AI-assisted dev loop — compile overhead (2026)
AI-assisted dev loop — compile overhead (2026)
Section titled “AI-assisted dev loop — compile overhead (2026)”Evidence baseline
Section titled “Evidence baseline”| Signal | Source |
|---|---|
CARGO_TARGET_DIR varies across agent terminals (target-agent-ssot, target-ci-prepush, default target/) | Recent Cursor terminal metadata (same machine, concurrent sessions) |
--quick pre-push was documented as ~30 s while implementation always runs doc lint + doctest-md + drift-check | Drift between local-ci-pre-push and pre_push.rs — corrected 2026-05-11 |
| CI already emits comparable test timings | Nextest JUnit → vox ci test-runtime-report (runner-contract) |
Canonical inner loop (agents)
Section titled “Canonical inner loop (agents)”vox ci dev-loop-audit— confirmCARGO_TARGET_DIRis unset or points at repotarget/.cargo check -p <crate>— fastest compile signal for the crate you touched.cargo nextest run -p <crate> --profile cior filteredcargo test -p <crate> …— prove behavior without compiling the workspace test graph when unnecessary.vox ci pre-push— push readiness only (default or--quick); avoid re-running as a substitute for steps 2–3.--full— parity with CI workspace nextest; use before merge or when touching cross-crate contracts.
Do not rotate CARGO_TARGET_DIR mid-task unless deliberately isolating (e.g. benchmarks); clear side dirs when done (.cursor/rules/build-environment.mdc).
Tooling
Section titled “Tooling”| Artifact | Command / path |
|---|---|
| Pre-push step timings | vox ci pre-push --report-json <path> → pre-push-report.v1.schema.json |
| Pre-push frequency (append-only log) | Env VOX_PREPUSH_AUDIT_LOG=<path> (JSON lines on success) |
| Target-dir / habit audit | vox ci dev-loop-audit [--json] → dev-loop-audit.v1.schema.json |
| Compile lane budgets | vox ci build-timings + docs/ci/build-timings/budgets.json |
Targets and rollout
Section titled “Targets and rollout”| Phase | Goal | Enforcement |
|---|---|---|
| A | Docs + dev-loop-audit warnings | Advisory |
| B | --report-json adoption in agent runbooks | Optional CI artifact later |
| C | Compare VOX_PREPUSH_AUDIT_LOG rates vs test-runtime-report medians | Maintainer review |
Success criteria (local): median edit → first cargo check -p cycle avoids redundant full workspace compiles; fragmentation_risk: none from dev-loop-audit during typical sessions. Stretch: 30–50% wall-clock reduction on crate-scoped tasks vs alternating CARGO_TARGET_DIR + repeated vox ci pre-push.