The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs several npm packages and plugins from the machinepulse-ai and quill-io organizations, which are external third-party sources.
[COMMAND_EXECUTION]: Instructions explicitly direct users to execute the Claude CLI with the --dangerously-load-development-channels flag to load plugins, which is a high-risk operation that bypasses safety and stability checks.
[DYNAMIC_EXECUTION]: The skill relies on dynamically loading sensors and bridge plugins to function, which executes code from external npm packages at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon signals from external sources like Hacker News, Reddit, and market data, which can contain malicious instructions.
Ingestion points: Signals consumed from HackerNewsSensor, RedditSensor, and arbitrary external API fetches defined in sensor code (SKILL.md).
Boundary markers: No delimiters or boundary markers are shown in the implementation examples to separate untrusted signal data from agent instructions.
Capability inventory: The skill documentation suggests capabilities such as writing to local logs via fs.appendFile and triggering external notifications like notifyPagerDuty (SKILL.md).
Sanitization: There is no evidence of sanitization, validation, or escaping of the signal data payload before it is processed by the agent.

world2agent-protocol