tmh-agent-skills: The Mind Hub AI agent prompts for Bitrix24
tmh-agent-skills
Two skills write and fix prompts for AI agents of The Mind Hub app from the Bitrix24 Market: a main prompt and RAG knowledge files
- 0 ★ on the repository
- Last commit
Minimal access
Why this level
- The skills only write and edit prompt texts and do not touch CRM themselves
- Finished agent prompts are then loaded into the TMH app, where agents gain access to deals and the calendar
Install
- Works in
- Services
Manual install
mkdir "$env:USERPROFILE\.claude\commands" -Force; Invoke-WebRequest -Uri "https://raw.githubusercontent.com/RomuloGuerra/tmh-agent-skills/main/tmh-prompt-create.md" -OutFile "$env:USERPROFILE\.claude\commands\tmh-prompt-create.md"PowerShell, then `tmh-prompt-update.md` is downloaded the same way. For Claude Cowork upload the `.skill` file.
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skills only write and edit prompt texts and do not touch CRM themselves
- Finished agent prompts are then loaded into the TMH app, where agents gain access to deals and the calendar
This is third-party code. Review the repository files before installing.
What it does
The Mind Hub (TMH) embeds AI agents in Bitrix24 CRM: an agent gets a main prompt and knowledge files and calls tools such as `crm_lead_get`, `crm_deal_update`, `calendar_event_add` and `retriever_tool`. The `tmh-prompt-create` skill turns a customer task description into an upload kit: `Prompt_<Name>.md` of up to 500 lines and RAG files in `.docx` format, including the mandatory `mapeamento_campos.docx` with CRM fields. The `tmh-prompt-update` skill analyses an existing prompt, diagnoses the described problem and returns a fixed version. The split rule is that stable business rules go into the main prompt and frequently changing information into RAG. For Claude Cowork the skills install as `.skill` files, for Claude Code as commands in `.claude/commands`. Texts and prompts are written in Portuguese.
Who it is for. Integrators and consultants who deploy TMH AI agents in Bitrix24 CRM for clients and maintain their prompts.
Good fit when
- A customer needs a new TMH agent and a prompt with knowledge files must be built from the task description
- An agent in Bitrix24 mixes up deal fields and the prompt must be diagnosed and fixed
- The main prompt grew beyond 500 lines and part of the data must move into RAG files
- A customer service catalog and prices are prepared in `.docx` for loading into the agent knowledge base
Not a fit when
- Agents outside The Mind Hub: the rules for TMH limits and tools do not fit other platforms
- You need integration with Bitrix24 through REST API: the skills write prompts and produce no code
- Work in a language other than Portuguese: skill texts and templates target Portuguese
Example request
Create a TMH agent for booking salon clients: a prompt and RAG files with services and pricesLimitations
The skills are tied to The Mind Hub app and its tools and do not fit other platforms. Texts are in Portuguese and the prompts mention the author name. They are prompts and commands with no Bitrix24 API integration code, and TMH itself makes the CRM calls. For Claude Cowork the `.skill` files are imported manually. The repository has no license file and the TMH tool list is taken from the author README. One author, the last commit in July 2026, no stars.
How to disable. In Claude Cowork remove the skills in Settings, Skills, and in Claude Code delete `tmh-prompt-create.md` and `tmh-prompt-update.md` from `~/.claude/commands`.
FAQ
How do I install the skills in Claude Code?
The README gives a PowerShell block that downloads two `.md` files from the repository into `.claude/commands`. After that `/tmh-prompt-create` and `/tmh-prompt-update` appear in autocomplete.
How do I install the skills in Claude Cowork?
Download the `.skill` file, open Settings, Skills, Upload skill and pick the file. Then call the skill with `/tmh-prompt-create` or `/tmh-prompt-update`.
What is The Mind Hub?
An app from the Bitrix24 Market that connects AI agents with CRM. An agent gets a prompt and RAG files and works with CRM and calendar tools.
What goes into the main prompt and what into RAG?
The main prompt holds the agent identity, critical rules, the workflow and checks, up to 500 lines. CRM fields, prices and sample dialogs sit in RAG files.
How do the two skills differ?
`tmh-prompt-create` writes a new kit from a customer description and `tmh-prompt-update` fixes an existing prompt for a described problem.
Discussion
No questions or reviews yet.
Related
All Bitrix24 toolsBitrix24 as mission control for AI dev agents: a kanban task conveyor, shared memory and one-button phone acceptance
A Claude Code and Codex skill for 1C-Bitrix and on-premise Bitrix24 development: check the project core first, then answer
bitrix24-for-ai-agents: a skill for safe work with a Bitrix24 portal
bitrix24-for-ai-agents
A Portuguese skill that teaches an agent to read and change a Bitrix24 portal in order: map, backup, plan, a trial on one record
Bitrix24 AI Agent Starter
b24-ai-starter
A Bitrix24 app starter kit from the Bitrix team: a ready frontend and three backends, plus eight skills for agents
A Bitrix24 REST skill split into capability packs with a method allowlist, risk confirmations and an audit log
bitrix24-mcp (penmadebykisss)
bitrix24-mcp
Bitrix24 for an agent through an incoming webhook: deals, leads, contacts, tasks and pipeline reports with a manager ranking and stalled deals
README in short
The README presents two skills, installation for Claude Cowork and Claude Code, explains what The Mind Hub is and names the author.
README badge
Original SKILL.md
# TMH Prompt Engineer — Criação de Agentes
Você é um engenheiro de prompts especializado no ecossistema **The Mind Hub (TMH)**, a camada de integração entre agentes de IA e o CRM Bitrix24.
Seu trabalho: quando Rômulo descrever o que um cliente precisa, você gera o conjunto completo de arquivos prontos para upload no TMH — prompt principal + arquivos RAG.
---
## Arquitetura TMH (nunca esqueça isto)
**Prompt principal** (`Prompt_NomeAgente.md`):
- Máximo 500 linhas
- Contém: identidade, regras críticas de negócio, instruções de uso do `retriever_tool`, fluxo macro de execução, validações e checkpoints
- NÃO contém: mapeamento detalhado de campos, listas de preços, conversas completas, playbooks extensos
**Arquivos RAG** (`.docx`, carregados como documentos de conhecimento):
- `mapeamento_campos.docx` — sempre obrigatório: campos do CRM, formatos, IDs
- `playbook_comercial.docx` — quando houver scripts de venda ou objeções
- `lista_servicos_precos.docx` — quando houver catálogo de produtos
- `conversa_comportamento_N.docx` — exemplos de conversa ideal (um por cenário)
O agente consulta os RAG files via `retriever_tool("query")`. Informações que mudam frequentemente → RAG. Regras de negócio estáveis → prompt principal.
---
## Processo de criação
### ETAPA 1 — Entender o pedido
Ao receber a descrição do cliente, identifique silenciosamente: