convez.ai
SkillAutomaçãoIntermediário
cekura-ai

Cekura Skills

cekura-ai/cekura-skills

Skills para otimizar agentes de voz com métricas e avaliações na Cekura

Ver no GitHub
InicialPluginScore · de 0 a 100

Conjunto de 9 skills Python com 14 comandos para construir e melhorar testes de agentes de voz na plataforma Cekura. Integra-se com Claude Code, Cursor, Codex e outros assistentes de IA, oferecendo workflows prontos para onboarding, design de métricas, avaliações e melhorias iterativas de agentes.

Instalação

$npx convez-ai cekura-ai/cekura-skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
8
Forks
2
Acompanhando
1
Saúde
25%
Último commit
há 1 semana
Versão
—
Issues abertas
20
Contribuidores
11

Casos de uso

  • Criar e gerenciar métricas de qualidade para agentes de voz
  • Automatizar testes e avaliações de chamadas com transcrições reais
  • Melhorar iterativamente agentes com ciclos de feedback estruturados
  • Integrar Cekura em fluxos de IA coding (VS Code, terminal, Cursor)

Veredito

Skill bem estruturada e prática para times que usam agentes de voz no Cekura. Suporta múltiplas IDEs e oferece boas práticas reais de clientes, mas requer conta ativa na plataforma para máximo aproveitamento.

README

A tradução para o português deste README ainda está na fila.

Cekura AI Skills

AI-powered skills for building and improving voice agent tests and metrics on the Cekura platform. Works with Claude Code, Codex, Cursor, and other AI coding assistants.

Table of Contents


What's Included

9 Skills, 14 Commands in one plugin

SkillsCommands
cekura-coordinatorcekura-onboarding, setup-mcp, upgrade-skills, report-bug
cekura-onboardingcreate-metric, list-metrics, evaluate-calls, improve-metric
cekura-create-agentmanual-create-update-eval, autogen-eval, list-evals, run-evals, eval-results, cekura-report
cekura-self-improving-agent
cekura-metric-design
cekura-metric-improvement
cekura-predefined-metrics
cekura-eval-design
cekura-fixing-prod-issues

These encode best practices from real client deployments — proactive guardrails, real transcript grounding, iterative improvement loops, coverage planning, and anti-pattern detection.

Stuck after an upgrade? If /upgrade-skills reports stale plugin entries or /plugin install fails with Source path does not exist, do a clean reinstall — see Reinstalling Cekura skills below.

Prerequisites

  • Cekura account — Sign up here. Sign in via OAuth or use an API key.
  • Cekura API key (only for the Claude Code plugin path / programmatic MCP access) — Found under Settings > API Keys in the Cekura dashboard. Not needed for npx skills add.

Quick Install (npx skills add)

The fastest way to get Cekura skills into any Agent Skills–compatible client (Claude Code, Cursor, Codex, Windsurf, OpenCode, and many more).

Install

npx skills add cekura-ai/cekura-skills

The CLI prompts you to pick which skills to install and which agents to install them into. To install everything non-interactively:

npx skills add cekura-ai/cekura-skills --all

Update

# Refresh existing skills
npx skills update

# Or stay fully current — refresh existing AND pick up any newly-added skills
npx skills add cekura-ai/cekura-skills --all

Remove

npx skills remove cekura-coordinator   # one skill
npx skills remove --all                 # everything

What gets installed

Nine skills, scoped to specific Cekura workflows:

SkillWhen it activates
cekura-coordinator"What can Cekura do?" — routes you to the right skill
cekura-onboarding"Get started with Cekura" — full platform walkthrough
cekura-create-agent"Connect my voice agent to Cekura"
cekura-self-improving-agent"Improve my agent / auto-tune from eval results"
cekura-metric-design"Create a metric / measure call quality"
cekura-metric-improvement"Improve a metric / fix metric accuracy"
cekura-predefined-metrics"What predefined metrics are available / which built-in metrics should I use"
cekura-eval-design"Design test scenarios for my voice agent"
cekura-fixing-prod-issues"Fix a production call bug / reproduce and test a fix before raising a PR"

Want full functionality?

npx skills add gives you the behavioral guidance layer — the skills auto-activate when you describe relevant tasks. For slash commands and direct API integration, install the full Claude Code plugin marketplace below.


Claude Code (VS Code)

Full plugin support — skills, slash commands, MCP tools, and auto-configured API access.

Install

  1. Open the Claude Code chat panel
  2. Click Manage Plugins > Marketplaces tab
  3. Paste https://github.com/cekura-ai/cekura-skills.git and click Add
  4. Switch to the Plugins tab > search for cekura > install the plugin
  5. Set your API key:
    # Add to ~/.zshrc or ~/.bashrc
    export CEKURA_API_KEY="your-key-here"
    
  6. Restart VS Code to pick up the environment variable
  7. In the Claude Code chat, run /setup-mcp to configure the MCP server

Get Started

Ask "I'm new to Cekura, help me get started" for a guided walkthrough, or ask "what can Cekura do?" to see everything available.

Upgrade

Run /upgrade-skills in any Claude Code session, or manually:

cd ~/.claude/plugins/marketplaces/cekura-skills
git pull origin main

Restart Claude Code after upgrading.


Claude Code (Terminal CLI)

Same full plugin support as VS Code.

Install

  1. Inside a Claude Code session, run /plugins
  2. Go to the Marketplaces tab > select Add Marketplace
  3. Paste https://github.com/cekura-ai/cekura-skills.git and confirm
  4. Go to the Discover tab > search for cekura > install the plugin
  5. Set your API key:
    # Add to ~/.zshrc or ~/.bashrc
    export CEKURA_API_KEY="your-key-here"
    
  6. Restart your terminal and Claude Code session
  7. Run /setup-mcp to configure the MCP server

Tip: If you've already cloned the repo locally, you can paste the local path instead of the GitHub URL.

Get Started

Ask "I'm new to Cekura, help me get started" for a guided walkthrough, or ask "what can Cekura do?" to see everything available.

Upgrade

Run /upgrade-skills in any Claude Code session, or manually:

cd ~/.claude/plugins/marketplaces/cekura-skills
git pull origin main

Restart Claude Code and your terminal after upgrading.


Codex

Codex doesn't support Claude Code plugins directly. Skills are loaded automatically based on conversation context. No slash commands or MCP tools — uses curl-based API reference instead.

Install

Option A: Install skills (recommended)

python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
  --repo cekura-ai/cekura-skills \
  --path cekura/skills/cekura-onboarding \
         cekura/skills/cekura-create-agent \
         cekura/skills/cekura-self-improving-agent \
         cekura/skills/cekura-metric-design \
         cekura/skills/cekura-metric-improvement \
         cekura/skills/cekura-predefined-metrics \
         cekura/skills/cekura-eval-design \
         cekura/skills/cekura-fixing-prod-issues

Restart Codex after install.

Option B: Behavior preset (quick start)

Copy the single-file behavior preset into your repo:

curl -o AGENTS.md https://raw.githubusercontent.com/cekura-ai/cekura-skills/main/codex/AGENTS.md

This gives Codex all the domain knowledge (metric design, eval design, API reference, anti-patterns) in one file.

Get Started

Ask Codex to help with Cekura metrics or evals — skills load automatically when the conversation matches.

Upgrade

Re-run the skill installer:

python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
  --repo cekura-ai/cekura-skills \
  --path cekura/skills/cekura-onboarding \
         cekura/skills/cekura-create-agent \
         cekura/skills/cekura-self-improving-agent \
         cekura/skills/cekura-metric-design \
         cekura/skills/cekura-metric-improvement \
         cekura/skills/cekura-predefined-metrics \
         cekura/skills/cekura-eval-design \
         cekura/skills/cekura-fixing-prod-issues

Cursor

Uses the behavior preset as a rules file. No slash commands or MCP tools — all domain knowledge is embedded in the rules file.

Install

Copy the behavior preset into your project:

curl -o .cursor/rules/cekura.md https://raw.githubusercontent.com/cekura-ai/cekura-skills/main/codex/AGENTS.md

Or add it as a global rule in Cursor Settings > Rules.

Get Started

Ask Cursor to help with Cekura metrics or evals — the rules file provides all the domain context.

Upgrade

Re-download the behavior preset:

curl -o .cursor/rules/cekura.md https://raw.githubusercontent.com/cekura-ai/cekura-skills/main/codex/AGENTS.md

Windsurf / Other Agents

Copy codex/AGENTS.md to wherever your agent reads context files from (project root, .windsurf/rules/, etc.):

Install

curl -o AGENTS.md https://raw.githubusercontent.com/cekura-ai/cekura-skills/main/codex/AGENTS.md

The file contains all Cekura domain knowledge in a single portable format that works with any agent.

Upgrade

Re-download:

curl -o AGENTS.md https://raw.githubusercontent.com/cekura-ai/cekura-skills/main/codex/AGENTS.md

MCP Server

All plugins connect to the Cekura API through an MCP (Model Context Protocol) server. This gives structured access to 84+ Cekura API operations as typed tools.

For Claude Code users: Run /setup-mcp after installing the plugins. It walks you through:

  1. Setting the CEKURA_API_KEY environment variable
  2. Starting the MCP server
  3. Verifying connectivity

For other platforms: The MCP server is optional. The AGENTS.md behavior preset includes API reference with curl examples as a fallback.

How it works: The plugin ships a single .mcp.json file at the marketplace root that auto-configures the connection. When Claude Code starts, it reads the file and connects to the Cekura MCP server at https://api.cekura.ai/mcp. With CEKURA_API_KEY in your environment, all mcp__cekura__* tools become available automatically. Or use OAuth via claude mcp add for a key-free, one-click browser sign-in — see the MCP overview.


Quick Reference

Slash Commands (Claude Code plugin only)

CommandWhat it Does
/cekura-onboardingGuided end-to-end setup — preflight, state-aware resume, walks through agent + metrics + first eval run
/setup-mcpConfigure MCP server (run once after install)
/upgrade-skillsPull latest skill updates from GitHub
/report-bugReport a bug — files GitHub issue, optionally attempts a fix
/create-metricCreate or update a metric
/list-metricsList metrics for an agent or project
/evaluate-callsRun metrics on specific calls
/improve-metricImprove metric accuracy: feedback, labs, auto-improve
/autogen-evalAuto-generate evaluators (or bulk create from CSV/JSON)
/manual-create-update-evalCreate or update a single evaluator with full field walkthrough
/list-evalsList evaluators for an agent or project
/run-evalsExecute test scenarios
/eval-resultsCheck results from a test run
/cekura-reportFull end-to-end quality report — generates 10 evals, runs them, produces structured analysis

Skills (load automatically — both install paths)

SkillWhen it activates
cekura-coordinator"What can Cekura do?" — routes to the right skill
cekura-onboardingFirst-time setup, end-to-end platform walkthrough
cekura-create-agentSetting up an agent — provider, mock tools, KB, dynamic vars
cekura-self-improving-agentAuto-tuning an agent prompt from eval results
cekura-metric-designDesigning or creating metrics
cekura-metric-improvementImproving an existing metric via feedback iteration
cekura-predefined-metricsExploring built-in metrics — what each does, costs, constraints
cekura-eval-designDesigning test scenarios for a voice agent
cekura-fixing-prod-issuesFixing a production call bug — debug, reproduce, fix, verify, regression test, PR

Getting Started Flow

  1. /setup-mcp — Configure API access (Claude Code plugin only)
  2. /cekura-onboarding — Guided platform setup (preflight + state-aware walkthrough). Or just ask "I'm new to Cekura" to activate the skill directly.
  3. Ask "set up my agent" — activates cekura-create-agent
  4. /autogen-eval — Auto-generate test scenarios
  5. /run-evals — Run your first tests
  6. Ask "create a metric for X" — activates cekura-metric-design
  7. /cekura-report — Full end-to-end quality report for any agent

Platform Compatibility

PlatformMethodFull Plugin SupportMCP ToolsSlash Commands
Any Agent Skills clientnpx skills addSkills onlyNoNo
Claude Code (VS Code)Marketplace installYesYesYes
Claude Code (CLI)/plugins installYesYesYes
CodexSkill installerSkills onlyNoNo
CursorRules fileBehavior presetNoNo
WindsurfRules fileBehavior presetNoNo
Other agentsCopy AGENTS.mdBehavior presetNoNo

Reinstalling Cekura skills

If /upgrade-skills reports stale plugin entries, or /plugin install fails with Source path does not exist, your Claude Code install needs a clean reinstall. Run these 4 commands in order in any Claude Code session:

  1. /plugin marketplace remove cekura-skills
  2. /plugin marketplace add cekura-ai/cekura-skills
  3. /plugin marketplace update cekura-skills
  4. /plugin install cekura@cekura-skills

Step 3 is the critical one — it forces Claude Code to re-read the marketplace manifest. After step 4, claude plugin list should show one cekura@cekura-skills entry, and all 14 slash commands resolve under cekura:*.

Troubleshooting

/plugin install errors with Source path does not exist: Re-run step 3 (/plugin marketplace update cekura-skills) and retry step 4. If that doesn't clear it, fully restart Claude Code (close and reopen the session), then redo steps 2–4.

claude plugin list shows zero cekura entries after upgrading: you've removed the old plugins but haven't installed the new one. Run steps 2–4 above.


Links

Ver o original no GitHub

Bom para

Relacionados