The Agent Skills Directory

[COMMAND_EXECUTION]: Executes various shell commands via the amp CLI (amp plugins list, amp plugins repositories, amp clone, amp plugins import) and git commands (git init, git commit, git push) to manage plugin lifecycles.
[EXTERNAL_DOWNLOADS]: Fetches external content using read_web_page for URL inspection and amp clone to retrieve plugin repositories from the vendor's infrastructure or user-provided URLs.
[DYNAMIC_EXECUTION]: Loads and executes TypeScript or JavaScript plugins using the load_plugin tool. This is the core functionality intended for extending the platform's capabilities.
[INDIRECT_PROMPT_INJECTION]: Processes untrusted content from external URLs via read_web_page.
Ingestion points: read_web_page is used to inspect plugin URLs before import (SKILL.md).
Boundary markers: None explicitly mentioned for handling the output of read_web_page before the agent processes it.
Capability inventory: The skill has the ability to write files to persistent locations (~/.config/amp/plugins/, .amp/plugins/), execute shell commands (git, amp), and load executable code (load_plugin).
Sanitization: The instructions recommend manual review ("Review the imported files") but do not specify technical sanitization or automated validation for the fetched content.

building-plugins