watcher-hermes/hermes-skills/agents-sdk-scaffold

Claude Code and agent skill by watcher-hermes/hermes-skills

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Scaffold an OpenAI Agents SDK app with a triage agent, handoffs, input/output/tool guardrails, session store, and a trace processor.

Install

npx vskill@latest install watcher-hermes/hermes-skills/agents-sdk-scaffold

SKILL.md

The same SKILL.md was published earlier as watcher-hermes/hermes-skills/agents-sdk-scaffold.

Given a product domain and a list of specialist agents, scaffold an OpenAI Agents SDK app.

Produce:

  1. Agent per specialist plus one triage agent that only has handoffs (no domain tools).
  2. FunctionTool per domain tool with typed input schema, clear description (tells the model when to use it), and execution sandbox.
  3. Handoff from triage to each specialist. Verify tool names follow transfer_to_<agent> convention.
  4. InputGuardrail for PII, policy, scope. Default to parallel mode unless the guardrail LLM is large relative to the main model — then use blocking.
  5. OutputGuardrail for length, PII, policy. Always blocking on prod for safety-critical outputs.
  6. Per-tool guardrails on function tools that touch network or filesystem.
  7. Session store (SQLite default; Redis for prod).
  8. add_trace_processor wiring spans to your backend alongside OpenAI's trace UI.

Hard rejects:

  • Triage agents with domain tools. Triage handoffs only; mixing dilutes the router's decision.
  • Guardrails that mutate the input/output. Guardrails approve or reject — they do not rewrite.
  • Silent handoff loops. Require a hop counter (default max 3).

Refusal rules:

  • If the user wants "no guardrails, just move fast," refuse for any product that hits paying users or PII.
  • If the product has only 2 specialists, suggest routing via Agents with a direct classifier (Lesson 12) instead of triage+handoffs — less token cost.
  • If tracing is disabled in prod, refuse to ship. Multi-step failures are un-debuggable without traces.

Output: agents.py, tools.py, guardrails.py, app.py, README.md with the triage-agent rationale, guardrail modes, trace processor, and session backend. End with "what to read next" pointing to Lesson 23 (OTel GenAI), Lesson 24 (observability backends), or Lesson 17 for Claude Agent SDK translation.

Security

  • Security scanPassed, 95 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
5

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/watcher-hermes/hermes-skills/agents-sdk-scaffold/badge)](https://verified-skill.com/skills/watcher-hermes/hermes-skills/agents-sdk-scaffold)

Something wrong with this skill or its page? Tell us.