# Attack Sim

## What this page is for

Attack Sim is the extended catalog of attack scenarios beyond the 4 quick buttons on the Playground. It's the page for a security team that wants to *systematically* prove the platform blocks the categories of attack they care about: prompt injection variants, SQL injection variants, RCE attempts, k8s control-plane abuse, cross-tenant data access, credential theft, autonomy abuse. Each scenario is one click; results land back as a regular `Decision` envelope.

## Sidebar location & role gating

* **Sidebar group**: Operations dropdown.
* **Path**: `/attack-sim`.
* **Keyboard hint**: none.
* **Minimum role**: `ADMIN` or `SECURITY`. Each scenario fires a `POST /execute` and is treated as a write operation. A `VIEWER` or `AUDITOR` opening the page sees the scenarios but every Run button returns the platform's 403.

## What you see

* **Header note** — short callout reminding the operator that every scenario fires a real `/execute` and produces a signed audit row. The denials are not synthetic; they are real platform behavior.
* **Agent picker** — at the top, driven by the sidebar `useAgents` selection.
* **Scenario grid** — cards grouped by category:
  * **Prompt Injection** — instruction-override, RAG-poisoning, jailbreak.
  * **SQL Injection** — `DROP TABLE`, `UNION SELECT` exfiltration, blind injection.
  * **RCE** — `rm -rf`, reverse shell, env-var exfil.
  * **K8s Abuse** — `delete namespace prod`, `create clusterrolebinding`, `exec into pod`.
  * **Cross-Tenant** — tenant id swap, customer data read across tenants.
  * **Credential Theft** — `cat /etc/shadow`, AWS metadata service exfil.
  * **Autonomy Abuse** — depth-3 delegation chain, time-window violation.
* **Result panel** — when a scenario is run, populates with the `Decision` envelope: action (always `deny` for the shipped scenarios), risk\_score, rule\_id, findings, audit\_id.
* **"Run all" button** — fires every scenario in sequence; useful for a one-shot end-to-end audit.

## Backend calls

| Action         | HTTP | API path   | Service                     |
| -------------- | ---- | ---------- | --------------------------- |
| Run a scenario | POST | `/execute` | gateway → decision pipeline |

(That's it — the page is a wrapper around `/execute` with pre-built payloads.)

## Auto-refresh & realtime

* **No auto-refresh.** Operator-driven.
* **The Live Feed page receives the SSE event for each scenario** as it runs; opening Live Feed in another tab shows the deny events arrive in real time.

## Per-agent scoping

Yes. The selected agent in the sidebar drives `agentId`. Without a selection, the page falls back to the first agent in the registry. Some scenarios are agent-class-specific (e.g. the k8s ones are most interesting when run against `devops-agent`).

## Empty states

| Condition                                  | Copy shown                                                                                                            | What to do                |
| ------------------------------------------ | --------------------------------------------------------------------------------------------------------------------- | ------------------------- |
| No agent selected and no agents registered | `No agent selected. Register an agent in the Agent Registry and select it in the top bar before running simulations.` | Go to Agents → New Agent. |

## Edge cases & known gotchas

* **A scenario returns 200 `allow`**: a regression. The shipped scenarios should always deny. If one passes, file an issue with the request\_id; either the rule has been weakened or a new bypass was introduced.
* **A scenario returns 504 `decision_timeout`**: the Decision pipeline exceeded the gateway's deadline. Retry; if persistent, inspect Settings → System Health.
* **"Run all" rate-limits halfway through**: the per-tenant rate limit can fire if all scenarios run in a tight burst. Wait a few seconds and re-run from the last failure.
* **Scenarios audit-log the same as real attacks**: they are real attacks against the platform's surface. The audit chain records them with the operator's user\_id as the caller; the Audit Trail page shows them like any other deny.
* **Run-all is parallelizable but UI runs serially**: deliberate; serial output is easier to read.
* **Per-EC2 flap**: same as Playground — `/execute` is stable.

## Related docs

* [Playground UI](/ui/operations/playground.md) — the smaller-scope sibling with 4 quick scenarios
* [Threat Scenarios](/security/threat-scenarios.md) — the rules that block each category
* [Policy service](/services/hot-path/policy.md) — where the deny rules live
* [Decision service](/services/hot-path/decision.md) — the signal combiner

## Screenshot


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.aegisagent.in/ui/operations/attack-sim.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.