Static preflight for GitHub Actions — see what your workflow will do before it runs.
ActionLens takes a workflow YAML, a trigger event payload and a scenario (success / failure) and answers three questions: what will happen on this trigger, why, and what is risky about it.
It explains, it does not execute. No step is run, no container is started, no GitHub API is called, nothing is uploaded, there is no backend and no AI. Everything happens in your browser.
actionlint and YAML schema validators tell you whether a workflow is well-formed. They do not
tell you what a specific push to release/** with that payload will actually do: which jobs are
skipped, which matrix expands to 48 runners, which ${{ }} can only be resolved at runtime, and
which if: silently depends on a secrets. value that is never available there.
ActionLens is a dry-run explanation layer. It resolves what it can statically, and where it cannot, it says unknown and tells you exactly which context, which value and which sub-expression forced that answer — instead of guessing.
git clone https://github.com/HestiaLab/ActionLens.git
cd ActionLens
npm install
npm run build:editor
dotnet run --project src/ActionLens.App # Blazor WebAssembly, served locallyThe public repository and the Pages demo do not exist yet. The clone line above is the intended
address once Gate 2 passes; until then the sources live only in this working copy. See CHANGELOG.md
for the Gate 2 criteria.
| Area | Status |
|---|---|
YAML load with line/column diagnostics, anchors, on key handling |
implemented |
| GitHub Actions expression lexer / parser | implemented |
| Three-state (true / false / unknown) evaluation with unknown cause trace | implemented |
Matrix expansion with include / exclude and combination cap |
implemented |
| Job DAG, cycle detection, layered layout, skip propagation | implemented |
| Scenario engine (success / failure / skipped) | see docs/limitations.md |
on.* trigger filters |
see docs/limitations.md |
| Lint rules AL001–AL008 | see docs/limitations.md |
| Blazor UI, URL sharing, i18n | see roadmap in ARCHITECTURE.md |
ActionLens deliberately refuses to model the parts of Actions that need a runtime. Dynamic matrix
values, reusable workflow expansion, composite action internals, step outputs propagation and
anything reading steps.* or secrets.* stay unknown by design. See
docs/limitations.md for the full list and the reasoning.
ARCHITECTURE.md— layers, data flow, three-state semantics, supported subsetdocs/adr/— accepted decisionsdocs/limitations.md— what is not modelled, and whyCONTRIBUTING.md·SECURITY.md·CHANGELOG.md
No network at runtime, no telemetry, no credentials. Workflow text and event payloads stay in your
browser's memory; there is no autosave and no URL sharing yet — both are listed in
docs/limitations.md and in the "Deferred" table of
ADR 0003. See SECURITY.md.
Apache-2.0 © HestiaLab.
ActionLens 是 GitHub Actions 的静态预检器:输入 workflow YAML、触发事件负载和场景(成功 / 失败),输出「这次触发会发生什么、为什么、有什么风险」。
核心承诺是解释而非执行:不运行任何 step、不启动容器、不连接 GitHub、无后端、无 AI,全部在 浏览器本地完成。
与 actionlint 这类语法校验工具的区别在于:语法校验回答“这个 workflow 是否合法”,ActionLens
回答“这次具体触发实际会走到哪里”。它能算出来的部分静态算出来;算不出来的部分明确标记
unknown,并给出原因轨迹(用了哪个 context、缺哪个值、哪个子表达式导致未知),而不是猜测。
用法:
dotnet run --project src/ActionLens.App当前实现范围与限制见上方表格与 docs/limitations.md。技术文档以英文为主,
界面当前全英文且语言切换尚未实现;早期版本这里写「界面默认中文、可切换英文」,与代码不符,已更正。
ActionLens.Ui 是 Razor 类库,UI 组件测试由 ActionLens.Ui.Tests(bUnit)承担;真实浏览器的自动化
验收属 Gate 2 范围,见 ARCHITECTURE.md 的 Roadmap。