The outbox your agent reads before it starts
Between one session and the next you think of four things the agent should know. By the time it is running you have forgotten two of them, and the other two arrive as interruptions in the middle of its work.
Keep a branch the agent reads first, so the things you thought of between sessions are waiting for it rather than lost.
Set it up
- Point the client at https://pando.ink/mcp, which speaks streamable HTTP.
- Authenticate with OAuth 2.1 and dynamic client registration, or paste a key from Agents and API keys as a Bearer token.
The pair
The inbox is the direction everybody thinks of: the agent hits something only you can do and writes you a task. The outbox is the other direction, and it is the one that saves the most time, because an instruction given before the work starts costs nothing and the same instruction given halfway through costs the work already done.
What belongs in it
- Things you noticed since the last run: a bug, a decision, a constraint that changed.
- A correction, so the agent does not repeat something you already told it once.
- The order you want things in, when the order matters and nothing in the code says so.
Not a to-do list of everything. An outbox that fills up is an outbox the agent reads and then ignores, which is worse than not having one.
How the run starts
The agent reads the branch before anything else, does what it says, and empties it as it goes: a handled line is ticked, with a word in the note about what it did. Then you can see, without asking, which of your four things actually landed.
Deeper
Connect an agent, about two minutes · The twelve tools · Which note apps an agent can reach · Who runs this