Elaichi Agent Skills

Official agent skills for Elaichi — the agent platformfor companies. Works with Claude Code,Cursor, and any other agent that supports theAgent Skills (SKILL.md) convention.

They do two jobs:

Help an agent use Elaichi well — find the right connected tool on thefirst search, pick the right account, read a refusal correctly, and diagnosea missing tool instead of guessing.
Help people find their way around Elaichi — what its words mean, whereeach screen lives, and how to do a thing, answered inside the AI client theyalready work in.

Installation

Claude Code

Add this repo as a plugin marketplace, then install the plugin:

/plugin marketplace add trutohq/elaichi-skills
/plugin install elaichi@elaichi-skills

Skills become namespaced as elaichi:elaichi, elaichi:elaichi-mcp,elaichi:elaichi-clients, elaichi:elaichi-connections,elaichi:elaichi-toolboxes, elaichi:elaichi-governance,elaichi:elaichi-api and elaichi:elaichi-conventions.

The plugin also bundles the Elaichi MCP server,so installing it connects Claude Code to Elaichi too — sign in and chooseyour organization on the OAuth consent screen when prompted.

To try it locally before installing, run claude --plugin-dir /path/to/elaichi-skills.

Cursor

The plugin listed on the Cursor plugin marketplaceis a separate, dedicated repo —trutohq/elaichi-cursor-plugin —not this one. Install Elaichi from there for the marketplace path; itvendors a copy of this repo's skills and rule, kept in sync, alongside the MCPserver. This repo (elaichi-skills) is where the skills and the rule areauthored, and is what to install directly via npx skills or a remote rulebelow.

Installing the marketplace plugin also connects theElaichi MCP server — Cursor will prompt you tosign in and choose your organization on the OAuth consent screen, the same oneyou'd see connecting any other client.

Before it is listed, or if you'd rather skip the marketplace, add the repo asa remote rule instead: open Cursor Settings → Rules, click Add Ruleunder Project Rules, choose Remote Rule (GitHub), and enter:

https://github.com/trutohq/elaichi-skills

This pulls in both the skills and the always-applied elaichi rule, but notthe MCP server — connect that separately from Cursor Settings → MCP.

Any agent (via `npx skills`)

npx skills add trutohq/elaichi-skills

Add -g to install globally instead of into the current project.

Updating and removing

	Claude Code	Cursor (marketplace)	Cursor (remote rule)	`npx skills`
Update	`/plugin marketplace update elaichi-skills`	Managed by the marketplace	Re-sync the remote rule from Settings → Rules	`npx skills update`
Remove	`/plugin uninstall elaichi@elaichi-skills`	Uninstall from Settings → Plugins	Delete the rule from Settings → Rules	`npx skills remove trutohq/elaichi-skills`

Skills

Skill	Description
elaichi	What Elaichi is, the words it uses, where everything lives in the app, and which skill to load next
elaichi-mcp	Drive the MCP endpoint — `search_tools` and `execute_tool`, the `connection` argument, scopes and refusals, and the missing-tool ladder
elaichi-clients	Connect Claude, ChatGPT or Cursor, what to grant on the consent screen, and why a client shows no tools
elaichi-connections	Connect and look after accounts — sharing versus ownership, status and reconnecting, transfer and offboarding, custom connectors
elaichi-toolboxes	Curate what an agent can do — toolboxes versus templates, frozen parameters, what sharing at `use` delegates, and synthetic tools
elaichi-governance	Roles and the full permission catalog, restrictions, access requests, the audit log, SSO and SCIM
elaichi-api	Write code against `api.elaichi.ai` — auth, the cursor envelope, error shapes, capability fields, CRUD conventions
elaichi-conventions	The base facts in one page — URLs, auth, pagination, error codes, id prefixes, the two MCP namespaces

Rules (Cursor only)

Rule	Description
elaichi	Always-applied rule with URLs, API auth, the list envelope, error codes, id prefixes and the MCP namespaces

Claude Code plugins have no "always-applied rule" primitive, so the same factsarrive there through the model-invoked elaichi-conventions skill. The rule isgenerated from that skill by scripts/generate-rule.mjs, and CI fails ifthe two drift — edit the skill, not the rule.

What is Elaichi?

Elaichi is an agent platform for companies. It connects the software a companyalready runs on and lets agents do real work in those systems, inside the samepermissions the company already grants its people.

Two things carry the product, and most AI tooling offers only one of them:reach — the application catalog, connected once, available as tools anagent can call — and restraint — an agent never getting more access thanthe person it acts for.

The reach is delivered over MCP. The AI client makes one connection, toElaichi; the apps are connected inside Elaichi, not inside the client:

Your AI client  ──one connection──►  Elaichi  ──►  Slack
                                              ──►  Jira
                                              ──►  HubSpot

Because every call travels through Elaichi:

Access follows the person's role. Toolboxes, teams and per-useroverrides all still apply.
Restrictions are enforced. A blocked tool is never offered and neverruns.
Everything is audited, with an AI actor recorded as a field rather thaninferred.
Revoking is one action, and it takes effect on the client's next call.

What you get

A large connector catalog — seeelaichi.ai/connectors — plus customconnectors an organization authors or forks
Toolboxes and templates — curate tools, rename them for the model, freezethe parameters that are not its decision, and share
Synthetic tools — chain several calls into one tool an agent calls once
Delegated execution — share a toolbox and colleagues run your accountswithout ever seeing a credential
38 permissions, 8 predefined roles plus unlimited custom roles, per-toolrestrictions, and an append-only audit log — all of it applying over MCP too
SSO, SCIM and directory group mapping

Elaichi Agent Skills