tools-only Stateless Agent Methodology (SAM)
A constraint-driven development framework that compensates for LLM limitations through architectural structure rather than behavioral instructions.
This methodology has companion documentation at:
- Canonical specification: Stateless Software Engineering Framework
- Prompt/automation helper: SAM framework generator
The Core Insight
Claude is not a knowledge worker - Claude is a stateless computation engine that happens to have noisy, stale priors baked in.
Treat Claude like a pure function:
- Input: Complete context (task file with all answers)
- Output: Verified result
- No side effects: Fresh context each time
- No memory: Everything externalized to artifact files
Part 1: The Problem
This section is maintained in the canonical framework document:
Part 2: The Architectural Solution
This section is maintained in the canonical framework document:
Part 3: The Pipeline Architecture
This section is maintained in the canonical framework document:
- Stateless Software Engineering Framework: Pipeline architecture + artifact flow
- Stateless Software Engineering Framework: Agent specifications (artifact templates)
Part 4: Input Constraints
This section is maintained in the canonical framework document:
Part 5: Stage Specifications
This section is maintained in the canonical framework document:
Further Reading (canonical)
- Framework: Why this works
- Framework: Theoretical foundations
- Framework: Anti-patterns
- Framework: Success metrics
Summary
SAM treats the model as a stateless computation engine and enforces correctness via externalized artifacts + verification gates. The detailed normative specification lives in the framework document linked above.