DEMO_GALLERY.md

Demo Gallery

This page lists runnable demos that can be used for videos, social posts, and reproducible onboarding.

1) Interactive Code Agent (Permissioned Tools)

Path: examples/code-agent/

What it demonstrates:

• ReAct loop with streaming responses
• Built-in tools (Read, Write, Edit, Bash, Glob, Grep)
• Permission workflow for sensitive tools
• Session save/resume

Run:

cd examples/code-agent
bun install
GEMINI_API_KEY=your_key bun dev

2) Quickstart Package (Basic + Session + Tools)

Path: examples/quickstart/

What it demonstrates:

• One-shot prompt call
• Multi-turn session workflow
• Tool usage (Glob, Bash, WebSearch)

Run:

cd examples/quickstart
bun install
GEMINI_API_KEY=your_key bun test

3) Skill System Demo

Path: examples/skills-demo.ts and examples/skills/

What it demonstrates:

• Skill loading from local directories
• Skill instruction injection into system prompt
• Task-specialized behavior with reusable skill files

Run:

OPENAI_API_KEY=your_key bun run examples/skills-demo.ts

4) Structured Output Demo

Path: examples/structured-output-demo.ts

What it demonstrates:

• Typed/structured result handling
• Stable output shape for downstream pipelines

Run:

OPENAI_API_KEY=your_key bun run examples/structured-output-demo.ts

5) SWE-bench Lite Smoke Eval

Path: benchmark/swebench/

What it demonstrates:

• Reproducible local eval harness workflow
• Prediction/report artifact generation
• Batch run and report summarization

Run:

cd benchmark/swebench
./scripts/run_oas_smoke_one.sh
python ./scripts/summarize_reports.py --reports-dir ./outputs/reports --limit 20

Artifact Convention

For demos intended for external sharing, store outputs under:

• ./outputs/ for local demo outputs
• benchmark/.../outputs/ for benchmark artifacts

Suggested social posting bundle per demo:

• 20-30 second terminal recording
• exact run command
• output snippet or report JSON