The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides scripts and commands that demonstrate how to access and download data from cloud resources if they are misconfigured or if credentials (like a Service Account key) are available.
Evidence: Python and JavaScript snippets for downloading files from Google Cloud Storage (bucket.getFiles() and f.download()), fetching images from Artifact Registry, and querying Firestore documents.
Evidence: Bash commands using curl to list S3 buckets, Azure Blob containers, and MinIO instances.
[COMMAND_EXECUTION]: The skill includes shell commands and Python code that use standard tools to interact with cloud APIs.
Evidence: Use of curl for testing HTTP methods, Firebase sign-ups, and Firestore access.
Evidence: Python code utilizing the requests and cryptography libraries to generate GCP access tokens from Service Account keys.
[INDIRECT_PROMPT_INJECTION]: As a documentation-focused skill, it has a surface for indirect prompt injection if an agent were to process untrusted cloud metadata or bucket contents based on these instructions.
Ingestion points: The skill describes reading data from Cloud Functions, Firestore, S3, and Artifact Registry.
Boundary markers: None present in the instructions.
Capability inventory: Contains file system writes (/tmp/layer_i.tar.gz), network requests (requests.get, curl), and token generation.
Sanitization: No sanitization is mentioned for the data retrieved from external cloud sources.

cloud-iam-deep