Skip to content

armgit / avibe

Public Imported from GitHub
0

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.

3.8 KB
Raw Blame

Hermetic integration and distribution checks

Use the repository's existing pytest environment (uv sync --no-install-project), Node supported by the Runtime manifest, and a real avault custody binary on PATH. The fixture exclusively uses a synthetic PAT, a temporary file custody store, temporary HOME/XDG/AVIBE_HOME, app ledger and loopback GitHub HTTP server. It does not use Keychain, installed user secrets or live GitHub writes.

Run focused tests:

PYTHONPATH=. .venv/bin/pytest -q tests/test_feedback_intake.py

Obtain the normal deployment show-runtime-manifest.json and the archive for the local test platform from the accepted release artifact. Set absolute paths:

AVIBE_FEEDBACK_RUNTIME_MANIFEST=/absolute/path/show-runtime-manifest.json \
AVIBE_FEEDBACK_RUNTIME_ARCHIVE=/absolute/path/vibe-show-runtime-node-linux-x64.tgz \
PYTHONPATH=. .venv/bin/pytest -q tests/test_feedback_intake_http.py

The test verifies the fixed manifest SHA, Runtime version 5e31 and platform archive SHA against exactly the executing host platform before extracting into temporary state. The app imports runtime_platform_tag and safe_extract_tar from the normal installed Avibe package; standalone invocation needs that Python environment (or PYTHONPATH=. in the repository). It never executes an unverified archive. Linux x64 must match ea1079eca7…. macOS uses its entry in the same frozen manifest. Platform-specific native modules require the matching archive; a macOS pass is not Linux deployment evidence. The existing core CI pin 5a9a6a52 remains untouched.

vibe-fixture.py invokes the real Avibe CLI and real avault delivery. Its only custody adaptations select the test binary and file store, avoiding automatic macOS Keychain use. Actual named secret resolution, encrypted envelope stdin, start_new_session semantics and child lifecycle execute unmodified product code. The report is read from its app ledger, not Vault stdin. No exec-style fake Vault stands in for the CLI. Production must never install this fixture.

The HTTP case runs actual uvicorn public admission, packaged Runtime, app TS, worker, real Vault CLI, GitHub fake and receipt GET. It checks Unicode/newlines, status-only POST, repeated/conflicting ID, malformed encoding, size/method, minimal status privacy, DB/WAL/SHM route probes, proxy exclusion and a completed upstream write followed by a disconnect. Focused tests cover quotas, capacity, real concurrency, process death, bounded read-only reconciliation and helper resume/deduplication. Real custody tests skip without avault, and HTTP skips without artifacts; those skips do not satisfy commissioning.

For normal wheel publication, build the UI and normal wheel as in the app README. tests/test_feedback_intake_distribution.py accepts AVIBE_FEEDBACK_TEST_WHEEL and verifies helper bytes from the wheel through installed-source selection and built-in snapshot publication. Normal artifact installation remains PM/ops-owned.

Recovery/crash consumers: tests/test_feedback_intake_recovery.py runs actual helper subprocesses, real HTTP receipt/admission and the production app SQLite reserve/claim path with an isolated GitHub write counter. It kills at durable intent boundaries and races original/recovery delivery without another Issue.

Artifact consumers: tests/test_feedback_intake_artifacts.py runs normal and optimized Python, platform/hash rejection and safe/malicious fixture extraction. Set AVIBE_FEEDBACK_LEGACY_PYTHON to an isolated Python environment containing Avibe dependencies on an actual older interpreter without tarfile.data_filter (e.g. Python 3.10.11) for the legacy compatibility check. It must not point at a production service environment. The modern path and existing tests/test_managed_runtime_composite_artifacts.py remain useful alongside it.