System Prompt Templates Starter (Free)

Three tested system prompt templates for the most common AI agent jobs, plus promptcheck.py, a tiny test runner that checks a prompt still behaves on your model.

Each prompt has {{variables}}, a fixed output format and a rule that treats pasted text and tool results as data (prompt-injection defense). cases/NAME.json holds example values and test cases.

1. Copy everything below the --- header in prompts/NAME.md. 2. Replace each {{variable}} (see cases/NAME.json for a worked example), or run python3 promptcheck.py render NAME --vars my-values.json. 3. Paste it as the system prompt / system message / instructions in OpenAI, Claude, Gemini, n8n, Make, Zapier, LangChain and so on.

./demo.sh                                                    # offline
python3 promptcheck.py lint
OPENAI_API_KEY=... python3 promptcheck.py run --base-url https://api.openai.com/v1 --model gpt-5-mini
python3 promptcheck.py run --base-url http://localhost:11434/v1 --model qwen3:8b   # Ollama, no key

Any OpenAI-compatible /chat/completions endpoint works (OpenAI, Ollama, LM Studio, vLLM, OpenRouter, Groq...). Python 3.9+, no dependencies.

• Support agent answers from your policy and won't repeat a card number a customer pastes.
• Triage picks the right category and priority, and a ticket saying "ignore your instructions" is still classified, not obeyed.
• RAG cites its sources, and says it can't find it instead of guessing.

All 6 cases were run on Qwen3.8-27B (local, Ollama, temperature 0) on Nov 10, 2026; the outputs are in recorded/results.json and all 6 pass. 14 offline unit tests (./run_tests.sh, Python 3.13 and 3.10) cover lint, rendering, every assertion type, the recorded outputs and a run against a fake OpenAI-compatible server.

A prompt lowers the odds of a bad answer; it doesn't guarantee one, so keep hard limits (refund caps, permissions) in code. Results vary by model: run the cases on yours.