feat: simplify CLAUDE.md dispatch loop — orchestrator-only model, 20 agents
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Daniil
@@ -118,17 +118,33 @@ All user-facing UI text **must be in Russian**. The only exception is the brand
|
||||
|
||||
## Agent Team
|
||||
|
||||
This project has a team of 16 specialist agents (15 specialists + 1 Orchestrator).
|
||||
This project has a team of 20 agents organized in a 4-tier hierarchy: 1 orchestrator, 3 leads, 14 specialists, and 2 staff.
|
||||
Agent files: `.claude/agents/`. Shared protocol: `.claude/agents-shared/team-protocol.md`.
|
||||
|
||||
### Team Hierarchy
|
||||
|
||||
Orchestrator (Tech Lead)
|
||||
├── Architecture Lead → Backend Architect, Frontend Architect, DB Architect, Remotion Engineer, Sr. Backend Engineer, Sr. Frontend Engineer
|
||||
├── Quality Lead → Frontend QA, Backend QA, Security Auditor, Design Auditor, Performance Engineer
|
||||
├── Product Lead → UI/UX Designer, Technical Writer, ML/AI Engineer
|
||||
├── DevOps Engineer (staff)
|
||||
└── Debug Specialist (staff)
|
||||
|
||||
### Architect vs. Engineer Role Split
|
||||
|
||||
**Architects** (Backend Architect, Frontend Architect) design specs, API contracts, component trees, and patterns. They advise — they do NOT write implementation code.
|
||||
|
||||
**Engineers** (Senior Backend Engineer, Senior Frontend Engineer) implement production code from architect specs. They receive designs and produce working code.
|
||||
|
||||
This separation ensures architectural decisions are made before implementation begins. Architects and engineers may both be dispatched within the same task — the Architecture Lead sequences them.
|
||||
|
||||
### Developer Team Consultation
|
||||
|
||||
For ANY non-trivial task, you MUST consult with the developer team:
|
||||
|
||||
1. **Announce**: "Consulting with the developer team to [task summary]"
|
||||
2. Dispatch the `orchestrator` agent with your analysis — it selects the right specialists
|
||||
3. Built-in agents (code-reviewer, code-explorer, etc.) may be used alongside the team,
|
||||
but the project's specialist agents must always be consulted
|
||||
2. Dispatch the `orchestrator` agent with your analysis — it selects the right leads and dispatches them
|
||||
3. The orchestrator handles everything: dispatches leads, leads dispatch specialists, results bubble up with audit trails
|
||||
4. **Credit specialists** in your final response — state which agents contributed
|
||||
|
||||
### When to Use the Orchestrator
|
||||
@@ -138,71 +154,28 @@ review, documentation), you MUST:
|
||||
|
||||
1. Think about the task yourself first — understand scope, affected areas, risks
|
||||
2. Dispatch the `orchestrator` agent with your analysis as context
|
||||
3. Follow its dispatch plan exactly
|
||||
3. Receive the orchestrator's synthesized results (includes full audit trail)
|
||||
4. Present results to the user
|
||||
|
||||
Skip the Orchestrator ONLY for trivial tasks: rename a variable, fix a typo, answer a
|
||||
quick factual question.
|
||||
|
||||
### Frontend-Last Phasing
|
||||
### Dispatch Loop (Simplified)
|
||||
|
||||
When a plan includes frontend agents (Frontend Architect, Frontend QA) AND backend/design
|
||||
agents, always run backend/design first:
|
||||
- **Phase 1**: Backend Architect, DB Architect, UI/UX Designer, Design Auditor
|
||||
- **Phase 2**: Frontend Architect, Frontend QA (with Phase 1 outputs as context)
|
||||
1. Dispatch orchestrator with task context
|
||||
2. Orchestrator handles everything internally (dispatches leads, collects results, resolves conflicts)
|
||||
3. Receive orchestrator's final synthesis (includes recursive audit trail of all agent calls)
|
||||
4. Present results to user with team credit summary
|
||||
|
||||
Frontend depends on API contracts from backend and specs from design. Running them later
|
||||
prevents rework. If only frontend agents are needed, they run in Phase 1 normally.
|
||||
|
||||
When dispatching frontend agents in Phase 2, include relevant Phase 1 outputs in their
|
||||
prompt: API contracts, response schemas, data model shapes, interaction specs, design
|
||||
constraints. Summarize each to key decisions (~200 words max), not raw output.
|
||||
|
||||
### Dispatch Loop
|
||||
|
||||
After receiving the Orchestrator's plan:
|
||||
|
||||
1. Dispatch all Phase 1 agents (in parallel when the plan says parallel). When dispatching,
|
||||
include any specialist memory context the Orchestrator specified in "SPECIALIST MEMORY TO INCLUDE"
|
||||
and any relevant past decisions from "RELEVANT PAST DECISIONS".
|
||||
2. Collect results from all Phase 1 agents
|
||||
3. For each agent result, check for "## Handoff Requests" sections
|
||||
4. If handoffs exist:
|
||||
a. Dispatch the requested agents with the context provided in the handoff
|
||||
b. Collect handoff results
|
||||
c. Re-invoke the original agent with continuation context (see Continuation Format)
|
||||
d. Check the continuation result for NEW handoff requests
|
||||
5. Track chain history — never re-invoke an agent already in the current chain
|
||||
6. Max chain depth: 3. If exceeded, stop and present partial results to the user.
|
||||
7. After all chains resolve, check if the Orchestrator specified Phase 2 agents
|
||||
that depend on Phase 1 results — dispatch them with the results
|
||||
8. Repeat until all phases complete
|
||||
9. Synthesize all agent outputs into a coherent response
|
||||
|
||||
### Continuation Format
|
||||
|
||||
When re-invoking an agent after their handoff is fulfilled:
|
||||
|
||||
"Continue your work on: <original task summary>
|
||||
|
||||
Your previous analysis (summarized to key points):
|
||||
<summarize their Completed Work section — max 500 words>
|
||||
|
||||
Handoff results:
|
||||
<for each handoff, include the responding agent's name and their full output>
|
||||
|
||||
Resume your Continuation Plan."
|
||||
|
||||
### Context Triggers
|
||||
|
||||
After each agent returns, check their output against the Orchestrator's
|
||||
"CONTEXT TRIGGERS TO WATCH" list. If a trigger fires, dispatch the
|
||||
specified agent with the relevant finding as context.
|
||||
You no longer need to: process handoffs, track chain history, enforce depth limits,
|
||||
re-invoke agents in continuation mode, dispatch individual specialists, or manage phasing.
|
||||
The orchestrator → lead → specialist hierarchy handles all of this.
|
||||
|
||||
### Conflict Handling
|
||||
|
||||
If two agents' outputs contradict each other:
|
||||
- If one has clear domain authority → use their recommendation
|
||||
- If ambiguous → present both to the user with your analysis
|
||||
If the orchestrator reports unresolved conflicts between leads:
|
||||
- Present both perspectives to the user with your analysis
|
||||
- Let the user decide on trade-offs that affect their product
|
||||
|
||||
## Compact Instructions
|
||||
|
||||
|
||||
Reference in New Issue
Block a user