Install the executor
RouteRTL is the tool surface an agent can call repeatedly. Python 3.10+ is enough for the first loop; vendor tools enter later.
Build one small project an agent can reason about: a requirement, a declarative project file, a locked dependency, and the first checks. Board validation follows reviewable checks.
RouteRTL is the tool surface an agent can call repeatedly. Python 3.10+ is enough for the first loop; vendor tools enter later.
Start from a concrete requirement: a sensor hub project that needs a UART path, tests, and a reproducible build.
Add a UART IP from the registry. RouteRTL resolves versions, checks license metadata, and writes the lock file an agent can inspect.
Before synthesis, make the loop prove something cheap: dependency shape, lint results, and a runnable simulation entry point.
The point is not only that an agent can call rr. The point is that the flow is explicit enough for an agent to infer the first contract, generate reviewable artifacts, run the checks, and loop with the engineer over concrete evidence.
Generate the contract-first artifacts for this IP from the public interface specification. Explain the inferred ports, timing assumptions, and review points before writing tests.rr sim contract-init hw/ip/uart/uart_rx.vhdrr sim scaffold hw/ip/uartrr sim coverage-map && rr sim run sim/cocotb/test_uart_rx.pyRouteRTL does not replace engineering judgement. It gives an agent a repeatable sequence and gives you artifacts to review at every stage. Use rr hardware doctor as the connectivity preflight before board runs.
rr queue submit synthLong synth, implementation, and bitstream jobs run through the queue so an agent can submit, poll, and preserve the log instead of blocking blindly.
rr loop openBudgeted loop sessions give an agent a supervised way to attempt, inspect, and close with evidence rather than grinding until a green bar appears.
requirements.ymlcontracted behavior and assumptionsproject.ymltarget, top entity, lint profile, and source rootsip.lockresolved IP versions, hashes, and licenseslint findingsstyle and synthesis-risk diagnosticssim resultrequirement coverage and failing witnessesqueue jobsynth, implementation, and bitstream logshw run.jsonboard validation command, result, and context