The Agent Skills Directory

[Command Argument Interpolation]: The skill directs the agent to run yarn flow $ARGUMENTS. Interpolating arguments into shell commands is a potential surface for command injection if inputs are not strictly validated against a known set of allowed values.
[Indirect Prompt Injection Surface]: The skill involves reading and reporting on Flow type errors within source code, which presents a surface for potential indirect prompt injection.
Ingestion points: The agent parses React and Flow source files in the local workspace as part of its type-checking routine.
Boundary markers: No specific delimiters or explicit safety instructions are provided to guide the agent when it encounters malicious patterns hidden in source code comments or strings.
Capability inventory: The skill leverages shell command execution tools to run build-related scripts.
Sanitization: The instructions do not specify sanitization or filtering steps for the agent when it aggregates and reports type error output to the user.
[Project Script Execution]: The instructions involve executing yarn flow and yarn flow-ci, which are typically defined as scripts in a project's configuration. This triggers the execution of local code, which is the standard expected behavior for type checking but assumes the project environment is trusted.

flow