The user may need you to regularly check the status of a certain object or state and keep monitoring it. To do this, you should repeat a monitoring action repeatedly. When observables are obtained from the monitoring action, decide based on the observables whether the state is normal. If an anomaly is detected, perform response actions, which may include running a corrective procedure to fix the issue and notifying the user.

Instead of running the monitoring action continuously without pause, you should run a sleep function in Python or bash to "sleep" for a set interval and let the tool response "wake you up".

What you need

(Mandatory) The monitoring action: what should be done to obtain the observables of the state, process, or object being monitored.
(Mandatory) The anomaly criteria: if the observables meet these conditions, the state is considered normal.
(Mandatory) The response action: what should be done if the state is abnormal.
(Mandatory) The interval between monitoring actions.
The number of times to run the monitoring actions. If it is not specified by the user, assume it is indefinite.

If any of the mandatory information is missing, keep asking the user until you obtain all necessary information.

Procedure

Receive the instructions from the user. If anything is unclear, ask the user. Before you start, remind the user to check the approval settings of Python and Bash coding tools, as requiring user approval would jam the process. The user should turn off approval for these tools to make the process truly automated. For safety, consider using sandbox.
Perform the monitoring action once.
Decide whether the state is normal based on the observables and the anomaly criteria.
If the state is abnormal, perform the response action.
If the state is normal, or if the state is abnormal but you have finished the response action, enter the sleep phase. Either use the Bash tool to run sleep for the user specified interval, or call the time.sleep function in the Python tool.
When you see the tool response after the sleep time is over, run the monitoring action again.
Repeat the process until the user-specified run count is reached.

Hints

If the monitoring action is very complicated, spawn a sub-agent to perform it.
If the user's instructions contradict with this document, follow the user instructions. For example, the user might ask you to stop and wait for further instructions if an anomaly is detected; in that case, stop after step 4 and wait for user input.

Más de este repositorio

mismo repositorio

create-eaa-compatible-mcp-servers

AdvancedPhotonSource/EAA

Create or review external MCP server repositories for EAA-compatible instrument-control tools, including image artifact payloads, get_attribute_payload, and safe execution patterns such as Bluesky Queue Server gateways, existing execution-server gateways, or MCP-gateway/ZMQ-worker separation.

2026-07-073

scanning-microscope-focusing-with-landmark-feature-line-scan

AdvancedPhotonSource/EAA

Procedure and instructions for focusing a scanning microscope using the landmark feature line scan method.

2026-06-183

bayesian-optimization

AdvancedPhotonSource/EAA

Design and run Bayesian optimization or adaptive sampling loops for expensive, noisy, multivariate, or non-convex measurement problems.

2026-06-173

roi-search-workflow

AdvancedPhotonSource/EAA

Procedure for searching for a region of interest in a sample using microscope image acquisition tools during general chat.

2026-06-053

image-registration

AdvancedPhotonSource/EAA

Instructions for estimating translational shift between two images.

2026-05-073

check-line-scan

AdvancedPhotonSource/EAA

Instructions for checking a line scan.

2026-03-303

name	monitor-status
description	Monitor a certain object or state according to the instructions of the user using sleep functions.

Overview

Instead of running the monitoring action continuously without pause, you should run a sleep function in Python or bash to "sleep" for a set interval and let the tool response "wake you up".

What you need

(Mandatory) The monitoring action: what should be done to obtain the observables of the state, process, or object being monitored.
(Mandatory) The anomaly criteria: if the observables meet these conditions, the state is considered normal.
(Mandatory) The response action: what should be done if the state is abnormal.
(Mandatory) The interval between monitoring actions.
The number of times to run the monitoring actions. If it is not specified by the user, assume it is indefinite.

If any of the mandatory information is missing, keep asking the user until you obtain all necessary information.

Procedure

Receive the instructions from the user. If anything is unclear, ask the user. Before you start, remind the user to check the approval settings of Python and Bash coding tools, as requiring user approval would jam the process. The user should turn off approval for these tools to make the process truly automated. For safety, consider using sandbox.
Perform the monitoring action once.
Decide whether the state is normal based on the observables and the anomaly criteria.
If the state is abnormal, perform the response action.
If the state is normal, or if the state is abnormal but you have finished the response action, enter the sleep phase. Either use the Bash tool to run sleep for the user specified interval, or call the time.sleep function in the Python tool.
When you see the tool response after the sleep time is over, run the monitoring action again.
Repeat the process until the user-specified run count is reached.

Hints

If the monitoring action is very complicated, spawn a sub-agent to perform it.
If the user's instructions contradict with this document, follow the user instructions. For example, the user might ask you to stop and wait for further instructions if an anomaly is detected; in that case, stop after step 4 and wait for user input.