The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process potentially untrusted data, including application configuration files, container definitions, and system logs.
Ingestion points: The agent is directed to read fly.toml, Dockerfile, and the output of various diagnostic commands like fly logs and sprite-env services list.
Boundary markers: The instructions include clear constraints, stating that "Reviews and investigations are read-only" and explicitly forbidding resource mutations unless specifically authorized for a task.
Capability inventory: The skill facilitates the use of the flyctl and sprite CLI tools, which have the capacity to modify cloud infrastructure, manage secrets, and execute remote commands.
Sanitization: The skill provides robust guidance for verifying the authenticity of internal requests, specifically instructing the agent to verify Fly-Src-Signature headers using Ed25519 public keys provided by the platform.
[EXTERNAL_DOWNLOADS]: The skill references several official resources and documentation sites for well-known infrastructure and data services.
Evidence: The references include links to fly.io, litestream.io, tigrisdata.com, sprites.dev, and the official GitHub repository for the superfly organization. These are recognized as legitimate service providers.

fly-io