The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from ArvanCloud's API endpoints and OpenAPI specifications, creating a potential vector for malicious content to influence agent behavior.
Ingestion points: API responses from napi.arvancloud.ir, ecc.arvancloudapis.ir, and other subdomains, as well as OpenAPI YAML/JSON files downloaded from www.arvancloud.ir/api-docs/.
Boundary markers: No explicit instructions are provided to the agent to treat external data as untrusted or to ignore instructions embedded within it.
Capability inventory: The agent can perform network requests (curl), write to local files (~/.config/arvan/config.json), and execute remote transfers and commands (scp, ssh).
Sanitization: The skill relies on jq for some JSON manipulation but lacks broad sanitization for all external inputs.
[DATA_EXFILTRATION]: The skill instructs the agent to access and transfer sensitive TLS private keys located in the ~/.acme.sh/ directory. While this is essential for certificate deployment, it involves the handling of high-value credentials.
[COMMAND_EXECUTION]: The skill utilizes several command-line tools including curl, jq, dig, ssh, scp, and acme.sh to manage infrastructure and certificates.
[DYNAMIC_EXECUTION]: The skill provides patterns for dynamically generating shell commands, such as the RELOADCMD for acme.sh hooks, by interpolating variables from local configuration and API responses. This mechanism could be vulnerable to injection if the underlying data source is compromised.

arvancloud-api