Repo-Arch is meant to stay boring to script and easy to read.
repo-arch init
repo-arch flow run --repo .
repo-arch flow run full --repo .
repo-arch flow inspect --repo .repo-arch mine-history --repo .
repo-arch classify --repo .
repo-arch cards --repo .
repo-arch cards --invalidate --repo .
repo-arch accept <card-id>
repo-arch reject <card-id>
repo-arch review listrepo-arch why src/core.ts --json
repo-arch check-diff --base main --json
repo-arch check-stale --json
repo-arch similar "why auth middleware token-only?" --jsonrepo-arch index
repo-arch eval
repo-arch dataset
repo-arch train prepare --repo .
repo-arch train cycle --repo .
repo-arch train resume --repo .
repo-arch train status --repo .
repo-arch train list --repo .
repo-arch train run --repo .Use --json when another tool will read the output. Keep the shape stable and avoid one-off fields unless they are versioned.
- prefer local evidence over guesses
- keep generated cards reviewable
- cache by repo + HEAD where possible
- keep
repo-arch.config.jsonin the repo root - do not make release docs depend on adapters
- keep the CLI contract stable
Use the card review state to tighten training and warnings:
- generate cards
- accept the useful ones
- reject the noisy ones
- run
repo-arch flow run full --repo .again - continue the training loop with
repo-arch train cycle --repo .