Scheduled brief

A scheduled brief is a run with no person watching. Allocait’s instructions ask the agent to read your account, compare it with its last run, and report only what changed, as facts about your own rules and recorded levels. A brief is not advice or an approval. Allocait can’t enforce what the agent does, and with a write connection the agent’s tools can still change data. Allocait doesn’t run the schedule for a brief. The schedule you set in your agent’s app decides when it runs. Allocait’s own notifications are the webhooks you set up.

​

Where the behavior comes from

Allocait’s MCP server sends server instructions that describe how to run when no person is present. Whether a client passes them to its agent is up to the client. An Allocait plugin also includes the allocait-brief skill, which spells out those steps. See Install the Claude Code plugin, or Connect ChatGPT and Codex for ChatGPT’s plugin. With no plugin, the server instructions are what the agent has. Connect by signing in, as Connect Claude to Allocait or Connect ChatGPT and Codex shows. A signed-in connection carries standing, which a brief reads from. See Standing.

​

What a run reads

1. list_portfolios. The brief covers each portfolio you have.
2. get_portfolio_snapshot with no prices. Allocait supplies none, and the agent never uses a remembered or estimated one. See Market prices.

On a signed-in connection, the snapshot carries standing with no prices passed. Allocait’s instructions ask the agent to give as_of and state no price. A portfolio with no active scenario has no standing, and the agent is asked to say so and not pick a scenario for you. A range in standing reads below, inside or above. A monitor’s state reads a near_ label only when that monitor has a threshold_pct. The agent is asked to add no near threshold of its own. Over an API key there is no standing. The agent reports a range or a monitor price state only when you give it prices, and it runs no check without them.

​

What counts as a change

The first run states the whole picture and says it is the starting point. Each later run reports only a change since the last run:

• A range left or re-entered.
• A monitor whose state or close_state changed. Only close_state (CONFIRMED or CLEARED) says a daily close confirmed a break.
• A playbook newly listed in playbooks_needing_input, which lists the playbooks with blank cells today.
• A held option newly past its expiry, listed in expired_options. The agent reports it as expired and does not guess what it became.

Allocait stores no run history, so an agent needs its own note of the last run to report only what changed. Allocait’s instructions ask the agent to add no threshold of its own, such as how near an edge counts or how old a level is, and not to tell you to buy, sell or hold anything.

​

What a run doesn’t do

Allocait’s instructions tell an agent with a write connection and no person present to change nothing and to put what it would change in the brief. A read-only connection has no tools that change your portfolio or research data. The allocait-brief skill and the instructions ask for these steps:

• Input values. If you ask an agent to keep an Input current from your email, it is asked to record nothing on its own and to list the values it would record, read as written from the source the Input’s description names. It is asked to send no null, because a null deletes a value.
• Rules and holdings. It is asked to leave them as they are.

To act on a proposed change, ask the agent for it in a chat.

​

An example brief

Illustrative example. The figures are hypothetical, not a recommendation.

Standing as of Oct 21, 5:35 pm. 1 range left: ALFA is above its size range. 1 monitor changed: BRAV closed past its level (1 close, Oct 21). 3 blanks today in the Levels playbook.

The instructions ask the agent to keep a brief short: counts first, then one line per change.

​

Next steps