Skip to content

[cli-tools-test] [CLI exploratory testing] compile diagnostics include concurrent run-artifact download output #70793

Closed as not planned

Description

@github-actions

Overview

Bounded exploratory testing observed unrelated run-artifact download messages inside the compile tool's compiler-warning diagnostic while logs/audit calls overlapped. The windows workflow entry was valid, but its overall dry-run gate failed. This is an observed diagnostic-isolation problem; its cause and contribution to the gate failure are unverified because a real deprecated-alias warning was also present.

Reproduction

  1. Through the agenticworkflows server, start three targeted compile calls concurrently: compile({"workflows":["windows"],"dry_run":true,"strict":true}), and equivalent calls for archie and actions-aw-linter-factory.
  2. While they run, call logs({"engine":"copilot","start_date":"-1d","count":3}) and logs({"workflow_name":"test-quality-sentinel","start_date":"-7d","count":3}). Audit representative returned completed runs; this session used 39547935839 and 39547935852.
  3. Inspect the windows compile response: its batch compiler_warning contains artifact-download activity for existing workflow runs.

Expected: compile diagnostics belong to that compile request. Actual: artifact-download warnings and progress messages appear in the compiler warning. Frequency: observed once; concurrency-dependent reproduction is not established. Severity: medium, because unrelated failures can confuse diagnosis and may affect strict gating. No workaround was verified.

Redacted diagnostics and measurements

The windows.md entry returned valid: true, no workflow errors or warnings, and a compiled file. The batch compiler entry returned valid: false, type compiler_warning, including:

Failed to download artifacts for run 39547935839: failed to flatten unified artifact ... no such file or directory
Downloading missing artifacts for run 39546274407: [agent usage]
Downloading missing artifacts for run 39546578576: [agent usage]
Deprecated alias for report-missing-tool.
Scanning pre-compilation inputs for the dry-run warning baseline
Running shellcheck on run steps in windows.lock.yml

The final dry-run gate was failed; shellcheck passed. Archie and actions-aw-linter-factory gates passed. All calls reported compile-only operation and execution unauthorized.

Observed tool durations: status 2.96 s; basic logs 7.18 s; engine-filtered logs 11.62 s; windows-filtered logs 13.26 s; sentinel-filtered logs 17.65 s; future-date logs 4.88 s; missing-workflow logs 0.09 s. Concurrent compile durations: windows 43.86 s, archie 42.76 s, actions-aw-linter-factory 43.14 s. These exceed the aspirational compile target, but overlap and validation costs prevent attributing a regression. Sentinel audit completed in 28.31 s.

Coverage

  • Passed checks: status discovery returned workflow names, engines, triggers and dependencies. Six logs calls used count 3. Exposed JSON summaries showed correct count limits, dates and Copilot engine selection; the nonempty workflow filter returned only Test Quality Sentinel. Missing workflow produced an actionable message; future date and windows lookback returned zero runs. Partial-result warnings were explicit.
  • Passed checks: two existing-run audits completed. Sentinel metadata and duration agreed with logs. Reports exposed job/step timing, token metrics, network and MCP analysis. PR Code Quality Reviewer identified a submitted review on PR 70763 and tool-failure metrics; successful execution must not be equated with zero internal tool failures. Both sampled runs had logs head SHA 2f8b00be1da8121800d713f3835c737b1fd08d2f.
  • Passed checks: all three trusted workflows emitted parseable YAML with strict/dry-run metadata, 64-character frontmatter hashes, agent environment mappings and safe-output jobs. Two of three gates passed. Targets cover a small import-free workflow, MCP plus imports, and scheduled execution with imports. Generated test locks were restored; no workflows were published or executed.
  • Observed problems: mixed compiler diagnostics and the windows gate failure described above. Artifact completeness was not independently checked.
  • Blocked or unfinished required checks: successful strict dry-run acceptance of windows remains unverified because its gate failed. This is not counted as a passing compile.
  • Not tested (expected exclusion): malformed YAML, invalid tool configuration and missing imports require prohibited synthetic fixtures; private raw-log inspection and independent artifact completeness are outside scope; newly compiled lock execution is prohibited; historical/failed-run variants were not available in the selected sample; in-progress audit, very-old-date query, invalid audit ID, repeated compilation/hash stability and further date ranges were omitted as optional variants. Existing-run audit of windows was unavailable because its seven-day query returned no runs. Repository-wide compilation health, full frontmatter preservation and live lifecycle behavior remain unverified.

Environment and next action

Repository: github/gh-aw. Date: 2026-10-21. CLI version was not exposed by status. Investigate request-local diagnostic capture during overlapping server calls and separate unrelated download output from compile warnings. Assess the legitimate deprecated alias independently.

References:

Generated by 🧪 Daily Cli Tools Tester · codex · gpt60 · 150.7 AIC · ⌖ 39.8 AIC · ⊞ 20.7K · ◷

  • expires on Oct 27, 2026, 9:30 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