The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes npx skills add <source> -y to install packages from URLs defined in catalog.yaml. While the default catalog contains a entry for a well-known vendor, the system is designed to fetch and execute code from arbitrary, untrusted third-party sources defined by the user at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill provides a mechanism to fetch remote content and write it into persistent 'guides' in the guides/ directory, which are then used as instructions for the agent.
Ingestion points: The ADD.md file instructs the agent to fetch remote sources and create instructional deltas in local markdown files.
Boundary markers: The instructions lack boundary markers; SKILL.md explicitly directs the agent to 'Follow the docs' and 'Run every step this agent can run' from these generated files.
Capability inventory: The agent has shell access to run npx, go install, and various platform CLI tools.
Sanitization: There is no evidence of sanitization or validation of the remote content before it is transformed into agent instructions.
[DYNAMIC_EXECUTION]: The skill involves the dynamic generation of instructions (guides) based on remote content, which the agent then executes. This runtime generation and subsequent execution of logic from untrusted sources is a high-risk pattern.
[COMMAND_EXECUTION]: The skill uses shell commands for core functionality, including npx for skill installation and various CLI tools (Cursor, Claude, Codex, Junie) for environment detection and plugin management. It also references go install for specific toolchain setups in the included guides.

install-from-catalog