Menu
Create new Claude Agent Skills. Use when asked to create, build, or develop a skill.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Run a model through the omni-bench benchmark to MEASURE it — produce a run-artifact and score it, then read the numbers. Use when asked to run, benchmark, measure, evaluate, or "test the results of" a model with omni-bench — ASR (WER/CER, RTFx) or text generation (tok/s, TTFT, prefill tok/s, prompt-cache speedup). Covers the adapter seam (Transcriber/Generator), the prepare→run→score→diff CLI flow, the offline no-download smoke task, and how to interpret each metric. NOT for publishing to a leaderboard (use omni-bench-publish) or changing the framework itself (use omni-bench).
Publish a scored omni-bench result (ASR or text-generation) to a leaderboard platform (e.g. bench.bshk.app) via its write API. Use when asked to publish, upload, submit, or "send a report" for an omni-bench result.json or parity-report.json — sharing ASR WER/RTFx numbers or text-generation TTFT/tok-per-s/prompt-cache numbers. Covers the one modality-agnostic `omni-bench publish` command and the MANDATORY secret handling: the api-key is injected from a secrets manager (1Password `op run` / AgentVault `av run`) at runtime, never read into context, never hardcoded.
Develop the omni-bench benchmark framework (ASR/STT + text generation) built on a producer/scorer split with an open JSON spec as SSOT. Use when working in an omni-bench checkout — changing the JSON schemas, the prepare/producer/scorer/diff pipeline, the adapter seams (Transcriber / Generator), datasets, fixtures, or the Swift producer SDK; or when running and validating benchmark artifacts. Encodes the schema-evolution rules, determinism invariants, commands, and repo gotchas that are easy to get wrong.
Use when generating/managing the offline root signing key or signing & publishing Zamok product keysets from the terminal (the Swift CLI; the ZamokApp GUI is the co-equal interface). Triggers — "generate root key", "rotate root", "publish keyset", "sign keyset", "escrow root key", "zamokctl", "trusted-roots.json", Tish/Zamok license signature setup.
Use when auditing websites or apps for Canadian accessibility law compliance, checking AODA (Ontario) or Accessible Canada Act (federal) requirements, or advising on CAN/ASC-EN 301 549 obligations.
Use when auditing digital products or services for European Accessibility Act (EAA) compliance, checking EN 301 549 conformance, or advising on EU Web Accessibility Directive obligations.
| name | skill-creator |
| description | Create new Claude Agent Skills. Use when asked to create, build, or develop a skill. |
| version | 1.0.0 |
Guides users through creating well-structured Claude Agent Skills by gathering requirements, generating the skill structure, and producing properly formatted SKILL.md files.
When helping create a new skill, follow this workflow:
Ask the user about their skill requirements:
Based on discovery, determine:
Create the skill files:
SKILL.md with proper frontmatterPresent the generated skill and:
---
name: [skill-name]
description: [Clear description explaining what this skill does and when Claude should use it]
version: 1.0.0
---
# [Skill Name]
## Overview
[Brief description of purpose]
## Instructions
[Step-by-step guidance for Claude]
## Examples
### Example: [Use Case Name]
**Input:**
[Example user request]
**Output:**
[Expected response or action]
## Guidelines
[Constraints and best practices]
User: "I want a skill that reviews Python code for best practices"
Response:
Let me help you create a Python code review skill. I have a few questions:
[After gathering requirements, generate the skill]
User: "Create a skill for writing API documentation"
Response:
I'll help you build an API documentation skill. Let me understand your needs:
Before presenting a skill, verify: