v0.1.0 - Initial Public Release

Why This Exists

Agent runs are often flaky and expensive:

• Non-determinism: The same prompt can produce different outputs.
• Slow feedback: Integration tests spend 90% of their time waiting on network calls.
• Cost: Repeated debugging burns tokens and money.

Cassette turns "It failed yesterday" into a replayable, immutable artifact.

---

Enterprise Use Case: Reliable Code Generation

(See examples/node-red-generator.ts)

This tool is designed for platforms like FlowFuse or Node-RED where AI agents generate executable code.
Agent Cassette provides a Regression Testing Harness that ensures:

1. Strict Schema Validation: Agents must output valid JSON structures (e.g., correct wires and coordinates).
2. Semantic Safety: If an agent generates unsafe code (e.g., missing return msg), the system detects it and swaps in a safe fallback.
3. Deterministic Replay: We record a "Golden Run" of a complex flow generation. CI pipelines can replay this instantly (0 cost) to prove that model upgrades (e.g., GPT-4o → GPT-5) don't break the JSON schema.