The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the prlctl utility to manage virtual machines and execute shell commands inside guest systems, including switching snapshots and running automated test lanes.
[COMMAND_EXECUTION]: Employs sudo and root-level execution via prlctl to perform administrative tasks such as service restarts and global package installations.
[CREDENTIALS_UNSAFE]: Instructs the agent to source sensitive environment variables from the user's profile ($HOME/.profile) and handle API keys for AI providers.
[EXTERNAL_DOWNLOADS]: Fetches software binaries and installers from the public npm registry and local harness servers.
[REMOTE_CODE_EXECUTION]: Installs and executes downloaded software packages and dynamically generated shell scripts within virtualized guest environments as part of the testing workflow.
[DATA_EXFILTRATION]: Facilitates interaction with Discord servers for automated testing validation, which involves reading and sending messages using bot tokens.
[PROMPT_INJECTION]: Processes data from potentially untrusted sources like guest logs and external messages while having access to high-privilege execution tools. 1. Ingestion points: Guest phase logs, Discord channel messages, and registry metadata. 2. Boundary markers: Employs log markers and JSON output flags to delimit data. 3. Capability inventory: Includes prlctl command execution, sudo access, and network operations. 4. Sanitization: The instructions do not specify sanitization or validation of the processed data.

openclaw-parallels-smoke