The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch and analyze metadata from the muapi.ai service, which serves as a potential vector for indirect prompt injection.
Ingestion points: Data from the external service enters the context through the muapi workflow discover and muapi workflow get commands.
Boundary markers: The agent instructions lack explicit delimiters or instructions to ignore embedded directives in the fetched data.
Capability inventory: The skill can execute shell commands and run multi-step workflows with arbitrary inputs through the provided scripts.
Sanitization: Metadata is processed by the agent's reasoning engine without prior filtering or sanitization.
[COMMAND_EXECUTION]: All provided scripts serve as wrappers for the muapi CLI, executing shell commands with arguments derived from user or agent input.
Evidence: scripts/run-workflow.sh constructs a command using $WORKFLOW_ID and ${ARGS[@]} for workflow execution.
[EXTERNAL_DOWNLOADS]: The skill documentation requires the user to install or upgrade a CLI tool from a local path.
Evidence: pip install --upgrade /path/to/muapi-cli in SKILL.md.

muapi-workflow