An Agent Skill is one SKILL.md file in the Anthropic format. An agent loads it and follows the instructions in it. Provely ships two kinds: the product skill, and one skill for each compiled integration.

The product skill is provely-outcome-verification. It teaches an agent to verify a side-effecting action before it claims that the action is complete, and Provely renders it for each agent host. Read the agent integrations.

Each compiled integration ships one skill of its own, for example provely-stripe. It names the contracts of that provider and states the loop: begin, act, action-result, verify. The machine plan stays in the signed manifest, and never in the file.

An Agent Skill from someone else can call a provider itself. Provely guards it at the call, exactly as it guards an MCP server. Read what Provely refuses when your agent calls an MCP server.