Ir al contenido

Releases, tags and versions

Esta página aún no está disponible en tu idioma.

The rule in every repository here is that a claim in a release note has something behind it in the repository that you can run yourself.

The convention is semantic versioning, MAJOR.MINOR.PATCH. Patch for a fix that leaves behaviour alone, minor for something added while existing commands keep working, major for a change that breaks what you depend on. Tags carry the version with a v prefix, and in molo the tag is what publishes: CI builds the macOS .app and .dmg, the Windows executable and the Android APK on every push, then attaches all three to a release on a v* tag.

A licence change is a version boundary and gets recorded as one. Plancia is GPL-3.0-or-later today. Versions up to 0.2.0 were MIT and stay MIT.

Four things: what changed, what breaks, what was checked, and what the release does not promise. The fourth is the one usually left out, so here it is in practice, from the repositories themselves.

  • Molo’s builds are unsigned. packaging/sign_and_notarize.sh performs the whole Developer ID flow, and it has not been run end to end, because this machine has no certificate.
  • Molo’s Windows executable and Android APK compile in CI and have never been run on real hardware. Only the macOS build has been launched and used for a full download.
  • Molo verifies the yt-dlp wheel it fetches against the sha256 PyPI publishes. That catches a truncated or corrupted download, the same protection pip gives over HTTPS. It is no signature and would not stop a compromised release.
  • claude-codex-bridge supports the Claude Code 2.1.* and Codex 0.146.* session formats, and every target write is gated on the version. --allow-unsupported-version relaxes the source read gate alone. A note that omits those version prefixes is incomplete, since the tool refuses to run outside them.

If a release claims a check, the check is in the repository

Section titled “If a release claims a check, the check is in the repository”
repository command what it runs
plancia python3 tools/prova.py one hundred and thirty eight checks in about twenty seconds, against a throwaway archive that never touches yours
rada python3 tools/prova.py seventy checks in a couple of seconds, no model and no real memory allocated, including a four hundred round adversarial simulation of the fairness rules
argano pytest and node --test web/lib/*.test.mjs samples produced with ffmpeg, broken, repaired and compared against the originals, with frame counts required to match exactly
claude-codex-bridge python3 tools/verify-drift.py --json a live probe of both CLIs, run before a release or an upgrade

Plancia’s suite also runs in CI, in the prova.yml workflow, and contributors turn it on before every push with git config core.hooksPath .githooks. How a plancia release is cut is written down in docs/RILASCIO.md. Molo enforces its one hard limit with a test, so a patch carrying circumvention code fails CI before it reaches anyone’s inbox. The drift probe in claude-codex-bridge consumes two Claude calls and one Codex call and removes only the probe sessions it created, so it runs manually and weekly on a private macOS runner, while standard CI validates its non-networked command surface and never spends quota.

Synced on 3 September 2026, across 26 public repositories.
standard state
README 12 of 12
licence 12 of 12, nine MIT and three GPL-3.0
description 12 of 12
contributing guide 2 of 12, molo and claude-codex-bridge
pull request template 1 of 12, molo
code of conduct 1 of 12, molo
issue template 0 of 12, as reported
community health score molo 100, claude-codex-bridge 71, the other ten 42

A README, a licence and a description are in place across all of them. Molo is the only repository carrying the full set of community files. Ten repositories have no contributing guide and eleven have no pull request template.