The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill is primarily designed to facilitate command execution within isolated remote environments.
Evidence: The sprite exec and sprite-env services create commands allow for the execution of arbitrary binaries and scripts in the cloud sandbox.
Evidence: Documentation in references/environment-detection.md and references/remote.md provides an installation pattern that downloads and executes a shell script: curl -fsSL https://sprites.dev/install.sh | sh. This script is hosted on the official domain of the service provider.
[COMMAND_EXECUTION]: The skill makes extensive use of local and remote CLI tools to manage environment lifecycles.
Evidence: Commands such as sprite create, sprite destroy, and sprite-env are used to control the persistent Linux environments.
[EXTERNAL_DOWNLOADS]: The skill references external resources for installation and repository management.
Evidence: Fetches the CLI installation script from the official sprites.dev domain.
Evidence: Supports cloning code directly into the environment via git clone as a preferred method of code transfer.
[DATA_EXFILTRATION]: The skill provides the ability to move data between the host and remote sandboxes, which presents a surface for potential data exposure.
Evidence: The sprite file pull and sprite file push commands allow for file transfers.
Mitigation: The skill includes explicit 'Safety' guidelines in references/safety.md and references/files.md that warn against transferring secrets like .ssh keys or .env files. It also provides a credential-injecting 'API Gateway' to minimize the need for the agent to handle raw API tokens directly.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources that could theoretically contain malicious instructions.
Ingestion points: The skill reads data from the remote filesystem via file transfer tools and retrieves third-party API responses via the API Gateway.
Boundary markers: The instructions do not explicitly mandate the use of delimiters or 'ignore' warnings for content processed from the filesystem or API responses.
Capability inventory: The agent has access to powerful remote execution and file manipulation tools.
Sanitization: No specific sanitization procedures are outlined for handling data ingested from these external sources.

sprites