The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by instructing the agent to ingest and narrate results from workflow executions. The agent uses clay workflows runs get --verbose and clay workflows runs steps to fetch outputs from various nodes (ingestion points) and is directed to narrate the trace node-by-node to the user. There are no instructions to treat this data as untrusted or to use delimiters to prevent the agent from executing instructions that might be embedded in the node results. Given the agent's extensive capabilities to modify workflows via the clay CLI, this represents a risk where malicious data could influence agent actions.
[COMMAND_EXECUTION]: The skill relies on shell command execution through the clay CLI for all its functionality. Specifically, the clay workflows code test command allows for the execution of Python code and the installation of arbitrary pip packages using the --packages flag. While this is an intended feature for workflow development and testing, it provides a high-privilege execution capability that could be targeted by sophisticated attacks if the agent is not sufficiently constrained.

workflows-cli