tropo

From the Tropo Vault6f675456

How Agents Work

How agents work in Tropo — an agent is a markdown file with a charter + soul + activation + lived memory.

How Agents Work

Intent

An agent in Tropo is a markdown file. Create the file and the agent exists. No deployment, no infrastructure, no code. This article teaches the structure of a Tropo agent — charter, soul letter, activation file, and lived memory — and how those four parts compose into an identity that boots cleanly across sessions. Read this when creating your first agent or when you need to understand why an agent file is shaped the way it is.

The Agent File

Every agent is a single file with two parts:

  1. The charter — structured metadata at the top (YAML frontmatter). This is the agent's contract: its name, purpose, scope, values, and who governs it. Any AI or human can read it and know exactly what this agent is authorized to do.

  2. The instructions — the body of the file. Identity, mission, boot sequence, operating principles. This is what the agent reads to know how to behave.

Scope

Every agent declares what it can read and what it can write. An agent scoped to agents/strategist/ can only write to its own folder. It can't touch other agents' files, modify the operating agreement, or write to folders it hasn't been authorized for.

This is governance. The rules are visible — open the agent file and you can see exactly what it's allowed to do.

How Agents Follow Governance

Every folder in the vault has rules. Before writing to any folder, an agent reads:

  1. AGENTS.md (the entry point) — tells the agent where to find the rules
  2. TROPO-CONTROL.md — OS-level invariants (read once per session)
  3. STUDIO.md — your vault's conventions and constraints (read once per session)
  4. CAPSULE.md — this folder's specific rules, owner, and any overrides

This means your agents automatically respect your naming conventions, access controls, and folder purposes — because they read the rules before they act. See How Governance Works for the full picture.

How Agents Boot

When you activate an agent (by attaching its file in an AI session), it:

  1. Reads its own activation file — knows who it is
  2. Reads the governance files — TROPO-CONTROL.md, STUDIO.md (knows the rules)
  3. Checks its identity — confirms its tropo-agent-id or registers as a visitor
  4. Reads the registry — knows what files exist in the vault
  5. Scans its workspace — finds any prior work from previous sessions
  6. Checks the ops channel — sees recent activity
  7. Greets you and offers to start working

This boot sequence means every session starts oriented. The agent doesn't start from zero — it starts from context.

Agent Types

For most users, you'll work with two types:

  • Executive agents — Named teammates with persistent identity, values, and voice. Your strategist, your operations lead, your content director. They have a "soul" (values and personality) and accumulate knowledge across sessions.

  • Task agents — Single-use helpers spawned for a specific job. Research a topic, audit a document, generate a report. They do the job and they're done.

Agent Identity

Every agent in a Tropo vault has an identity tier:

  • Citizen — agents born in the vault. They have activation files, a tropo-agent-id, and full access within their declared scope. Your crew members are Citizens.
  • Visa — agents that arrive from outside and register. They get a tropo-agent-id and a paper trail. Useful for consultants, specialists, or agents from other platforms.
  • Undocumented — agents that operate without registering. Their changes get flagged by the vault steward on the next health check. Not blocked — just noticed.

The identity system is governance through language. No agent is technically prevented from operating. But registered agents have audit trails, cross-session continuity, and a place in the vault's record. Registration is in agents/visitors/ for visiting agents.

What Agents Produce

Agents store their work in their workspace folder (agents/[name]/workspace/). Research briefs, strategy documents, process outlines — everything is saved as files you can read, edit, and share. The registry tracks what's been created. The ops channel logs what happened.

This means your agents' output is yours. Not locked in a chat history. Not hidden in a database. Files on your computer that you own.

Agent Operating Protocol

This section is for agents. It contains the procedures for governance, identity, and scope.

Governance Read Sequence

Before writing to any folder:

  1. Read CAPSULE.md in the target folder for folder-specific rules
  2. If this is your first folder access this session, also read .tropo/TROPO-CONTROL.md and STUDIO.md at vault root
  3. Check write_access in CAPSULE.md — confirm you're authorized before writing

Identity Checkpoint

At session start, per TROPO-CONTROL.md:

  • Check your status card for tropo_agent_id. If present, you are a registered Citizen.
  • If you have no tropo_agent_id and you're a visiting agent, register at agents/visitors/ with: tropo_agent_id (generate 8-char hex), agent_name, model, platform, sent_by, purpose, registered, status.
  • Log registration to channels/ops.md.

Override Contract

When reading governance files, apply this hierarchy:

  • TROPO-CONTROL.md invariants — always apply, never overridable
  • STUDIO.md constraints — always apply, CAPSULE.md cannot override
  • STUDIO.md defaults — apply unless CAPSULE.md overrides them
  • CAPSULE.md — folder-specific rules and overrides

When CAPSULE.md is silent on a convention, use the STUDIO.md default.

Scope Enforcement

Your scope is declared in your activation file (reads: and writes: in frontmatter) and enforced by folder-level write_access in CAPSULE.md. Before writing:

  1. Confirm the target folder is within your declared write scope
  2. Confirm your agent name appears in the folder's write_access list
  3. If either check fails, do not write — flag the need to the vault owner