The Ten-Step Working Method

Definition

The Ten-Step Working Method

The ten-step working method is a structured operating procedure for moving an agent task from problem framing to verified completion. It separates intent, context gathering, plan review, implementation, testing, evidence collection, human review, commit hygiene, release decision and post-change observation so that an agent’s speed does not bypass engineering control.

Why It Matters

The recommended working method turns agent use into an engineering loop:

1. FrameDescribe the outcome, user need, scope and non-goals.
2. GroundPoint to relevant repository areas, standards and examples.
3. SpecifyProvide acceptance criteria, quality attributes and expected evidence.
4. ExploreLet the agent inspect before it edits; require clarification for material ambiguity.
5. PlanFor complex work, agree on decomposition and risk points.
6. ImplementWork in small coherent increments.
7. VerifyRun tests, linters, type checks, security checks, evals or visual inspection.
8. ReviewInspect behaviour, diff, architecture and unintended changes.
9. ReflectCapture recurring mistakes in tests, instructions or reusable skills.
10. Ship and observeRelease through normal controls and monitor real behaviour.

Failure mode

The recurring risk is treating this concept as a label rather than an engineering decision with trade-offs, evidence needs and operational consequences.

Evaluation check

Ask what observable evidence would show that this concept has been applied well in a real system, repository or learning exercise.

Related Concepts

• Task briefs and examples
• Evaluating coding agents

Further study: OpenAI, Codex best practices, and Google SRE guidance on safe change management.