Credential theft
| Rule | Severity | Meaning |
|---|---|---|
checkout-token-extraction | critical | Reads the token that actions/checkout stores in git config. prt-scan used this to steal the workflow token |
environment-dump | high | Encodes the whole environment, a common way to move every secret out in one value |
secret-file-harvest | high | Reads credential files such as ~/.aws/credentials, ~/.npmrc or SSH keys |
oidc-token-request | medium | Requests a GitHub OIDC token, which can be exchanged for cloud credentials |
runner-memory-read | critical | Reads CI runner process memory, where secrets are held unmasked |
Persistence and escalation
| Rule | Severity | Meaning |
|---|---|---|
install-hook-hidden-script | high | An install hook runs a hidden dotfile script. A wfx-loot variant delivered its stealer this way |
workflow-dispatch-from-code | high | Creates or dispatches workflows from build code |
workflow-injection | high | Writes GitHub Actions workflow files |
forged-status-check | medium | Posts success commit statuses from build code. Can be used to satisfy required checks without running them |
agent-persistence | high | Writes editor or AI agent autostart hooks |
Delivery and exfiltration
| Rule | Severity | Meaning |
|---|---|---|
exfil-endpoint | high | Contacts a known data exfiltration service |
bare-ip-request | medium | Sends requests to a hardcoded public IP address |
remote-shell | high | Pipes a downloaded script into a shell |
encoded-eval | high | Executes base64-decoded code |
obfuscated-code | high | Contains obfuscated code, or a long encoded block in a file that runs during CI |
bun-dropper | high | Downloads the Bun runtime |
secret-scanner-download | high | Downloads a secret scanner |
cloud-metadata | medium | Queries cloud instance metadata |
ai-agent-abuse | critical | Runs AI coding agents with safety checks disabled |
github-repo-creation | high | Creates GitHub repositories with the API |
Campaigns
| Rule | Severity | Meaning |
|---|---|---|
campaign-indicator | critical | Contains an indicator published for a known campaign. See Threat intelligence |