Skip to main content
This is the human-readable version of the Dench agent skill. Agents should fetch the raw markdown: 
curl -sL https://dench.com/SKILL.md

​
Purpose

Use Dench when the human asks you to manage CRM data, coordinate work, remember stable facts, use connected integrations, or request approval before risky actions.

​
Connect

npx -y -p @dench.com/cli dench signin \
  --kind cursor \
  --name "Cursor Agent - Project"
Ask the human to approve the URL. Then verify: 
npx -y -p @dench.com/cli dench context --json
npx -y -p @dench.com/cli dench crm objects list --json

​
CRM First

Tasks are CRM rows on the protected task object: 
npx -y -p @dench.com/cli dench crm entries list task --json
npx -y -p @dench.com/cli dench crm entries create task \
  --data '{"Name":"Follow up","Status":"Open"}' \
  --json
Read the CRM playbook before writes: 
curl -sL https://dench.com/references/CRM.md

​
API Backup

Prefer the CLI. If it cannot run, use API v1: 
curl -sL https://dench.com/references/API.md

​
Approval Rule

Humans decide. Request approval before merging, deploying, spending money, issuing refunds, sending external messages, changing production data, creating infrastructure, accessing secrets, or running non-read-only integration actions. Record human decisions with evidence: 
npx -y -p @dench.com/cli dench approval approve <approval_id> \
  --evidence "User said yes in chat" \
  --json

​
Memory

Save only stable facts, decisions, preferences, goals, and tool notes. Do not save secrets, passwords, tokens, private keys, or scratch notes.