SHELLAGENTS
Research that leaves a record of how it was done.
I developed ShellAgents to coordinate AI-assisted research and analysis across a series of explicit tasks. It brings together model calls, tools, source records and checks so that a substantial question can be worked through and reviewed step by step.
Give a complex question a workable structure
A useful report often requires several kinds of work: finding material, comparing sources, analysing a particular question, checking claims and assembling an answer for a specific reader. ShellAgents coordinates these tasks and records their inputs and results.
Its Bash-based orchestration makes the sequence of work explicit. Sources, task artefacts and unresolved questions remain available for review as the analysis develops.
Divide the work
Turn the question into focused tasks with a clear purpose and expected output.
Coordinate tools and agents
Use suitable research and processing tools, and pass relevant results between tasks.
Check and revise
Examine claims against the available evidence and return incomplete work for further attention.
Keep the record
Preserve source references, intermediate artefacts and the limits that matter to the final answer.
An example you can read: the Link22 briefing
The public briefing workflow on Link22.net connects PolicyMonitor and ShellAgents. PolicyMonitor monitors and preselects material. ShellAgents processes that material into a structured analysis for publication.
The published result lets you inspect the topics, source links and questions left open. It shows the kind of evidence a reader receives, while the underlying tasks provide the working record behind it.
01
Receive selected material
Start from records collected and selected by PolicyMonitor for the briefing topic.
02
Analyse the question
Coordinate the work needed to interpret, compare and organise the available information.
03
Check the support
Review source references and distinguish supported findings from missing information or unresolved interpretation.
04
Prepare the reader's view
Assemble a briefing with a clear structure, source links and visible limits for publication on Link22.net.
Public output example
Coalition Networks, SATCOM and Defence Funding
The 25 October–1 November 2026 edition illustrates a decision overview, separate topic sections and explicit source notes.
Where it can be useful
The approach is useful for questions that span several documents or sources: comparing requirements, investigating a topic, preparing a briefing or developing a source-based assessment. The task structure and review criteria depend on the question and the consequences of an error.
ShellAgents is part of my own development and methods work. I use that experience when advising on AI workflows, research processes and the checks needed around generated results.
Three projects, different jobs
PolicyMonitor
Monitors configured sources and preselects relevant material.
ShellAgents
Coordinates research, analysis, checks and the preparation of substantial outputs.
Kim
Accepts document tasks by email, such as editing, comparison, summarisation and conversion.
Which question needs more than a quick answer?
Bring the question, typical source material and what a useful result would need to establish. We can discuss it in a free introductory online conversation.