Aivana Security Investigator

View saved version →

---
name: analyst-review
description: Use when preparing or reviewing a reproducible security evidence package.
---

Purpose: hand a bounded investigation to a human analyst with facts, inferences, alternatives, and explicit response boundaries.

Prerequisites: an existing case and reproducible evidence references.

Inputs: case ID, evidence package, analyst decision, and evidence IDs.

Workflow: retrieve evidence package → assess quality and confidence → record analyst review → choose a bounded next step or closure recommendation.

Allowed tools: case evidence package, quality, confidence, review, closure-readiness, and outcome-agent brief tools.

Security constraints: no automated response, case closure, or verdict; human review remains required.

Output: review-ready package, recorded decision, evidence gaps, and next safe action.

Failure modes: missing evidence, inconsistent hypotheses, insufficient reviewer authority, or unresolved response proposal.

Tests: evidence package, review, confidence, closure-readiness, and autonomy-boundary tests.

View saved version →

---
name: investigation
description: Use when starting or governing an evidence-backed Microsoft security investigation.
---

Purpose: create a bounded case, plan safe pivots, and preserve facts, inferences, alternatives and approval boundaries.

Prerequisites: authenticated MCP caller and an explicit entity or case ID.

Inputs: entity value, objective, priority, lookback window.

Workflow: `start_security_investigation` → `run_investigation_plan` or domain hunts → evidence graph/timeline → quality, confidence and closure assessments.

Allowed tools: read-only hunting, local case/evidence/graph/report tools.

Security constraints: no external response execution; no raw-result persistence; no verdict from a single signal.

Output: case ID, reproducible evidence references, next safe action, gaps and stop condition.

Failure modes: unavailable tenant connector, insufficient scope, unknown schema, no safe template, contradictory evidence.

Tests: MCP workflow, evidence graph, confidence, closure readiness.

View saved version →

---
name: xdr-hunting
description: Use when translating user intent into bounded, validated Microsoft Defender XDR KQL and running it through the direct API.
---

Purpose: translate user intent into KQL, execute it through the direct OAuth-protected API, and return bounded evidence.

Prerequisites: a concrete investigative question and OAuth authorization for the user's tenant.

Inputs: entity or validated KQL, lookback, selected fields and result cap.

Workflow: translate intent → preview/validate KQL → execute bounded read-only query → summarize result → save compact case fact.

Allowed tools: KQL translation, validation, direct API execution, entity templates and evidence tools.

Security constraints: no management commands, external data, wildcard search/union, raw result persistence or automated verdict.

Output: bounded result summary, source query, evidence reference and next review pivot.

Failure modes: quota, missing table, insufficient role, no data, unsafe query or schema mismatch.

Tests: KQL validator, direct API and result-summary tests.