mhylle/prompt-generator

Prompt Generator

Generate structured implementation prompts for phase-based project execution using an orchestrator/subagent pattern.

Quick Start

When triggered, gather these inputs from the user:

| Variable | Description | Example | |----------|-------------|---------| | PHASE_NUMBER | Phase identifier | "1", "2", "3" | | PHASE_NAME | Descriptive phase name | "Foundation", "Data Pipeline" | | PHASE_DOC_PATH | Path to phase plan document | /project/docs/plans/01-foundation.md | | PROJECT_ROOT | Project root directory | /home/user/myproject | | GENERAL_PLAN_PATH | Path to general plan (optional) | /project/docs/plans/00-general-plan.md | | ADR_PATH | Path to ADR directory (optional) | /project/docs/decisions/ |

ADR Integration: The generated prompt will instruct the orchestrator to:

• Read any ADRs referenced in the plan before implementation
• Create new ADRs when architectural decisions arise during implementation
• Update the plan with ADR references when deviating from the original design

Workflow

1. Gather User Input

Ask the user for required variables. If context provides these values, confirm them:

To generate the implementation prompt, I need:
1. Phase number (e.g., 1, 2, 3)
2. Phase name (e.g., "Foundation", "Data Pipeline")
3. Path to the phase document
4. Project root directory
5. Path to general plan (optional)

2. Generate the Prompt

Read the template from references/implementation-prompt-template.md and substitute all {PLACEHOLDER} values with user-provided inputs.

3. Output and Save

1. Display: Output the complete generated prompt in the chat
2. Save: Use scripts/save_prompt.py to save to <PROJECT_ROOT>/docs/prompts/

echo "<generated_prompt>" | python scripts/save_prompt.py <project_root> <phase_number> <phase_name>

The file is saved as: docs/prompts/phase-<N>-<name>.md

Template Features

The generated prompt includes:

• Orchestration requirements: Main session coordinates, subagents implement
• Delegation patterns: File creation, testing, infrastructure, integration
• Progress tracking: TodoWrite integration for task management
• Validation workflow: Group validation and final success criteria checks
• Error handling: Failure analysis, rollback procedures, escalation guidance
• Handoff preparation: Documentation for next phase transition
• Plan updates: Update implementation plan with actual outcomes upon completion

Subagent Patterns Reference

| Task Type | Pattern | Concurrency | |-----------|---------|-------------| | File creation (independent) | Parallel | High (5-7) | | File creation (dependent) | Sequential | Low (1-2) | | Testing/Validation | Per test suite | Medium (3-5) | | Docker/Infrastructure | Per service | Medium (3-4) | | Integration | Per integration point | Low (2-3) |

Output Format

The generated prompt follows this structure:

Implement Phase {N}: {Name} of the trading platform...

IMPORTANT ORCHESTRATION REQUIREMENTS:
1. This is an ORCHESTRATION SESSION...
2. Use the implement-plan skill...
3. Spawn SUBAGENTS for all implementation...
4. Track progress using TodoWrite...
5. Coordinate subagents and validate...

## Implementation Strategy
[Phase overview, orchestration workflow, delegation patterns]

## Error Handling
[Failure analysis, rollback, escalation]

## Success Criteria Validation
[Automated and manual checks]

## Handoff to Next Phase
[Update implementation plan, completion documentation, context prep, clean state]

Integration with implement-plan/implement-phase

Generated prompts are automatically discovered and used by the implementation skills:

Workflow

1. prompt-generator creates: docs/prompts/phase-2-data-pipeline.md
                                    ↓
2. implement-plan discovers prompt via Glob("docs/prompts/phase-*.md")
                                    ↓
3. implement-plan passes prompt path to implement-phase
                                    ↓
4. implement-phase uses prompt for detailed orchestration instructions
                                    ↓
5. On completion, implement-phase archives to: docs/prompts/completed/

Naming Convention

Prompts must follow this naming pattern for auto-discovery:

docs/prompts/phase-<N>-<name>.md

Examples:
  docs/prompts/phase-1-foundation.md      → Phase 1
  docs/prompts/phase-2-data-pipeline.md   → Phase 2
  docs/prompts/phase-3-agent-system.md    → Phase 3

Directory Structure

docs/prompts/
├── phase-1-foundation.md        # Pending - ready for implementation
├── phase-2-data-pipeline.md     # Pending - ready for implementation
├── phase-3-agent-system.md      # Pending - ready for implementation
└── completed/                   # Archived after successful completion
    ├── phase-1-foundation.md    # Completed
    └── phase-2-data-pipeline.md # Completed

Benefits

• Pre-planning: Generate all prompts upfront before implementation
• Consistency: Same orchestration patterns across all phases
• Tracking: Completed folder shows implementation progress
• Review: Archived prompts document what instructions were used

Resources

references/

• implementation-prompt-template.md: Full prompt template with all placeholders and patterns

scripts/

• save_prompt.py: Saves generated prompts to docs/prompts/ with metadata header