Agents

Agent skill

A single Markdown file that teaches an agent how to work your accounts — which call to make, what to check first, and what never to do on its own.

Why a skill and not just an API

An agent handed six endpoints will use them literally. It will publish when you meant draft, answer a complaint it should have escalated, and target a platform you have not connected. None of that is a missing endpoint — it is missing judgement.

The skill carries that judgement: check the accounts before choosing platforms, prefer a draft when intent is ambiguous, never send a public reply unless the person explicitly asked for one.

Install

Two commands. The key comes from your environment; the skill file holds no secrets.

shell

export ECHOIA_API_KEY="eko_your_key_here"

mkdir -p ~/.claude/skills/echoia && \
  curl -s https://app.echoia.io/api/v1/skill > ~/.claude/skills/echoia/SKILL.md

The file is served publicly at /api/v1/skill, so you can read it before installing it — and re-fetch it later to pick up changes.

Agents other than Claude

Nothing about the file is Claude-specific: it is Markdown describing endpoints and rules. For another framework, drop it wherever that framework loads instructions from, or paste it into a system prompt. What matters is that the agent reads it before acting.

What it tells the agent

The base URL and how to authenticate from ECHOIA_API_KEY.
Each endpoint, its parameters, and which one fits which request.
That create_post drafts by default, schedules with a date, and publishes only when explicitly asked.
That publishing and replying are public and immediate — and to confirm wording first unless already approved.
How to read errors: a 403 names a missing scope, a 429 means back off or stop.
To check connected accounts before choosing platforms, and to flag ones needing reconnection.

Belt and braces

The skill shapes behaviour; the token enforces it. A skill telling an agent not to publish is useful, but a token without the publish scope is what makes it impossible. Use both — instructions guide, scopes constrain.

Using it

Once installed, ask in plain language. The agent picks the calls, and the skill decides whether a request is a draft or a publish.

examples

"What's scheduled this week?"
"Draft three posts about the new collection for Instagram and LinkedIn."
"Any unanswered questions in the comments?"
"How did last Friday's post do compared to the one before?"

Keep it current

The skill is a snapshot of the file at the moment you downloaded it. When endpoints change, re-run the curl command — an agent working from a stale skill fails in confusing ways rather than obvious ones.

Was this page helpful?