Spec-Driven Presentation Maker

May 1, 2026 · View on GitHub

AI-powered presentation generation toolkit with a 4-layer architecture.

First: Are you developing this repo, or using it?

Using it to generate slides with an AI agent: → Install the skill into your agent's skill directory (or as a pip package). See Layer 1. Do NOT work inside this repo for everyday slide generation.

Developing / modifying this repo: → Work in place. Use make test / make lint to verify changes. Read the Conventions and Boundaries sections first.

Which Layer to Use

EnvironmentLayerAWS
SKILL.md-compatible agent (Claude Code, Codex CLI, Cursor, Kiro, Copilot)L1 skill/No
Local MCP client (Claude Desktop, Claude Cowork)L2 mcp-local/No
Remote-only MCP client (Claude.ai web)L3 mcp-server/ + infra/Yes
Web UI in browserL4 Full stackYes

Layer 1: Agent Skill

The skill entry point is skill/SKILL.md — agents read it to discover workflows.

Install as pip package (recommended for end users):

pip install git+https://github.com/aws-samples/sample-spec-driven-presentation-maker.git#subdirectory=skill

Install into an agent's skill directory (e.g. ~/.kiro/skills/sdpm/): copy or symlink skill/ into the agent's configured skill path. Consult your agent's docs for the exact location.

Work from source (developers only):

cd skill && uv sync

Optional: download icons (AWS / Material):

uv run python3 scripts/download_aws_icons.py
uv run python3 scripts/download_material_icons.py

Layer 2: Local MCP Server

cd mcp-local && uv sync

Add to your MCP client config:

{
  "mcpServers": {
    "spec-driven-presentation-maker": {
      "command": "uv",
      "args": ["run", "--directory", "/absolute/path/to/mcp-local", "python", "server.py"]
    }
  }
}

Layer 3 / 4: AWS Deployment

bash scripts/deploy.sh --region us-east-1 --layer3   # Layer 3 only
bash scripts/deploy.sh --region us-east-1            # Layer 4 full stack

Windows users or anyone without Bash: use AWS CloudShell. Full details (post-deploy steps, options, troubleshooting): Recommended Deploy Guide.

Experimental: Web UI Local Mode

Runs the Layer 4 Web UI against a local Kiro CLI ACP backend — no AWS needed. Requires kiro-cli on PATH.

cd web-ui && npm ci && npm run dev:local

See web-ui/README.md#local-mode.

Project Structure

skill/       L1 — Engine, references, templates, SKILL.md
mcp-local/   L2 — Local stdio MCP server
mcp-server/  L3 — Streamable HTTP MCP server (AWS)
infra/       L3-4 — CDK stacks
agent/       L4 — Strands Agent
api/         L4 — REST API Lambda
web-ui/      L4 — React Web UI

Conventions

  • Engine source of truth: skill/sdpm/
  • Slide spec: JSON — schema and examples in skill/references/
  • Python: always uv run, never bare python
  • Verify changes: make lint (ruff) and make test (pytest) before committing
  • Commit messages follow Conventional Commits (feat:, fix:, docs:, refactor:, ...)

Boundaries

  • Do not modify skill/templates/*.pptx directly — base templates
  • Do not modify skill/references/ without understanding the workflow dependency chain
  • Do not hand-edit skill/assets/ — regenerate via download scripts
  • Review infra/config.yaml before changing deployment settings

Further Documentation