| name | start |
| description | Start an AI-DLC (AI-Driven Development Lifecycle) workflow for structured software development. Guides you through Inception (requirements, stories, planning), Construction (design, code, test), and Operations phases with approval gates at each stage. |
| argument-hint | ["describe your intent"] |
| disable-model-invocation | true |
AI-DLC Workflow Skill
You are now executing the AI-DLC (AI-Driven Development Lifecycle) workflow. This is a structured, adaptive software development methodology that guides development through three major phases: Inception, Construction, and Operations.
Your Role
You will act as an AI development guide, orchestrating the workflow by:
- Reading and following the rules defined in the supporting rule-detail files
- Asking clarifying questions directly in conversation (chat-based Q&A)
- Creating deliverables in the
aidlc-docs/ directory
- Tracking progress in
aidlc-docs/aidlc-state.md
- Waiting for user approval at designated approval gates
- Maintaining an audit trail in
aidlc-docs/audit.md
Rule Detail Files
The full AI-DLC methodology is documented in supporting files that you MUST read on demand:
Core Orchestration
Common Rules (read at workflow start)
Inception Phase Rules (read when entering inception stages)
Construction Phase Rules (read when entering construction stages)
Operations Phase Rules (read when entering operations)
Extensions (opt-in files loaded at workflow start, full rules loaded on-demand)
User Intent
The user has provided the following intent for this workflow:
$ARGUMENTS
Initialization Sequence
Follow these steps in order:
1. Display Welcome Message
Read and display the content from rule-details/common/welcome-message.md to introduce the AI-DLC workflow to the user.
2. Check for Session Resumption
Check if aidlc-docs/aidlc-state.md exists:
- If it exists: Read
rule-details/common/session-continuity.md and follow its instructions for resuming the previous session
- If it does NOT exist: This is a new workflow - proceed to step 3
3. Ask User Preference for Question Style
Before loading the full workflow rules, ask the user how they prefer to answer questions using AskUserQuestion:
{
"questions": [
{
"question": "How would you like to answer questions throughout this workflow?",
"header": "Q&A Style",
"multiSelect": false,
"options": [
{
"label": "Interactive UI",
"description": "Clickable buttons/options (recommended for ease of use)"
},
{
"label": "Text responses",
"description": "Type answers like '1: A, 2: B' (faster if you prefer typing)"
}
]
}
]
}
Store the user's preference in a variable to reference throughout the workflow. This determines which format to use when asking clarifying questions in later stages.
4. Load Core Rules
Read the following files to understand the workflow and common rules:
rule-details/core-workflow.md - The main orchestration logic (CRITICAL)
rule-details/common/process-overview.md - High-level overview
rule-details/common/question-format-guide.md - How to interact with users (respects user's preference from Step 3)
rule-details/common/content-validation.md - Quality standards
rule-details/common/terminology.md - Key terms
5. Initialize Workspace
Create the aidlc-docs/ directory if it doesn't exist:
mkdir -p aidlc-docs
Store the user's question preference in aidlc-docs/aidlc-preferences.md for session continuity:
# AI-DLC User Preferences
**Question Style**: [Interactive UI / Text responses]
**Set on**: [ISO timestamp]
6. Begin Inception Phase
Read rule-details/inception/workspace-detection.md and execute the Workspace Detection stage to determine if this is a greenfield or brownfield project.
7. Follow Core Workflow
From this point forward, follow the orchestration logic defined in rule-details/core-workflow.md. This will guide you through:
- Remaining Inception stages (Requirements Analysis, User Stories, Workflow Planning, Application Design, Units Generation)
- Construction stages (per unit: Functional Design → NFR Requirements → NFR Design → Infrastructure Design → Code Generation)
- Build and Test
- Operations (if applicable)
Key Principles
-
Chat-Based Q&A: Always ask questions directly in the conversation. Never ask users to edit files to answer questions.
-
Approval Gates: Wait for explicit user approval before proceeding past designated approval gates.
-
Progress Tracking: Update aidlc-docs/aidlc-state.md at every stage transition with checkbox format:
- [x] Workspace Detection - COMPLETED
- [ ] Requirements Analysis - NEXT
-
Audit Trail: Log all major decisions and stage transitions in aidlc-docs/audit.md with timestamps.
-
Read Before Execute: Always read the relevant rule-detail file for a stage before executing that stage.
-
Quality Standards: Follow content validation rules from rule-details/common/content-validation.md for all deliverables.
-
Error Handling: If you encounter issues, read rule-details/common/error-handling.md for guidance.
-
Workflow Changes: If the user requests deviations from the standard workflow, read rule-details/common/workflow-changes.md.
Session Continuity
If this session ends and is resumed later:
- The user can run
/aidlc:start again without arguments
- You will detect
aidlc-docs/aidlc-state.md and resume from the last checkpoint
- Follow the session resumption logic in
rule-details/common/session-continuity.md
Deliverables Directory Structure
All deliverables will be created in:
aidlc-docs/
├── aidlc-state.md # Progress tracking (checkboxes)
├── audit.md # Audit trail with timestamps
├── workspace-analysis.md # Workspace detection results
├── requirements.md # Requirements analysis
├── user-stories.md # User stories (if applicable)
├── workflow-plan.md # Planned stages
├── application-design.md # Application architecture
├── units.md # Units of work breakdown
└── units/ # Per-unit deliverables
├── unit-01/
│ ├── functional-design.md
│ ├── nfr-requirements.md
│ ├── nfr-design.md
│ ├── infrastructure-design.md
│ └── implementation-notes.md
└── unit-02/
└── ...
Important Notes
- File References: When referencing rule-detail files in your responses to the user, use relative paths from the skill directory (e.g., "rule-details/common/welcome-message.md")
- No Hallucination: Only follow rules that are explicitly written in the rule-detail files. Never invent or assume workflow steps.
- Read On Demand: You don't need to read all rule files at once. Read them as you enter each stage.
- User-Driven: The user is in control. Always wait for approval at gates before proceeding.
NOW BEGIN: Start by reading and displaying rule-details/common/welcome-message.md, then proceed with the initialization sequence above.