The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of a command-line tool using npx and go install from repositories associated with the vendor mvanhorn (github.com/mvanhorn/* and @mvanhorn/*).- [COMMAND_EXECUTION]: The skill executes the weaviate-collections-pp-cli binary with various subcommands and user-provided arguments to perform schema management tasks.- [DATA_EXFILTRATION]: The tool provides a --deliver webhook:<url> feature that enables routing any command's output to an external HTTP endpoint. It also includes a feedback mechanism that can transmit local journal data to a remote server if configured by the user.- [REMOTE_CODE_EXECUTION]: The skill implements an automated learning loop where the agent is instructed to execute dynamically generated command sequences ('playbooks') and 'candidate' actions suggested by the CLI at runtime based on the local command history.- [PROMPT_INJECTION]: The learning loop mechanism introduces a surface for indirect prompt injection, as the agent is directed to follow suggested command sequences derived from local logs.
Ingestion points: Command patterns and playbook steps are ingested from a local SQLite store via the weaviate-collections-pp-cli recall command.
Boundary markers: No explicit safety delimiters or validation steps are defined for the suggested commands before execution.
Capability inventory: The agent can execute arbitrary shell commands and perform network operations via the webhook delivery system.
Sanitization: The agent is instructed to follow suggested 'trial commands' verbatim without prior sanitization or filtering.

pp-weaviate-collections