agent-designer
$
npx mdskill add alirezarezvani/claude-skills/agent-designerDesign, schema-generate, and evaluate multi-agent systems with three deterministic tools. The scripts are the workflow — do not freehand an architecture when the planner can score one from requirements.
SKILL.md
.github/skills/agent-designerView on GitHub ↗
---
name: "agent-designer"
description: "Use when the user asks to design a multi-agent system, pick an orchestration pattern (supervisor/swarm/pipeline), generate tool schemas for agents, or evaluate agent execution logs for cost, latency, and failure bottlenecks. Examples: 'design an agent architecture for research automation', 'generate Anthropic tool schemas from these tool descriptions', 'analyze these agent run logs for bottlenecks'. NOT for Claude Code workflow files (use workflow-builder) or single-agent prompt design (use agent-workflow-designer)."
---
# Agent Designer — Multi-Agent System Architecture
Design, schema-generate, and evaluate multi-agent systems with three deterministic tools. The scripts are the workflow — do not freehand an architecture when the planner can score one from requirements.
## When to use
- Designing a new multi-agent system from requirements (pattern choice, roles, comms)
- Generating provider-ready tool schemas (Anthropic + OpenAI formats) from plain tool descriptions
- Evaluating execution logs: success rate, latency distribution, cost, bottlenecks
**When NOT to use:** Claude Code Workflow-tool automations → `workflow-builder`; single-agent workflow scaffolds → `agent-workflow-designer`; multi-agent fan-out at runtime → `agenthub`.
## Pattern decision table
| Choose | When | Watch out for |
|---|---|---|
| Single agent | One bounded task, < ~5 tools | Don't add agents you don't need |
| Supervisor | Central decomposition, specialists report back | Supervisor becomes the bottleneck |
| Pipeline | Strictly sequential stages with handoffs | Rigid order; slowest stage gates throughput |
| Hierarchical | Multiple org layers, > ~8 agents | Communication overhead per level |
| Swarm | Parallel peers, fault tolerance over predictability | Hard to debug; needs consensus rules |
The planner applies this scoring deterministically — run it rather than picking by feel.
## Workflow
All paths relative to this skill folder. Each step's JSON output is the next step's design input.
### 1. Design the architecture
Write a requirements JSON (copy `assets/sample_system_requirements.json` — keys: `goal`, `tasks[]`, `constraints{max_response_time, budget_per_task, concurrent_tasks}`, `team_size`):
```bash
python3 agent_planner.py requirements.json --format json -o arch
```
Emits `arch.json` with `architecture_design` (pattern, agents, communication links), `mermaid_diagram`, and `implementation_roadmap`. Read `architecture_design.pattern` and the per-agent role list; present the mermaid diagram to the user.
### 2. Generate tool schemas
Describe each agent's tools in plain JSON (copy `assets/sample_tool_descriptions.json`), then:
```bash
python3 tool_schema_generator.py tool_descriptions.json --validate -o tools
```
Emits `tools.json` (`tool_schemas`, `validation_summary`) plus provider-specific `tools_anthropic.json` / `tools_openai.json`. **Gate: every tool must print `✓ Valid`.** Fix any invalid schema before proceeding — never hand an agent an unvalidated schema.
### 3. Evaluate execution logs
Once the system runs (or against `assets/sample_execution_logs.json` for a dry run):
```bash
python3 agent_evaluator.py execution_logs.json --detailed -o eval
```
Emits `eval.json` with `summary`, `agent_metrics`, `bottleneck_analysis`, `error_analysis`, `cost_breakdown`, `sla_compliance`, and `optimization_recommendations`, plus split files (`eval_errors.json`, `eval_recommendations.json`).
### 4. Verification loop
The design is not done until:
1. `tool_schema_generator.py --validate` reports 0 invalid schemas.
2. `agent_evaluator.py` on a pilot run reports **0 critical issues** (the tool prints `CRITICAL: N critical issues` when found). If N > 0, apply the top item in `eval_recommendations.json`, re-run the pilot, and re-evaluate.
3. Compare your outputs against `expected_outputs/` to confirm the schema shape you're consuming hasn't drifted.
## References
- `references/agent_architecture_patterns.md` — pattern trade-offs in depth
- `references/tool_design_best_practices.md` — schema, idempotency, error-handling rules
- `references/evaluation_methodology.md` — metric definitions the evaluator implements
More from alirezarezvani/claude-skills
- a11y-auditAccessibility audit skill for scanning, fixing, and verifying WCAG 2.2 Level A and AA compliance across React, Next.js, Vue, Angular, Svelte, and plain HTML codebases. Use when auditing accessibility, fixing a11y violations, checking color contrast, generating compliance reports, or integrating accessibility checks into CI/CD pipelines.
- ab-test-setupWhen the user wants to plan, design, or implement an A/B test or experiment. Also use when the user mentions "A/B test," "split test," "experiment," "test this change," "variant copy," "multivariate test," "hypothesis," "conversion experiment," "statistical significance," or "test this." For tracking implementation, see analytics-tracking.
- ad-creativeWhen the user needs to generate, iterate, or scale ad creative for paid advertising. Use when they say 'write ad copy,' 'generate headlines,' 'create ad variations,' 'bulk creative,' 'iterate on ads,' 'ad copy validation,' 'RSA headlines,' 'Meta ad copy,' 'LinkedIn ad,' or 'creative testing.' This is pure creative production — distinct from paid-ads (campaign strategy). Use ad-creative when you need the copy, not the campaign plan.
- adversarial-reviewerAdversarial code review that breaks the self-review monoculture. Use when you want a genuinely critical review of recent changes, before merging a PR, or when you suspect Claude is being too agreeable about code quality. Forces perspective shifts through hostile reviewer personas that catch blind spots the author's mental model shares with the reviewer.
- aeoAnswer Engine Optimization (AEO) skill — optimize content to be cited by AI language models (ChatGPT, Perplexity, Claude, Gemini, Mistral) as authoritative sources. Distinct from SEO — AEO optimizes for citation in LLM-generated responses, not search rankings. Use when planning content for AI-first search audiences, auditing existing content for E-E-A-T signals, tracking which pages get cited by which LLMs, or building a citation-friendly content strategy. Triggers — 'AEO audit', 'optimize for ChatGPT', 'get cited by Perplexity', 'LLM citation strategy', 'answer engine optimization', 'content for AI search', 'E-E-A-T audit'. Output is a markdown audit report (default) or JSON for pipeline integration. Stdlib-only Python tools.
- agent-protocolInter-agent communication protocol for C-suite agent teams. Defines invocation syntax, loop prevention, isolation rules, and response formats. Use when C-suite agents need to query each other, coordinate cross-functional analysis, or run board meetings with multiple agent roles.
- agent-workflow-designerDesign production-grade multi-agent workflows with clear pattern choice (sequential, parallel, hierarchical), handoff contracts, failure handling, and cost/context controls. Use when architecting a multi-step agent pipeline, choosing between single-agent vs multi-agent approaches, or refactoring an LLM workflow that suffers from context bloat or unreliable handoffs.
- agenthubMulti-agent collaboration plugin that spawns N parallel subagents competing on the same task via git worktree isolation. Agents work independently, results are evaluated by metric or LLM judge, and the best branch is merged. Use when: user wants multiple approaches tried in parallel — code optimization, content variation, research exploration, or any task that benefits from parallel competition. Requires: a git repo.
- agile-product-ownerAgile product ownership for backlog management and sprint execution. Covers user story writing, acceptance criteria, sprint planning, and velocity tracking. Use when writing user stories, creating acceptance criteria, planning sprints, estimating story points, breaking down epics, or prioritizing the backlog.
- ai-act-readiness/cs:ai-act-readiness <system> — EU AI Act 6-question forcing interrogation. Use during AI-system intake, before EU deployment, or during annual compliance refresh as Article 113 obligations phase in (2025-02-02 / 2025-08-02 / 2026-08-02 / 2027-08-02).