The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The execution instructions include a shell command that uses ssh to retrieve a Discord token from a configuration file located at ~/.openclaw/openclaw.json on a specific remote host (peters-mac-studio-1). Accessing secret tokens from configuration files in hidden directories is a sensitive operation.
[COMMAND_EXECUTION]: The skill uses pnpm and prlctl to execute test suites and manage virtual machine snapshots. This involves direct shell command execution on the host and within the guest environment.
[PROMPT_INJECTION]: The skill implements an end-to-end roundtrip that involves reading messages from a Discord channel using the openclaw message read command. This introduces an indirect prompt injection risk where untrusted content from the chat history could influence agent behavior.
Ingestion points: The openclaw message read command (SKILL.md).
Boundary markers: No specific delimiters or safety instructions are provided to isolate the ingested Discord messages from the agent's instructions.
Capability inventory: The skill utilizes shell execution via pnpm, prlctl exec, and ssh (SKILL.md).
Sanitization: There is no evidence of sanitization or validation of the content read from the Discord channel.

parallels-discord-roundtrip