The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to use various platform-specific CLI tools (including kubectl, heroku, fly, ssh, and docker) to manage and interact with production environments.
[REMOTE_CODE_EXECUTION]: The skill enables the agent to execute arbitrary Python code on a remote production server using the plain portal exec command. This provides a direct method for the agent to manipulate the remote environment's runtime.
[DATA_EXFILTRATION]: The inclusion of a pull command allows the agent to transfer files from a remote production system to the local filesystem, creating a potential path for sensitive data exposure.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by consuming unvalidated output from remote production systems which may be controlled by an attacker or contain malicious data.
Ingestion points: Data is ingested via the output of plain portal exec and the real-time stream from plain portal connect.
Boundary markers: The instructions do not provide any delimiters or boundary markers to differentiate between intended output and potentially malicious embedded instructions.
Capability inventory: The skill has the capability to run shell commands via uv run, perform file system operations (pull/push), and manage network tunnels.
Sanitization: There is no evidence of sanitization or validation of the remote output before the agent processes it.

plain-portal