Feature: warn when workflow YAML validates but would fail at runtime (schema vs runner gap)

首先跟踪现有的 workflow YAML schema 验证入口，以及 API 树已经使用的已认证会话。检查 action.yml 文件和 workflow 触发器是如何解析的，然后确定四个 opt-in 运行时检查和 workflow-file-hash 缓存如何与验证并列配合。完成标准是：可忽略的 Information 诊断会在打开和保存时运行，且不会阻止保存。

Summary

Extension validates workflow YAML against the schema, but many mistakes only surface after git push triggers a runner:

• runs-on: referencing a self-hosted label that has no online runner
• uses: owner/repo@ref where ref no longer resolves (deleted tag/branch, moved SHA)
• secrets: inherit on a reusable workflow whose caller does not actually inherit
• permissions: narrower than what a step needs (e.g. contents: read + a step that pushes)
• if: expression referencing a context that is empty for the trigger (e.g. github.event.pull_request.* on push)

Each case validates green locally, then burns a runner minute and a red X on the PR.

Proposal

A "runtime-plausibility" pass, opt-in (github-actions.runtimeChecks.enabled), that runs alongside schema validation and surfaces Information-level diagnostics for:

1. Unresolvable uses: refs (HEAD probe via the authenticated session already used for the API tree)
2. runs-on: labels not present in the repo's runner list
3. permissions: narrower than the union of permissions declared by any resolvable action's action.yml
4. Context references that are empty for the declared on: triggers

None block save; all are dismissible. Runs on open + on save, cached by workflow-file hash.

Why not act / nektos

act runs the whole workflow in Docker; this is a static, seconds-scale lint. Complementary, not overlapping.

Related

• #593 (commit-pinned actions reported unresolved) - same "static analysis of uses:" surface
• #609 (false-positive missing-required-inputs) - related schema-vs-runtime gap in the other direction