Skip to content

[cli-tools-test] windows workflow uses deprecated missing-tool key, failing strict dry-run compile gate #71046

Closed as not planned

Description

@github-actions

Overview

Bounded exploratory testing found that the trusted windows workflow emits a lock but fails the strict dry-run gate because it still declares the deprecated missing-tool safe-output key. Two other selected workflows were rejected by dry-run safety checks; those rejections are coverage limits, not evidence of a compiler defect.

Finding and reproduction

  • Repository: github/gh-aw; tested 2026-10-22.
  • Source: .github/workflows/windows.md, safe-outputs.missing-tool.
  • Tool call: agenticworkflows compile {"workflows":["windows"],"dry_run":true,"strict":true}.
  • Expected: a maintained trusted workflow passes strict validation without deprecated declarations.
  • Actual: the workflow entry reports valid: true, but the compiler batch reports compiler_warning and the overall dry-run gate is failed. Shellcheck passed. Consumers must inspect batch results, not just the workflow entry.
  • Severity: Low; observed once. This blocks using this sample as a successful strict dry-run compile check.
  • Suggested correction: migrate the source declaration to report-missing-tool through the normal authoring and validation flow. No source changes or publication were performed during this test.

Coverage and timings

Passed checks: status discovery returned names, engines, triggers, and dependencies in 2.9s. Five logs queries all used count: 3: last-day metadata (3 runs, 6.3s), weekly-issue-summary over seven days (1 run, 7.3s), copilot with a before-run cursor (3 runs, 7.8s), nonexistent workflow (actionable empty response, 0.3s), and future date (empty summary, 1.2s). Explicitly exposed summary files were readable; workflow, engine, timestamps, and count matched filters. Count-limited queries correctly disclosed partial coverage.

Two existing-run audits matched logs IDs, names, conclusions, and durations: §39273204856 succeeded, took 15.4s to audit, and reported tool usage, network counts, job timings, and one created discussion; §39724437444 failed, took 7.8s to audit, and identified agent and conclusion failures. These are existing-run observations, not execution of newly compiled locks. The successful run SHA was 88d48aeea757ac1a3b316aecac1e08cc59b3bbb0; the failed run SHA was 273e7e1773d364c9ca7c400676882f89f35604f8.

Blocked or unfinished required checks: all 3 targeted compile calls used dry_run: true and strict: true, but none passed the overall gate. Windows took 29.0s; Archie 27.6s; Weekly Issue Summary 28.3s (concurrent measurements, above the suggested 5s target; no isolated performance conclusion). Successful compilation and generated-lock validation for all three targets remain unverified. The Windows lock had a frontmatter hash, strict/dry-run metadata, jobs, and staged safe outputs; complete syntax and semantic validation was not independently established.

<b>Bounded compile diagnostics</b>

Windows: The 'missing-tool' safe-output key is deprecated. Use 'report-missing-tool' instead.

Archie: .github/workflows/archie.md:1:1, --dry-run cannot safely execute pre-agent-steps, custom MCP servers.

Weekly Issue Summary: .github/workflows/weekly-issue-summary.md:1:1, --dry-run cannot safely execute steps.

The rejected operations were not retried or modified to bypass the checks.

Not tested (expected exclusion)

  • Synthetic malformed YAML, invalid tool configurations, and missing imports: restricted to trusted repository targets.
  • Private raw logs, directory structure, and independent artifact completeness: outside session scope.
  • Execution of newly compiled locks: dispatch and reruns prohibited.
  • Historical and in-progress audit variants: omitted in favor of two representative completed runs.
  • Redundant date variants and repeat-compilation hash comparison: optional expansion omitted after the three failed gates.

Other unverified details include independent completeness of tool/network detection, created-resource link validation, memory usage, and sub-agent error attribution. No repository-wide health claim is made.

Next actions

Migrate the deprecated Windows key and rerun the bounded compile check. Select future test targets compatible with dry-run restrictions without weakening those restrictions. Required compile coverage is reported incomplete separately.

References: Tester §39727564924, successful audit §39273204856, failed audit §39724437444.

Generated by 🧪 Daily Cli Tools Tester · codex · gpt60 · 124.2 AIC · ⌖ 39.3 AIC · ⊞ 20.7K · ◷

  • expires on Oct 28, 2026, 9:29 PM UTC-08:00
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions