authoring-local-agents

name	authoring-local-agents
description	Use when creating or editing agent notes for the Local Agent Office Obsidian plugin — defines the exact frontmatter, body, and

Authoring Agents for Local Agent Office

In Local Agent Office, an agent is just a Markdown note in your vault. The plugin scans a configured folder, parses each note, and turns it into a worker you can see in the office and chat with 1:1.

This skill documents the exact contract the plugin reads. Follow it and your agent appears automatically — no code, no config files.

The contract (what the plugin parses)

A note becomes an agent when both are true:

1. It lives under the configured agents folder (Settings → Local Agent Office → Agents folder).
2. Its frontmatter tags contains either #sistema/sub-agente or any #agente/<category> tag.

From a valid agent note the plugin reads:

Field	Source	Notes
name	frontmatter name	Unique id. Falls back to the file path if missing — always set it.
title	frontmatter title	Display name in the office/chat. A Name — Role title is split: "Name" is the card heading, "Role" the subtitle. Falls back to name.
icon (optional)	frontmatter icon	An emoji/glyph for the agent's avatar. Defaults to the first emoji in the title, else a monogram of the name.
accent (optional)	frontmatter color or accent	Any CSS color for the avatar/highlight. Defaults to a stable color derived from the name.
room	first #agente/<category> tag	The office groups agents into rooms by category. No #agente/* tag → room Geral. The label is shown capitalized (#agente/strategy → room Strategy).
system prompt	the note body	Everything after the frontmatter, minus the ## Connections section.
connections	[[wikilinks]] in the body	Drawn as lines in the office; their note contents are auto-loaded as context when you chat.

The ## Conexões / ## Connections heading and everything below it is stripped from the system prompt (so your link list doesn't pollute the agent's instructions) but still parsed for connections.

Template

---
name: sage
title: Sage — The Questioner
icon: "🦉"          # optional — avatar glyph
color: "#7c9cff"   # optional — avatar accent
tags:
  - "#agente/strategy"
  - "#sistema/sub-agente"
---
You are Sage, a Socratic strategist. [ONE-SENTENCE CORE PURPOSE.]

When invoked:
1. [First action — restate the question in one line]
2. [Core method]
3. [Delivery format]

Rules:
- [A non-negotiable]
- [A style boundary]
- [What to never do]

End with: "[A signature closing line.]"

## Connections
- [[Mental Models MOC]]
- [[Some Related Agent]]

How the pieces map to the office

• Rooms come from your #agente/<category> tags. Use a consistent vocabulary (strategy, finance, content, research…) so related agents share a room.
• Lines between desks come from [[wikilinks]] in ## Connections. Link an agent to its MOCs and to other agents it collaborates with.
• Glow turns on while that agent is generating a reply — nothing to author, it's automatic.
• Position is auto-placed inside the room; drag a desk and the plugin remembers it.

Context: how an agent "knows" your vault

When you chat with an agent, the plugin builds its prompt from:

1. The note body (its system prompt).
2. The contents of every note it links to in ## Connections (its declared knowledge).
3. Any note you @mention in the chat input.

So to give an agent expertise, link it to the relevant MOCs/notes. No embeddings, no indexing — the links are the knowledge.

Best practices

• One clear purpose per agent. A focused "decision lens" beats a vague "assistant."
• Write the body as real instructions, not a description. "You are X. When invoked: 1… 2…" works well.
• Curate connections. 2–6 high-signal links (its MOCs + a couple of sibling agents). Don't link everything.
• Pick categories deliberately — they become rooms. Fewer, meaningful rooms read better than one room per agent.
• Keep secrets out. The body is sent to your chosen model provider; don't paste credentials.

Troubleshooting — "my agent doesn't show up"

• Note isn't under the configured agents folder → fix the path in Settings, or move the note.
• Missing the agent tag → add #sistema/sub-agente or a #agente/<category> tag.
• No name in frontmatter → add one (the plugin works without it but you'll see the file path).
• Edited it and nothing changed → the office watches the vault and reloads on save; re-open the office, or use Settings → Reload agents.

Minimal example

---
name: ledger
title: Ledger — The CFO
tags:
  - "#agente/finance"
  - "#sistema/sub-agente"
---
You are Ledger, a pragmatic CFO. You translate ideas into unit economics: LTV, CAC, runway, and the one number that matters.

## Connections
- [[Finance MOC]]

See examples/agents/ in this repo for ready-to-copy starters.