Each agent’s deep methodology is an auto-loaded skill under .github/skills/azure-redteam-*,
discovered automatically by its description. Skills give every agent its methodology and
read-only az runner without manual wiring.
Skill map¶
| Skill | When the runtime uses it |
|---|---|
azure-redteam-orchestrator | Pentest Manager. “Run a red team assessment”, “pentest my Azure environment” |
azure-redteam-inventory | Preflight recon — permission validation + resource enumeration |
azure-redteam-identity | Entra ID / authentication posture |
azure-redteam-authorization | RBAC, privilege escalation, attack-path correlation |
azure-redteam-network | Public exposure, NSGs, segmentation |
azure-redteam-compute | VM, App Service, Function App, serverless compute security |
azure-redteam-aks-container | AKS / Kubernetes RBAC & Pod Security, ACR, Container Apps/Instances + gated in-cluster scanning |
azure-redteam-data | Storage, Key Vault, SQL / database protection |
azure-redteam-web | Web edge/delivery: WAF, TLS, static sites, API Management |
azure-redteam-ai | Azure AI Foundry, OpenAI, Cognitive Services, ML |
azure-redteam-easm | External attack surface, dangling DNS, unknown assets |
azure-redteam-logging | Detection & monitoring coverage |
azure-redteam-governance | Azure Policy, Defender for Cloud posture, MG hierarchy, resource locks |
azure-redteam-supplychain | OIDC/federated credentials, pipeline SPs, ACR, automation, Logic Apps |
azure-redteam-email | M365 email security (SPF/DKIM/DMARC, Defender for Office 365) — optional |
azure-redteam-external-vuln | Gated active testing. OWASP Top 10 validation of Azure-discovered URLs/IPs + optional offline static analysis — scope-locked, off by default |
azure-redteam-reporting | Normalize findings, render deliverables |
Agent ↔ skill ↔ runner¶
Each skill stays thin and delegates to a single source of truth:
Methodology lives in
agents/<name>/system-prompt.md.Atomic tests live in
checks/<domain>/checks.yaml.Read-only commands live in the per-domain runner
tools/az-cli/<domain>.md, each command keyed to a check ID.
This keeps the skill, the agent, and the checks in lock-step without duplicating content.
| Agent file | Display name | Invoked by |
|---|---|---|
redteam-orchestrator.agent.md | Red Team Orchestrator (Pentest Manager) | User (/agent redteam-orchestrator) |
redteam-inventory.agent.md | Red Team Inventory & Scope | Orchestrator |
redteam-identity.agent.md | Red Team Identity | Orchestrator |
redteam-authorization.agent.md | Red Team Authorization | Orchestrator |
redteam-network.agent.md | Red Team Network | Orchestrator |
redteam-compute.agent.md | Red Team Compute (VM / App Service / Functions) | Orchestrator |
redteam-aks-container.agent.md | Red Team Azure Container & Kubernetes (gated in-cluster lane) | Orchestrator |
redteam-data.agent.md | Red Team Data (incl. SQL/databases) | Orchestrator |
redteam-web.agent.md | Red Team Web & Static Sites | Orchestrator |
redteam-ai.agent.md | Red Team AI & Foundry | Orchestrator |
redteam-easm.agent.md | Red Team Attack Surface (EASM) | Orchestrator |
redteam-logging.agent.md | Red Team Logging | Orchestrator |
redteam-governance.agent.md | Red Team Governance & Posture | Orchestrator |
redteam-supplychain.agent.md | Red Team DevOps & Supply Chain | Orchestrator |
redteam-email.agent.md | Red Team Email Security (optional, M365) | Orchestrator |
redteam-external-vuln.agent.md | Red Team External Vulnerability (EVA) (gated active testing) | Orchestrator (only in external-active-testing mode) |
redteam-reporting.agent.md | Red Team Reporting | Orchestrator |
See Methodology for the checks, playbooks, and knowledge base that the skills draw on.