
Summary
This rule detects creation of secrets in GitHub-supplied CI/CD contexts (GitHub Actions, Dependabot, Codespaces) at organization, repository, environment, or user scope. Secrets injected at runtime can grant durable access to cloud or service credentials, enabling attacker-controlled automation to persist beyond initial access. The rule watches for specific GitHub audit log events: org.create_actions_secret, repo.create_actions_secret, environment.create_actions_secret, org.create_integration_secret, repo.create_integration_secret, and user.create_integration_secret. It excludes bot activity and surfaces key fields such as actor identity, source IP, hashed token, and the token’s programmatic access type to infer whether creation was interactive or token-based. It also captures the secret’s scope and name (github.org, github.repo, github.name) and assesses whether a workflow referencing the new secret was added or changed within a close time frame, indicating potential malicious wiring. The rule maps to MITRE ATT&CK techniques T1098 (Account Manipulation) with subtechnique T1098.001 (Additional Cloud Credentials) and T1552 (Unsecured Credentials), primarily under Persistence (TA0003) and Credential Access (TA0006). It includes an investigation guide, false-positive guidance related to routine secret provisioning or onboarding, and remediation steps such as secret deletion, credential rotation, token revocation, and review of workflows and actor activity. Suppression is supported to group alerts by repository and actor name to reduce noise, with a 60-minute window for suppression when fields are missing. This rule is intended as a building block to be correlated with actor behavior, token usage, and surrounding workflow changes to differentiate legitimate operations from abuse.
Categories
- Cloud
- Web
- Application
Data Sources
- Application Log
ATT&CK Techniques
- T1098
- T1098.001
- T1552
Created: 2026-09-22