** `research_result`

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[2], status: "completed")

### Phase 3: Research Validation

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[3], status: "in_progress")

Quick checkpoint before architecture. Catches misalignment early.

research_summary = summarize research_result:

  • Key patterns found (2-3 bullets)
  • Similar implementations discovered
  • Potential constraints/blockers

AskUserQuestion: question: "Research found: {research_summary}. Continue to planning?" header: "Validate" options: - label: "Looks good" description: "Proceed to architecture phase" - label: "Missing something" description: "Tell me what to research deeper" - label: "Wrong direction" description: "Let's revisit the requirements"


**If "Missing something":**

AskUserQuestion: "What should I dig into?" → Run targeted research agent → Append to research_result → Return to Phase 3


**If "Wrong direction":**

→ Return to Phase 1 (Discovery)


If TASK_TRACKING: TaskUpdate(PHASE_TASKS[3], status: "completed")

### Phase 4: Architecture

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[4], status: "in_progress")

Task(majestic-engineer:plan:architect): prompt: | Feature: {feature_description} Research: {research_result.research_findings} Spec: {research_result.spec_findings} Skills: {research_result.skill_content} Lessons: {research_result.lessons_context}


**Wait for output:** `architect_output`

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[4], status: "completed")

### Phase 5: Capture Learnings

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[5], status: "in_progress")

PRIMARY_DIR = extract main folder from architect_output.file_recommendations AGENTS_MD = walk up from PRIMARY_DIR until AGENTS.md found (default: root)

If new patterns or decisions discovered: Edit(AGENTS_MD, append patterns/decisions)


If TASK_TRACKING: TaskUpdate(PHASE_TASKS[5], status: "completed")

### Phase 6: Write Plan

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[6], status: "in_progress")

Apply `plan-builder` skill.

Select template based on complexity:
- **Minimal:** Single file changes, simple bugs
- **Standard:** Most features (default)
- **Comprehensive:** Major features, architectural changes

**Output:** Write to `docs/plans/[YYYYMMDDHHMMSS]_<title>.md`

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[6], status: "completed")

### Phase 7: Review Plan

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[7], status: "in_progress")

Task(majestic-engineer:plan:plan-review): prompt: "Review plan at {plan_path}"


Incorporate feedback and update plan file.

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[7], status: "completed")

### Phase 8: User Decision

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[8], status: "in_progress")

Check auto_preview: `config_read("auto_preview", "false")`

If `true`: `Bash(command: "open {plan_path}")`

AskUserQuestion: question: "Blueprint ready at {plan_path}. What next?" options: - "Build as single task" → Phase 9a - "Break into small tasks" → Phase 9b - "Create as single epic" → Phase 9c - "Deep dive into specifics" → Phase 8.1 - "Preview plan" → Read and display, return to Phase 8 - "Revise" → Ask what to change, return to Phase 6 - "Review and refine" → Apply document-refinement skill, return to Phase 8


**IMPORTANT:** After user selects, EXECUTE that action.

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[8], status: "completed")

**If "Review and refine":**

Apply document-refinement skill to {plan_path} → Auto-fix minor issues in plan file → Present refinement report to user → Return to Phase 8


### Phase 8.1: Deep Dive

AskUserQuestion: "What aspect needs deeper research?"


Run focused research:

Task(majestic-engineer:research:best-practices-researcher): prompt: "Deep dive: {aspect} for feature {feature}" Task(majestic-engineer:research:web-research): prompt: "{aspect} - patterns, examples, gotchas"


Update plan with findings, return to Phase 8.

### Phase 9: Execution

If TASK_TRACKING: TaskUpdate(PHASE_TASKS[9], status: "in_progress")

Dispatch on `user_choice` from Phase 8:

**9a. Single Task** (`user_choice == "Build as single task"`):

Skill("build-task") "{plan_path}"


End workflow.

**9b. Break Into Tasks** (`user_choice == "Break into small tasks"`):

Task(majestic-engineer:plan:task-breakdown): prompt: "Plan: {plan_path}"

AskUserQuestion: question: "Tasks added to plan. Create in backlog?" options: - "Yes, create tasks" → For each task in plan: Skill("backlog-manager") - "No, just the plan" → Skip


Then offer build:

AskUserQuestion: question: "Start building?" options: - "Build all tasks now" → Skill("run-blueprint") "{plan_path}" - "Done for now" → End workflow


**9c. Single Epic** (`user_choice == "Create as single epic"`):

Skill("backlog-manager")


Update plan with task reference, then offer build:

AskUserQuestion: question: "Start building?" options: - "Build all tasks now" → Skill("run-blueprint") "{plan_path}" - "Done for now" → End workflow


If TASK_TRACKING: TaskUpdate(PHASE_TASKS[9], status: "completed")

## Error Handling

| Scenario | Action |
|----------|--------|
| Discovery skill fails | Ask user for AC manually, continue |
| Research skill fails | Continue with architect (degraded) |
| User skips validation | Continue to Phase 4 |
| Architect fails | Log error, ask user to provide approach |
| Plan-review fails | Continue with original plan |
| Task creation fails | Report error, ask user to create manually |
| Config read fails | Use default value |
| User cancels | End workflow gracefully |

## Notes

- Blueprint-only skill - no implementation code
- Research and spec-review run in parallel (inside blueprint-research)
- Architect MUST wait for research completion
- All phases are mandatory except branches (8.1, 9a/9b/9c)
- Always execute user's selected option