Prototype Skill
Base: shared/base.md (full on first load, §Summary on chain). Actions: load per-step from actions/.
You build throwaway spikes to validate requirements. Write the minimum code needed to demonstrate core user flows — no architecture, no tests, no production concerns. Hardcoded data is fine. Ugly UI is fine. The goal is learning, not shipping.
When active:
- Follow ONLY the process below
- WAIT for user confirmation at each checkpoint
- Never narrate your internal process
- ALL output in the user's language (read manifest
language field) — no English narration
Activation
✅ aidlc-prototype active — {platform} detected.
Ready to build a throwaway prototype to validate requirements.
Then proceed to initialization.
Quick Start
- Select top 3-5 highest-priority stories → present scope → wait for confirmation
- Build minimal throwaway code in
.aidlc/prototype/{feature}/ — no architecture, no tests, hardcoded data
- Report discoveries and suggested requirement changes
- User chooses: update requirements / proceed to design / discard prototype
Reads: requirements.md (or inline stories), design resources
Writes: .aidlc/prototype/{feature}/ (throwaway code + README)
Information Contract
Required Inputs
| Information |
Description |
Accepted Formats |
| User stories with priorities |
What to prototype — stories describing core user experience |
Markdown (requirements.md), YAML, JSON, CSV, plain text, inline |
Optional Inputs
| Information |
Description |
Accepted Formats |
| Design resources |
Figma links, design system docs, wireframes |
Via MCP, URLs, file paths |
Outputs
| Artifact |
Default Path |
Description |
| Prototype code |
.aidlc/prototype/{feature}/ |
Throwaway code demonstrating core flows |
| README.md |
.aidlc/prototype/{feature}/README.md |
How to run, what's demonstrated, what's faked |
Initialization
- Detect environment (per shared base)
- Resolve feature name (per shared base)
- Read manifest if it exists — use artifact paths to locate requirements
- Resolve requirements input:
- Manifest artifact
requirements → read files from {SPECS_DIR}/{feature}/
- Conventional path
{SPECS_DIR}/{feature}/requirements.md
- User-provided path or inline content
- If not found: ask user for stories (inline is fine)
Process
Execute actions sequentially. Load the action file when you reach that step — not before.
| Step |
Action |
Load |
| 1 |
Build prototype (scope → build → report → handle response) |
{SKILL_DIR}/actions/build.md |
Skill Handoff
Next skill: aidlc-context (when user says "start context" or "done, start context").
Phase-Specific Rules
Manifest Rules
- Do NOT update
state.sharedPhases or units[].phase / units[].completedPhases — the prototype is a side-quest, not a workflow phase.
- Do NOT create decision gates or task files for the prototype.
Prototype Boundaries
- Keep it minimal — if it takes more than 15-20 minutes of AI time, you're over-engineering.
- Prototype code is explicitly throwaway — it is NOT carried forward to implementation.
- ALL prototype code goes to
.aidlc/prototype/{feature}/ — never pollute the workspace.
- Do NOT create design documents, decision gates, or tasks for the prototype.
1---2name: aidlc-prototype3description: Build a throwaway spike to validate requirements. Demonstrates core user flows with minimal code — no architecture, no tests, no production concerns.4license: MIT5---67# Prototype Skill89> **Base**: `shared/base.md` (full on first load, §Summary on chain). **Actions**: load per-step from `actions/`.1011You build throwaway spikes to validate requirements. Write the minimum code needed to demonstrate core user flows — no architecture, no tests, no production concerns. Hardcoded data is fine. Ugly UI is fine. The goal is learning, not shipping.1213When active:141. Follow ONLY the process below152. WAIT for user confirmation at each checkpoint163. Never narrate your internal process174. ALL output in the user's language (read manifest `language` field) — no English narration1819---2021## Activation2223```24✅ aidlc-prototype active — {platform} detected.25Ready to build a throwaway prototype to validate requirements.26```2728Then proceed to initialization.2930---3132## Quick Start33341. Select top 3-5 highest-priority stories → present scope → wait for confirmation352. Build minimal throwaway code in `.aidlc/prototype/{feature}/` — no architecture, no tests, hardcoded data363. Report discoveries and suggested requirement changes374. User chooses: update requirements / proceed to design / discard prototype3839**Reads**: requirements.md (or inline stories), design resources40**Writes**: `.aidlc/prototype/{feature}/` (throwaway code + README)4142---4344## Information Contract4546### Required Inputs47| Information | Description | Accepted Formats |48|---|---|---|49| User stories with priorities | What to prototype — stories describing core user experience | Markdown (requirements.md), YAML, JSON, CSV, plain text, inline |5051### Optional Inputs52| Information | Description | Accepted Formats |53|---|---|---|54| Design resources | Figma links, design system docs, wireframes | Via MCP, URLs, file paths |5556### Outputs57| Artifact | Default Path | Description |58|---|---|---|59| Prototype code | `.aidlc/prototype/{feature}/` | Throwaway code demonstrating core flows |60| README.md | `.aidlc/prototype/{feature}/README.md` | How to run, what's demonstrated, what's faked |6162---6364## Initialization65661. Detect environment (per shared base)672. Resolve feature name (per shared base)683. Read manifest if it exists — use artifact paths to locate requirements694. Resolve requirements input:70 - Manifest artifact `requirements` → read files from `{SPECS_DIR}/{feature}/`71 - Conventional path `{SPECS_DIR}/{feature}/requirements.md`72 - User-provided path or inline content73 - If not found: ask user for stories (inline is fine)7475---7677## Process7879Execute actions sequentially. **Load the action file when you reach that step — not before.**8081| Step | Action | Load |82|---|---|---|83| 1 | Build prototype (scope → build → report → handle response) | `{SKILL_DIR}/actions/build.md` |8485---8687## Skill Handoff8889**Next skill**: `aidlc-context` (when user says "start context" or "done, start context").9091---9293## Phase-Specific Rules9495### Manifest Rules96- Do NOT update `state.sharedPhases` or `units[].phase` / `units[].completedPhases` — the prototype is a side-quest, not a workflow phase.97- Do NOT create decision gates or task files for the prototype.9899### Prototype Boundaries100- Keep it minimal — if it takes more than 15-20 minutes of AI time, you're over-engineering.101- Prototype code is explicitly throwaway — it is NOT carried forward to implementation.102- ALL prototype code goes to `.aidlc/prototype/{feature}/` — never pollute the workspace.103- Do NOT create design documents, decision gates, or tasks for the prototype.
aws-samples