Skip to content

Latest commit

 

History

History
68 lines (55 loc) · 3.21 KB

File metadata and controls

68 lines (55 loc) · 3.21 KB

Agentic Development Workflow

This repository supports Router, Architect, Implementer, and Reviewer roles for AI-assisted development. The goal is to let a high-reasoning model own design, planning, and review while a lower-cost worker model performs narrowly scoped implementation.

All agents must follow the root AGENTS.md. It is the bootstrap and hard-policy index for this repository.

Start Here

  • Role bootstrap: role assignment, required agent-role preflight, required context, and Workshop launch caveats.
  • Workshop confinement: the mandatory Workshop execution boundary and agent UI modes.
  • Commands: canonical Workshop actions and lower-level command references.
  • Verification: standard loops and area-specific verification matrix.
  • Git boundary: Git inspection, commits/tags, and the no-Workshop-push rule.
  • Scope and hygiene: scope control, generated artifacts, and user changes.

Roles

Role Responsibility
Router Receive unstructured requests, answer read-only repository questions directly, route verification-only requests, coordinate manual handoffs, and hand implementation-oriented work to the Architect.
Architect Inspect the repo, make design decisions, and write one implementation packet.
Implementer Apply the packet exactly, run listed Workshop checks, and stop.
Reviewer Review the diff, verify scope, run checks, and create any follow-up packet.
Inline assistant Help with small local edits while a human is steering.

Do not ask the Implementer to plan. Do not ask the Reviewer to accept work without reading the diff.

Workflows

  • Delegation workflow: multi-agent enforcement, operating loop, panel role assignment, packet size, and failure handling.
  • Editor preflight: VS Code and Zed Workshop readiness checks.
  • Agent skills: repo-local slash-command workflows for /tdd, /grill-with-docs, and /diagnose.
  • Wiki workflow: documentation modes and wiki-specific checks.
  • Model selection: model and surface assignment.

Use templates/role-launch-prompts.md for minimal role prompts and templates/implementation-packet.md for implementation packets.

Reusable Policy Modules

The following modules are intended to be reusable by other Workshop-backed projects with project names and action lists adjusted: