Agent Briefs
These briefs are task contracts for coding agents that edit files, run commands, inspect output, and verify an end state. Each brief uses the same structure:
Goal · Prerequisites · Steps (files + commands) · Expected output · Verify (acceptance) · Pitfalls
If expected output does not appear, the agent should stop at that step and report the mismatch rather than infer success.
Optional Net skill
The Net Claude Code skill provides additional Net-specific API context and examples:
github.com/ai-2070/net-claude-skill — setup steps in Claude Skills
The briefs are self-contained. The skill is useful for broader implementation work or when the brief leads into APIs not covered by its steps.
The briefs
- Wrap and Use an MCP Server — put an existing MCP tool on the mesh and invoke it from an agent.
- Build a Recoverable Capability — serve a native capability and prove it survives a provider failure.
- Generate Typed Tool Bindings — turn a discovered tool into typed, compile-checked client code.
Commands and expected output target the CLI and SDK documented on this branch. Where a step depends on a running mesh, the brief states that prerequisite.