Skip to main content

SAFE AUTONOMOUS CODING

Plan with agents without hiding the decisions.

Start with one small feature idea in Claude. Keep intent explicit, give the agent focused instructions, validate the system model, generate documentation from it, and finish with a scoped, critic-reviewed plan and visible blockers.

Plan a governed change with Claude Learn safe agentic coding

Designed for developers with a Git repository, Claude Code, and the pinned AEP and ESS binaries. The published eight-step golden path ends with one story still active; its optional aep drive follow-on is experimental and does not complete implementation.

ONE PRACTICAL LOOP

From a feature idea to a verifiable plan.​

The first path is organized around what you are trying to accomplish—not around which repository happens to implement each boundary.

  1. 01

    Understand

    Make the outcome, limits, and authority visible before the model starts.

  2. 02

    Instruct

    Give Claude focused Agent Plugins instead of one giant prompt.

  3. 03

    Specify

    Use ESS when system behavior should validate and generate deterministic docs or contracts.

  4. 04

    Plan

    Decompose and scope the work, then retain critic verdicts and unresolved decisions.

  5. 05

    Verify

    Keep the validated model, generated docs, plan artifacts, blockers, and refusals inspectable.

Follow the complete learn → do → verify path

OTHER OUTCOMES

Go deeper when the work calls for it.​

Product deployment, agent infrastructure, and repository reference remain available without competing with a first practitioner experience.

REFERENCE

Already know what you need?​

Browse source-locked technical documentation or inspect the complete project map.

Open technical documentationExplore every project