AI-Assisted Development
sharedPractical setup for Cursor, Claude Code, Codex, and similar assistants in Athas.
Athas is already structured for AI-assisted workflows (contract-first APIs, typed shared packages, and explicit runbooks). This guide standardizes how we add assistant rules, skills, and reusable commands without creating tool lock-in.
Goals
- keep assistant outputs consistent with Athas architecture
- reduce repeated prompt boilerplate for common workflows
- preserve one shared source of truth across AI tools
Recommended structure
Use a shared .agents/ folder as the canonical source, then map provider-specific folders to it.
.
├─ .agents/
│ ├─ agents/
│ │ └─ code-reviewer.md
│ ├─ commands/
│ │ ├─ setup-new-feature.md
│ │ └─ release-drill.md
│ └─ skills/
│ ├─ better-auth-best-practices/
│ │ └─ SKILL.md
│ ├─ dokploy-deployment/
│ │ └─ SKILL.md
│ └─ contract-first-orpc/
│ └─ SKILL.md
├─ .cursor/
├─ .claude/
├─ .github/
├─ AGENTS.md
└─ CLAUDE.mdRule files
AGENTS.md
Use root-level AGENTS.md as the canonical rules file.
Keep it short and operational. Focus on Athas-specific constraints that AI cannot infer from code:
- module boundary rules (
corepackages framework-agnostic) - contract-first oRPC expectations
- Dokploy deploy/rollback defaults
- non-destructive git policy
CLAUDE.md
Use:
@AGENTS.mdThis avoids drift between rule files.
Skills (modular for Athas)
Keep skills focused by workflow unit (feature/module/migration/component/release):
athas-feature-slice- end-to-end feature change across contracts/backend/frontend
athas-backend-module- module layering and boundary-safe backend refactors
athas-db-migration- schema + migration changes with forward-safe rollout
athas-ui-component- reusable component implementation and variant discipline
athas-release-fix- CI/release gate triage with minimal unblock fixes
athas-contract-first-orpc- oRPC contract-first changes and compile-time drift prevention
athas-dokploy-release-drill- staged Dokploy rollout + rollback rehearsal
athas-docs-maintenance- docs IA, canonical checks, and maintenance-mode updates
Skill directory pattern:
.agents/skills/<skill-name>/SKILL.mdWith frontmatter:
---
name: contract-first-orpc
description: Use when editing oRPC contracts, procedures, or shared-api client integration.
---Commands
Use slash-style command docs for repetitive operations:
/setup-new-feature-> scaffold and integrate feature slices/release-drill-> run staged Dokploy deployment + rollback verification/fix-ci-failure-> triage failing workflow and apply minimal fix
Command files live in:
.agents/commands/*.mdSubagents
Useful baseline subagents:
code-reviewer(readonly): conventions, boundaries, contract driftsecurity-auditor(readonly): auth, secrets, billing/webhook surfacesdocs-auditor(readonly): canonical pointer policy + docs IA consistency
Best practices
- plan before coding: explore -> plan -> implement -> verify
- always include verification criteria (
typecheck,depcruise, smoke checks) - keep prompts scoped to feature/module boundaries
- start a fresh session when context drifts
- treat AI output as draft code and review diffs carefully
Verification commands
Common Athas checks assistants should run after significant changes:
bun run typecheck
bun run depcruise:backend
bun run depcruise:libs
node tools/check-docs-canonical.mjsFor release-readiness workflow:
bun run verify:libs