Devotd Github
NewSecurity

scs0209/laya-guard v0.6.1 — Forced push blocking and a web trial page

The local guardrail laya-guard for AI coding agents blocks forced pushes of the form `git push -f` as mass-delete reviews in v0.6.1, and adds a web trial page that can be checked without installation.

Repositoryscs0209/laya-guard ↗

The local guardrail laya-guard for AI coding agents blocks forced pushes of the form `git push -f` as mass-delete reviews in v0.6.1, and adds a web trial page that can be checked without installation.

GitHub description: Free local guardrail for AI coding agents: every tool_call judged on-device (regex + Laya), allow/review/block

stars
★ 27
Forks
4
Language
Python
License
MIT
Latest release
v0.4.1
Read original source ↗View on GitHub ↗

scs0209/laya-guard v0.6.1 was released on 2026-10-23. It is a local guardrail that reads, on-device with regular expressions and models, the tool calls an AI coding agent is about to execute, and decides whether to allow, review, or block them. The core of this release is that forced pushes such as `git push -f origin main`, which previously passed through, are now blocked as `mass-delete` review targets.

According to the CHANGELOG, the newly caught patterns have widened to `--force*`, `-f` (including when it appears inside `-uf`), and `+ref`, and arguments wrapped in single quotes are caught as well. Conversely, arguments are read split at shell separators, so `git push origin main && rm -rf build` is not mistaken for a forced push. The same release includes a web trial page on a Hugging Face Gradio Space, which lets you paste in commands without installing the package and see the verdict. This page does not change the package's behavior.

The commit message states that a wheel built from the branch was placed into a clean Python 3.12 virtual environment with `--no-deps` and put through offline checks. It confirms that `git push -f origin main` and `git push origin '+main'` are caught as mass-delete while `ls -la` passes, and that `install claude-code --dry-run` and `audit` were also executed. However, the files touched by this commit are only six — the CHANGELOG, the English and Korean guides, pyproject.toml, the Space configuration, and so on — and the pattern modifications are contained in the earlier PR #83.

The judgment should be made with caution. The verification is the project's own self-check, and the Korean 30-item 30/30 and AgentDojo 25/27·92/97 numbers in the document's comparison table are also self-measured and are explicitly labeled as judge-level. The other tools listed as comparison targets likewise have no verified third-party scores, so it is hard to confirm blocking performance from these numbers alone. Also, since the PyPI release only happens after the v0.6.1 tag is attached to the merge commit, 0.6.1 could not be installed at the time of this commit.