Build and test troubleshooting
For source checkouts. For packaged-app failures, see Troubleshooting.
Focused verification
Reproduce a code failure with the narrowest relevant command before editing:
cargo nextest run -p providers
cargo nextest run -p orchestration --lib
cargo nextest run -p desktop
npm --prefix crates/ui run typecheck
./scripts/check-architecture.sh./scripts/verify.sh is the canonical full gate. Provider fixture tests can't prove a real account's Codex entitlement; record the interactive live smoke separately.
MCP protocol checks
Deterministic local protocol lane:
./scripts/verify/mcp-live.shRust build disk usage
Cargo incremental compilation is off, so workspace feature and test variants can't grow target/debug/incremental without bound. Compiled dependencies and sccache stay reusable.
Scripted Cargo entrypoints refuse to start when target/debug exceeds 64 GiB or free space drops below 24 GiB locally (8 GiB on GitHub Actions). Override with OPENFLOW_MAX_DEBUG_CACHE_GIB and OPENFLOW_MIN_BUILD_SPACE_GIB; 0 disables that guard, so use it only on purpose.
When a guard trips, delete only the rebuildable debug cache:
./scripts/clean-rust-cache.sh --yesSource files, Git state, release artifacts, and cross-compiled targets are kept. The next Rust build is cold.
