How It Works
A written procedure, arithmetic that does not drift, and a record that stays true.
Every agent in the fleet is built on the same three layers. The skill does the reasoning, the engine does the arithmetic, and the graph is why the answer is still true next quarter.
- Layers under every agent
- 3
- Agents on those layers
- 9
- Graph they all resolve against
- 1
The Architecture
Three layers, and each one is there for a reason.
The split is not architectural neatness. Each layer exists because the layer above it is bad at that particular job.
Layer 1
The Skill
A set of markdown instructions installed directly into your AI. It knows the frameworks, asks the right questions, and guides you through every phase.
A skill is not a chatbot wrapper and not a form. It is a written procedure that an agent follows: what to ask, in what order, what to do with the answer, and when to stop and hand back to a person.
That it is markdown matters more than it sounds. You can read it. You can see exactly what the agent was told to do before it did it, which is not something you can say about most software that touches a compliance record.
Layer 2
The Engine
Anything that has to give the same answer twice is handed to code rather than to the model.
Language models are poor arithmetic engines and excellent interviewers. So the reasoning about what could go wrong is the agent's, and the maths is handed to code that produces the same answer every time it is run.
Where the job is exposure, that code returns a distribution rather than a single number, because an honest answer to what a breach would cost is a range with a shape, not a point estimate dressed up as precision.
Layer 3
The Graph
A live map of every asset, person, process, and control in your organisation. Compliance never goes stale: when your environment changes, your risk picture updates with it.
Held to an ontology, so a supplier, an incident and a control are typed things with typed relationships rather than paragraphs in a file. That is what lets one agent's work be read correctly by another.
It is also what keeps the agents accurate. No agent is handed the whole store and asked to find the relevant part. Each queries the slice its job needs, which is a far harder thing to get wrong.
The agents do not share a runtime. What keeps them consistent is that third layer: the incident your response agent logs is the same incident your management review reads, and the supplier your vendor agent flags is the one your risk register already knows about.
Running One
What using one actually looks like.
Not a console and not a dashboard. You install it once, then you ask it things in the same place you already work.
Once
Install it
The agents run inside your own Claude Code, against your own records. There is no tenant to provision and nothing to migrate.
claude plugin marketplace add askara-agents \ --source github:Askara-Solutions/askara-agents claude plugin install risk-investigation-agent@askara-agentsTwo ways
Start one
A slash command
/start
The six operating agents: risk, continuity, incidents, awareness, suppliers, endpoints.
Just ask
plain conversation
The three shared agents: governance, assurance, documents.
Then
Ask it something
Real prompts, from four different parts of the standard. This is the whole interface: there is no console to learn.
Clause 4. “We're a new company standing up our ISMS. Walk me through scope, objectives and KPIs.”
Clause 5. “Draft our Information Security Policy Statement as a one-page intent statement, ready for the MD to sign.”
Clause 7. “Compose an awareness campaign for the new Acceptable Use Policy and send it to everyone in the employee database.”
Clause 10. “Capture this audit finding as a nonconformity and link it to the internal audit event.”
The Year
What this actually costs you in hours.
Five cadences, from a weekly beat to an annual set-piece, and almost all of the work sits in the fastest one. The agents are built around that shape rather than around a project plan with an end date, because a management system does not have one.
Your certification body sets the audit dates. Everything else is the tempo the work falls into rather than a schedule the standard imposes: it asks for planned intervals, not particular ones, so those are yours to set.
The Gate
Before any of that: no agent commits anything on its own authority, at any of the three layers. See how the gate works.
See it doing the work.
The architecture is easier to judge from an agent that uses it. The Risk Investigation Agent is the one you can buy today, and it exercises all three layers: the skill interviews you, the engine quantifies the exposure, and what comes out is structured enough to be read by everything downstream.
Need help?
Submit a request below and we'll get back to you within one business day.
Or email us at support@askara.solutions