kestrel-mcp-role-spec-author

To write an RFC specification for weak models, define a single unambiguous objective and constraints with prescribed SEARCH/REPLACE blocks, mandatory tests, and rollback plans. This deterministic guidance reduces ambiguity and ensures verifiable outcomes for lower-capability agents.

A model instruction RFC is a policy-style specification defining exact task constraints and acceptance criteria for agents. You need one for multi-agent governance coordination to ensure safe, measurable, and consistent machine-assisted engineering iterations.

Draft task instructions for lower-capability local models by prescribing exact code blocks, setting strict guardrails, and defining measurable acceptance criteria. Including a mandatory rollback plan ensures safe iteration and prevents unintended execution errors.

The best way to structure an RFC for multi-agent coordination is to mandate sections for problem, approach, tests, and rollback. This provides measurable acceptance criteria and safe iteration plans required for reliable multi-agent collaboration.

Yes, an RFC for weak models needs inline executable test blocks to comply with the strict template. These test blocks provide deterministic guidance and verifiable acceptance criteria, ensuring the lower-capability agent produces correct outcomes.