The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads and executes the enumerate-iam tool from a third-party GitHub repository (andresriancho/enumerate-iam). It also installs the dsnap forensic tool and the impacket library via pip, and utilizes the official firebase-admin package from Google.
[COMMAND_EXECUTION]: The skill makes extensive use of powerful cloud CLI tools (aws, gcloud, az) and system utilities (ssh, curl, guestfish) to perform complex infrastructure operations and forensic analysis.
[PRIVILEGE_ESCALATION]: Includes methodologies and payloads for bypassing cloud security controls, such as GCP metadata injection to gain root access and AWS CodeBuild privileged container escapes to target the host system.
[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves ingesting and acting upon untrusted data from external sources (e.g., leaked credentials, S3 bucket contents, SQS message bodies). It lacks sanitization or boundary markers, creating an attack surface where malicious data could influence the agent's shell commands.
Ingestion points: Challenge descriptions, IMDS metadata, SQS messages, bucket objects, and Firestore data.
Boundary markers: None present.
Capability inventory: High-privilege access to AWS/GCP/Azure CLIs, shell execution, and local script runners.
Sanitization: None present.
[DYNAMIC_EXECUTION]: The skill utilizes and provides examples of dynamic code execution techniques, including YAML deserialization vulnerabilities (!!python/object/apply:os.system) and the BASH_FUNC_id%% shell function injection to bypass container entrypoints.

cloud-ctf