mrkeyoor.comAsk

Gitlawb/zerotested 07 Oct 2026commit 99721c7

zero review

Trial Zero if hosted and local model choice fits your workflow and you can supervise the evaluation.

  • 83 packages
23 sto install
    Dev Toolsevaluationupdated 31 Oct 2026
    +3stars / 7d

    Choose Zero for a supervised trial if you want a terminal coding agent with hosted or local model options. Our run on 2026-10-07 at commit 99721c7 recorded a successful npm installation. Build and test stages were skipped. We did not test provider-backed coding, task completion, edit quality or native sandbox enforcement.

    Verdict

    Trial Zero if hosted and local model choice fits your workflow and you can supervise the evaluation. The recorded installation succeeded, but engine warnings, skipped verification and audit advisories remain material limits. Pin the version, investigate dependency findings, inspect sandbox policy and check a representative task in a disposable repository before expanding access.

    We ran it

    Lab card: what happened when we ran zero
    Lab card from our sandbox run
    Screenshot of zero (zero.gitlawb.com)
    zero.gitlawb.com
    Install✓ · 23s83 packages · 186 MB
    Buildn/ano build script
    Testsn/ano test script
    Known vulns31 critical · 0 high · 2 moderate · 0 low (npm audit)
    Repo1539 files~421,590 lines of source · 15.2 MB · 5 CI workflows

    Answers from our run

    Does zero build from source?

    Dependencies installed in 23 seconds (83 packages), and the project has no separate build step. We cloned commit 99721c7 into a clean Debian container with 3 CPUs and no project-specific setup.

    Does zero have tests you can run?

    Not through a standard command: the project exposes no test script or target that our harness could run.

    Does zero have known vulnerabilities in its dependencies?

    npm audit flagged 3 known advisories in the dependency tree, including 1 critical at the time of our run.

    Who should not use zero?

    Teams seeking measured coding quality, runtime performance or successful automation from this review.

    What are the alternatives to zero?

    Codex, Aider. Trial Zero if hosted and local model choice fits your workflow and you can supervise the evaluation.

    Setup3/5Installation completed in the stored 23-second summary, but engine warnings and skipped build/test stages leave runtime setup unverified. See the visible setup rubric.
    Docs5/5The supplied README explicitly covers installation, providers, headless operation, permissions and Go checks. This rates coverage under the visible rubric, not verified correctness.
    Community4/5Issue 1041 contains substantive maintainer guidance and a recorded resolution. The visible rubric assesses that engagement, not adoption or support response times.
    Maturity2/5Installation succeeded, but coding tasks, Go checks and native sandbox enforcement were not exercised. Recorded dependency advisories remain unresolved in the supplied evidence.

    Who it’s for

    Developers evaluating hosted providers and local models through one terminal interface.
    Teams interested in documented interactive sessions and headless JSON workflows.
    Operators prepared to check runtime compatibility, permissions and sandbox enforcement before expanding repository access.

    Who it’s NOT for

    Teams seeking measured coding quality, runtime performance or successful automation from this review.
    Users who require a verified clean dependency audit before installation.
    Contributors expecting our npm-path checks to establish whether the Go suite passes.
    Operators who need evidence of effective native isolation before evaluating an agent.

    Setup reality

    The stored parser summary records a 23-second install, 83 dependency packages and a 186 MB dependency footprint. The raw npm footer says it added 70 packages and audited 71 in 14 seconds. The stored summary and raw npm footer disagree on elapsed time and package counts. Their relationship has not been reconciled; we retain both records without attributing the differences to a verified cause.

    The raw log identifies Node v18.20.8 and npm 10.8.2, despite the summary's lab-node:22 image label. It contains unsupported-engine warnings, including an agent-browser requirement for Node >=24.0.0. Installation exited zero; runtime compatibility remains unverified.

    The README pinned to the measured checkout requires Go 1.26.6+ for source development and documents a separate Linux sandbox helper. The supplied current README requires Go 1.26.10+, says Linux native sandboxing needs Bubblewrap, and describes the main binary acting as its own helper when the separate helper is absent. Current instructions do not establish what the measured checkout enforced.

    What our run establishes

    Zero review: our run recorded a 23-second install. That result supports an installation assessment, while coding quality remains untested. The run took place on 2026-10-07 at commit 99721c7, with full recorded SHA 99721c762f37cd43ac511007a5f51d1846df959e. No new execution was performed for this correction.

    Installation returned exit code 0 without timing out. Build and test were separately marked skipped with the reason "no script". Those skipped stages establish neither passing nor failing tests. The overall "ran" status does not establish completion of a coding workload.

    The detector selected npm even though it recognized Node and Go and found a Makefile. The pinned README documents Go development checks. Our harness did not run them, so the missing npm scripts say nothing about whether the Go suite passes.

    The MrKeyoor testing methodology supplies provenance context. Its current description does not establish that this older run used every procedure it describes.

    Installation records disagree

    The stored parser summary records 23 seconds for installation, 83 dependency packages and 186 MB on disk. The raw npm footer reports 70 packages added and 71 audited in 14 seconds. The stored summary and raw npm footer disagree on elapsed time and package counts. Their relationship has not been reconciled; we retain both records without attributing the differences to a verified cause. The supplied count-scope warning also says parser summaries do not establish exhaustive unique coverage.

    The summary labels the image lab-node:22, while the raw install log identifies Node v18.20.8 and npm 10.8.2. Unsupported-engine warnings show that installation success alone does not demonstrate supported runtime compatibility.

    Our recorded npm audit reported 3 advisories: 1 critical and 2 moderate. These are dependency advisories, not demonstrated vulnerabilities in a Zero workflow. The supplied record does not identify affected packages, reachable exploit paths or scannedPackages, so scan coverage is unknown. Investigate the package-level findings before granting access to source credentials. This record does not establish whether a current release still has those findings.

    Why consider a supervised trial?

    Zero can inspect repositories, edit files, run commands and retain local sessions. The Zero README documents hosted providers alongside Ollama, LM Studio and compatible endpoints. It also describes interactive model selection, resumable sessions and headless text, JSON and stream-JSON operation.

    These are documented capabilities. Our run did not connect a provider, submit a coding task, exercise an isolated worktree or validate CI exit behavior. If those capabilities fit your workflow, the next step is one representative task in a disposable repository, followed by an independently checked diff.

    Check sandbox guidance against the installed version

    The pinned README tells Linux source builders to build zero-linux-sandbox for native sandboxing. Issue 1041 reports reduced isolation after a manual build at a different commit, f30f550. That report is not a sandbox failure observed in our run.

    The supplied issue page records closure as completed on October 12, 2026. The current README describes a different helper arrangement: Linux needs Bubblewrap, and the main Zero binary re-executes itself as the helper when zero-linux-sandbox is absent. A separate helper next to the binary or on PATH takes precedence. The historical report therefore cannot establish an unchanged current helper requirement.

    Before granting shell or write access, use the documented zero doctor and zero sandbox policy commands to inspect setup and sandbox state. These commands were not executed for this correction, and our stored run did not verify enforcement. Inspecting policy is a preliminary check; it does not replace testing the installed version's behavior.

    Alternatives for specific workflows

    Codex CLI supports signing in with a ChatGPT account. Its README makes it a candidate replacement when that account-based workflow is your priority. The supplied evidence does not establish stronger isolation or better coding results than Zero.

    Aider automatically commits changes with commit messages. Its README also documents linting and testing after changes. Consider it when automatic commits are central to reviewing and undoing edits. We did not compare its results with Zero.

    These alternatives answer workflow preferences. The supplied evidence does not establish which tool is faster, safer or cheaper.

    What the ratings mean

    The ratings are editorial judgments on a 1-5 scale, not benchmark results. Setup ease rates the observed installation path: 1 means installation was blocked, 3 means it completed with unresolved setup caveats, and 5 requires verified operation without material setup barriers. The score is 3 because installation completed with engine warnings and skipped verification.

    Documentation rates the specificity of the supplied instructions: 1 means little actionable guidance, 3 means basic setup guidance, and 5 means explicit coverage of installation, providers, headless operation, permissions and development checks. The score is 5 for that coverage, without claiming every instruction was validated.

    Community rates visible maintainer engagement in the supplied discussion: 1 means no substantive response in that evidence, 3 means useful guidance, and 5 requires broader evidence of sustained support. The score is 4 because issue 1041 contains substantive maintainer guidance and a recorded resolution. It does not measure adoption or guarantee response times.

    Maturity rates the evidence supporting operational confidence: 1 means a demonstrated blocker prevents the evaluated path, 3 requires a representative workload and verification results, and 5 requires substantially broader operational evidence. The score is 2 because installation succeeded, but coding behavior and sandbox enforcement remain untested and the recorded dependency advisories are unresolved in this evidence.

    Alternatives

    ProjectWhat it isPick it when
    Codex gh↗OpenAI's coding agent that runs locally on your computer.Consider this replacement when signing in through a ChatGPT account is your preferred workflow. The supplied evidence does not establish superior coding quality or isolation.
    Aider gh↗A terminal pair programmer with automatic Git commits and documented linting and testing.Consider this replacement when automatic commits are central to reviewing and undoing edits.

    What people are saying

    1. [hackernews] LeCun has "zero concerns" about AI wiping out humanity, recent "rogue" incidents
    2. [mastodon-trends] Imminent Zero-Day Attack: KiteWorks Urges Customers to Shut Down Servers
    3. [mastodon-trends] From 15,000 to 185 to Zero: Trump’s Claims of Illegal Voting in Nevada Fizzle
    4. [hackernews] First Shader from Zero in Godot 4
    5. [hackernews] ZK-JPEG: Zero-Knowledge Image Editing and Compression
    6. [hackernews] WangNet – 1.8 MB, zero-dependency Numberwang adjudication in 11 languages

    Sources

    1. Zero repository and README
    2. Zero v0.9.0 release
    3. Issue 1041: degraded Linux sandbox after a manual build
    4. Issue 1103: non-hermetic tests
    5. Issue 1118: missing ACP usage records

    More dev tools reviews

    Related reading