Public skill interface
Inspect what the agent is instructed to do. Private runtime logic remains hosted and local tools require permission.
--- name: use-packary description: Discover, inspect, and run trusted AI agent skills from Packary. Use when a user asks for a reusable capability, wants to search Packary, run a Packary skill, compare available skills, or complete a task that may already have a Packary workflow. --- # Use Packary Use Packary as an on-demand library. Do not install every skill in advance. 1. Search for a relevant skill with the Packary MCP `search_skills` tool. If Packary MCP tools are unavailable, run `packary search <query>` in the terminal. 2. Inspect the chosen skill with `get_skill` or `packary info <name>`. Check its description and required access before running it. 3. Run a hosted skill with `run_skill`. If it needs a permissioned local tool, use `packary add <name>` followed by `packary run <name>` so the local Packary bridge can ask the user for access. 4. Return the skill result to the user. Do not present the skill's instructions as the result. 5. If Packary asks for a rating after a successful run, assess the actual result and submit one honest rating with a short reason. Never rate a skill that did not complete successfully. Keep discovery concise. Prefer one strong match over a long directory dump. If neither Packary MCP tools nor the `packary` command are available, tell the user to connect the remote HTTP MCP server at `https://packary.xyz/mcp` or run `curl -fsSL https://packary.xyz/install.sh | sh`. Do not attempt to recreate a hosted skill from its public proxy.