Skip to content

Latest commit

 

History

History
49 lines (37 loc) · 2.58 KB

File metadata and controls

49 lines (37 loc) · 2.58 KB

AGENTS.md

Subagent routing guide. For documentation rules, tech stack, and PR protocol see CLAUDE.md.

Agent Team — Subagent Routing

Use this table to decide which subagent to dispatch for a given task:

Signal Subagent When
Need to understand existing code/patterns research-analyst Before any implementation, exploring unfamiliar areas
Build component logic, hooks, pages frontend-engineer React state, data fetching hooks, page structure
Styling, layout, responsive, accessibility ui-designer Tailwind classes, animations, design polish
API routes, services, caching, subgraph api-architect Data layer, server-side fetching, performance
Contract calls, wallet, transactions web3-specialist wagmi/viem, ABIs, onchain data, Web3 flows
Write or update documentation docs-writer Docs in docs/, code comments, ADRs

Parallel Dispatch Rules

  • Independent tasks → dispatch multiple subagents in parallel
  • Research then implement → research-analyst first, then implementation agent
  • Frontend + styling → frontend-engineer for logic, ui-designer for visual (can run in parallel on different files)
  • Avoid file conflicts → never dispatch two agents that edit the same file

Spawning as Agent Team (Experimental)

When the task benefits from inter-agent communication (not just parallel work), use agent teams instead of subagents. Example spawn prompts:

Create an agent team with:
- A research-analyst to investigate the current auction implementation
- A frontend-engineer to build the new auction countdown component
- A web3-specialist to implement the bidding transaction flow
Have the researcher share findings with the other two before they start implementing.
Create an agent team with 3 teammates:
- frontend-engineer: implement the proposal detail page layout and state
- ui-designer: style the proposal cards and voting interface
- api-architect: create the proposal data service and caching layer
Each teammate owns different files. Coordinate through the shared task list.

Use subagents (not teams) when:

  • Tasks are independent and don't need to share findings
  • Only one agent is needed
  • The task is quick/focused (research, single component, docs update)