The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill instructs the user to install the memrise or memrise-cli package globally via npm or bun. There is a naming inconsistency between files (the primary README suggests memrise while the reference guide suggests memrise-cli), which increases the risk of a user or agent installing the wrong, potentially malicious, or deprecated package from a public registry.
COMMAND_EXECUTION (LOW): The skill is designed to construct and run shell commands using the memcli binary. While this is the intended use case, it provides a direct interface for the agent to interact with the host system's shell.
PROMPT_INJECTION (LOW): The skill is vulnerable to indirect prompt injection because it ingests untrusted data from the Memrise API (such as course titles or word definitions) that could contain hidden instructions for the AI.
Ingestion points: Data entering the context via memcli courses, memcli words, and memcli levels commands.
Boundary markers: Absent; there are no instructions or delimiters shown to ensure the agent ignores instructions embedded within the retrieved course data.
Capability inventory: Execution of shell commands through the memcli utility.
Sanitization: No evidence of sanitization or filtering of the CLI output before it is processed by the agent.

memrise