Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.
CI test-value audit and latency reduction
Intent and acceptance
Owner request (2026-10-13): remove redundant, obsolete, tautological, or
implementation-pinning tests and optimize the valuable remainder. Aim for a
complete PR lint at or below the early-September reference (~9m21s), not a
faster check achieved by hiding failures. Report both elapsed workflow time
(including queue gaps) and execution critical path / total runner seconds.
Use complete successful exact-head runs; a single fast sample is not a stable
performance guarantee.
Base: b85e3c567e598eb826c9774c545bd56aad4c3e47 (PR #2518 already merged).
Recent measurements: PR run 38258683681 = 18m33s, master run 38261750469 =
18m15s. Python files run in separate processes; a stale timing snapshot and
repeated fixture work amplify a ~536-file suite. UI validation and remaining
browser suites currently execute serially in one job. Earlier Model Hub UI
regression now runs daily/manual; Windows platform suites run on master, with
Windows latest installer smoke retained for PRs.
Test-value rule
For each deletion record the exact test/file, the asserted contract, why the assertion has no independent value, and any remaining owning-boundary coverage. Credible candidates are tests of source spelling/internal layout, duplicate regressions for an already-covered failure class, tests whose mocks implement what they assert, dead-feature requirements, and exhaustive matrices with no additional boundary risk. Age, passing history, AI authorship, runtime, or test count alone are not a deletion reason. No deletion quota.
Keep unique security/authorization, data preservation, migration/compatibility, installer, cancellation/failure propagation, and real runtime boundary coverage. Do not add test-only production APIs, relax assertions, catch failures, add retries, or mark valuable checks optional just to meet the latency target. Optimize expensive fixture preparation without sharing mutable test state; keep migration tests genuinely unseeded and hermetic filesystem/process access.
Integration contracts and ownership
One integration PR, owned/reviewed by session sesp2bd67rp9c; no lane opens a
stacked PR or merges. Internal lane commits are integrated and independently
reviewed by the orchestrator before the single PR is pushed.
- Orchestrator:
.github/workflows/,scripts/ci_*, timing snapshot,tests/test_ci_*.py, plan/evidence integration, review/CI delivery. Retain existing required context names and fail-closed aggregates. Every retained Python file must be selected exactly once. Split independent UI tasks only with explicit required dependency wiring. Reuse existing tools. - Python lane:
tests/excludingtest_ci_*.py, plus its own evidence filedocs/plans/ci-test-value-python.md. Optimize measured slow test fixtures and remove proven low-value tests. No production behavior changes, workflow or shard changes; report cross-boundary opportunities rather than editing them. - UI lane:
ui/test files/tooling/config and its own evidence filedocs/plans/ci-test-value-ui.md. No application behavior changes; preserve existing package script interface unless coordinated. No workflow edits. - Independent audit lane: read-only across all tests. Identify concrete deletion/overconstraint candidates and challenge coverage assumptions; no code writes. Its report is evidence, not final review authority.
No API, wire format, persistent schema, allocated IDs, credential or deployment changes. No remote environments or local Avibe restart. Keep the primary checkout and unrelated active worktrees untouched.
Active execution
- Integration branch:
vibe/20261013-ci-test-value, orchestrator worktree.worktrees/avibe/ci-test-value-audit. - Python implementation: session
sesm4rsqtgcq5, branchvibe/20261013-ci-value-python. - UI implementation: session
sesswafk5qzrr, branchvibe/20261013-ci-value-ui. - Independent read-only audit: session
sesvtw3epcdgv, worktree.worktrees/avibe/ci-test-value-independent. - Shared contract commit before dispatch:
e089ae4429aae72d49b448ca4cc59d806cc2f82e. - All three lanes are complete. Python commit
47ff8390eand UI commits9eacb1c7d,a6a6e3e1d,12f71d572were independently inspected and integrated. The read-only auditor approved the two Python duplicate deletions and identified unique routing/state contracts to retain. - Orchestrator has removed shell/toolchain self-tests, split required UI lanes,
and rebuilt a complete timing snapshot. See
ci-test-value-evidence.md. Master169c31333was integrated before the first full-CI measurement. Required checks, native process isolation and real browser coverage remain.
Work sequence
- Collect complete per-file timings from successful PR/master runs; preserve source SHAs/run IDs and inspect phase breakdown for actual work vs waits.
- Parallel test-value audits and targeted fixture/tooling reductions; record deletion rationale and before/after focused measurements.
- Integrate justified removals; refresh shard timings and evaluate balanced plans. Prefer reducing real work before increasing runner concurrency.
- Parallelize the retained UI lanes if still on the critical path. Keep cancellation/skipped/failure dependencies fail-closed and names stable.
- Focused validation + changed Python Ruff + UI build; push a non-draft PR, arm one durable review/CI Watch, and validate full current-head CI and Codex.
- Compare actual CI against target. If unmet, diagnose the remaining measured tail and continue bounded, evidence-led changes rather than claiming success.
Merge requires fresh explicit owner authorization for this new PR; prior permission applied to #2518 only. Final report includes deleted contracts, retained coverage, actual time saved, runner cost, and remaining risks.