AgentLayerโ–ธdocs
Start here

Your first ten minutes

A first session that needs no keys. Say hello, teach the agent something, and see the next session start with it.

This tour needs no API keys and no packs. Launch your host (claude or codex) from the agent's folder. The examples use an agent named Oppi; lines in brackets describe what comes back.

Say hello

The banner is proof the plugin loaded. On the first session it carries a welcome:

  ๐Ÿ™ Agent:     Oppi ยท v1.0.0
  ๐Ÿ‘‹ Welcome:   first session, say hi and I'll suggest where to start
  ๐Ÿง  Knowledge: ~/Documents/Agents/Oppi/knowledge
  ๐Ÿ“ Projects:  ~/Documents/Agents/Oppi/projects

> hi
[Oppi asks "Where should we start?" and offers: build your roadmap (recommended),
 absorb the documents you dropped in at setup, set up skill packs, or a morning brief]

It leaves out any that do not apply. Pick one. The roadmap is an interview that turns the goals you gave at setup into roadmap.html. The morning brief is a read-out of what the agent already knows and what is on today. Whatever you skip, it names again at the end.

Tell it something to remember

> capture this: Acme renews in March, and Alex owns the renewal
[writes a dated note to knowledge/raw/inbox/]

> save as feedback: keep replies under ten lines unless I ask for more
[appends to knowledge/raw/user/feedback.md]

A fact goes to the inbox. A correction about how the agent should work goes to the feedback log. See Capture.

Give it work, then ask what to do

> create a project called "acme-renewal"
[confirms the folder name and a one-line purpose, then creates
 projects/acme-renewal/ with a README and a tasks folder]

> add this to my list: draft the renewal proposal for Acme
[asks when: today, this week, this month or later; creates the task]

> what should I work on?
[proposes up to three things for today, each with a reason, and asks you to confirm]

Tasks are one markdown file each. See Tasks and Attention.

Open the dashboard

> open the dashboard
[rebuilds dashboard.html in the agent's folder and opens it, or gives you its path]

One page shows the day you planned, the project, and what the agent knows so far. See Dashboard.

See it in the sidebar (macOS, optional)

If you run your agents in cmux, the AgentLayer sidebar keeps the same projects, reports and goals beside your terminals, next to every running session.

Compile, exit, come back

> compile my knowledge
[reads the inbox note and the feedback line, files them into knowledge/]

> /exit

On exit the conversation is saved to knowledge/raw/sessions/, with secret values removed. Launch again and ask:

> what do you know about Acme, and where did we leave off?
[answers from the compiled note and the tail of the last session]

You did not paste anything back in. That is the loop.

Where the context came from

WhenWhat loads
Every sessionAGENTS.md (the manual), SOUL.md, IDENTITY.md and USER.md (character, role, you), knowledge/index.md and knowledge/memory/index.md (the catalog and what is active), projects/TASKS.md (the task board)
Each session startToday's date in your timezone, the tail of the last session, today's reports, your pinned reports and recent git activity, capped at 10KB
On demandNotes about you (knowledge/user/*.md), concept articles and project READMEs, read when relevant
By file pathCoding rules, applied when a matching file is in play

Claude Code reads the first row through the @-imports in .claude/CLAUDE.md (run /context to see them). Codex receives the same files when the session starts. The reference for each lane is What loads at session start.

Things to try

  • briefing pulse for a one-minute status check, and sync when you have been away. See Daily rhythm.
  • I'm overwhelmed to empty your head and get today cut to three.
  • tldr when a reply runs long: the agent restarts with the answer in three lines, then walks you through the rest one step at a time.

Skills answer to plain words, as above, or to a command: /agentlayer:briefing morning in Claude Code, $briefing morning in Codex. See Hosts.

On this page