Index of every page on this site: llms.txt
All skills
microsoft avatar

/nvx-triage-ci-run

@fef56ac
by microsoftmicrosoft/nvx857 stars
66
Run it yourself

Triage a failed, cancelled, or hanging microsoft/nvx GitHub Actions run: classify supersession and queueing, find the first causal failure for the exact head, reproduce it on the matching platform, and fix it locally without committing. Use for workflow run URLs or IDs, red pull-request checks, runner, matrix, and artifact-handoff failures, and performance-gate or benchmark-host failures.

  • 1 file
  • 5.7 KB
  • Updated 4 days ago
  • GitHub
Use this Skill: https://skilld.dev/gh/microsoft/nvx/nvx-triage-ci-run

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈103 tokens for metadata: the name and description. ≈1.3k when used: this file.

Description uses 5.1% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈411 characters, or 5.1%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Triage An NVX CI Run

Find the first causal failure of one exact head, reproduce it at the narrowest scope, and fix repository defects without committing or pushing. Commands use python3; use python on Windows.

  1. Record the branch, HEAD, the status of NVX and openvmm/, and any Git operation in progress. Never reset, clean, discard, switch, stash, or update the submodule without permission.

  2. Classify the run before treating it as a defect:

    python3 scripts/ci_triage.py <run URL, run ID, or #PR>

    It resolves exactly one run and reports its attempt, event, and head; the pull request's current head and draft state; a classification; the failed jobs with their first failed step, runner, and matrix values; the first error in each failed log; earlier failures of the same jobs on dev; and open issues that name them. Act on the classification:

    • superseded: report the replacement run and stop, unless the cancelled run's cleanup failed.
    • merge-group-cancelled: check the pull request's queue outcome first; see Superseded merge groups.
    • dev-push-cancelled: dev pushes queue rather than replace each other, so investigate it.
    • benchmark-host-changed: the runner is no longer the host of its series' history, an infrastructure change; accept the host and reset the history only on request, as Expected benchmark hosts describes.

    A run that started while its pull request was Ready keeps running after the pull request becomes a draft; that is not the draft guard failing. Bind every conclusion to that run's attempt and head, and never take an older green run as evidence for the current head.

  3. Read the complete logs of the failed jobs where the excerpt is not enough (gh run view RUN --repo microsoft/nvx --job JOB --log-failed), and the workflow, actions, scripts, and tests as they were at the run's commit (git show SHA:PATH). Separate the first cause from matrix cancellations, cleanup failures, and secondary errors. State one falsifiable hypothesis that a decisive log line and the owning code path support, and classify it as a product, test, workflow, dependency or toolchain, transient infrastructure, or missing-prerequisite failure. Change no code for an external or transient failure unless the repository has a concrete robustness defect. A failed OpenVMM CI of the pinned revision job (openvmm-ci) means that OpenVMM's own CI has not passed a job that NVX requires for the pinned revision, as its annotations say, unless it could not read OpenVMM's runs; python3 scripts/nvx.py check-openvmm-ci checks the pin again. Classify it as OpenVMM's CI result, not an NVX defect: rerun a failed OpenVMM job in OpenVMM, then the openvmm-ci job once OpenVMM's CI has passed the pin, as OpenVMM CI describes.

  4. Reproduce the earliest actionable failure with the narrowest equivalent command, on the run's platform and backend:

    • quality and CLI failures: python3 scripts/local_gates.py --run --only GATE --files PATH..., with the paths that the run's change touched (git diff --name-only BASE SHA), since --only keeps only the gates that those paths select and fails for any other; or the failed step's command. A failed public-API check of check-nvx prints how the public API of nvx differs from its snapshots in nvx/public-api/; a deliberate change updates them with python3 scripts/public_api.py --update;
    • boot, microVM, snapshot, guest-console, and benchmark acceptance failures: nvx-run-microvm, after nvx-prepare-host selects a profile with the run's backend and host type;
    • performance regressions: compare the platform, ABI, processor count, metric direction, sample, and baseline window first, and reproduce only on the same host and series, with nvx-collect-benchmarks.

    Do not switch the checkout to the run's commit, install packages, change runners, elevate privileges, or repair hosts without approval, and report when the exact commit cannot run.

  5. Fix the controlling code path with the smallest coherent change, never by weakening assertions, disabling tests, broadening retries, swallowing errors, or hiding output. Rerun the reproduction, then the original failed command, then python3 scripts/local_gates.py --run, and add regression coverage when behavior changes.

  6. Do not commit, push, rerun or cancel runs, or change the pull request unless asked. When asked to push and follow CI, follow only the run of the pushed head, with one watcher, and rebind once when a newer head supersedes it.

Report

Report the run's URL, attempt, and head; the failed job and step; the classification; the decisive evidence and the root cause; the files changed; the reproduction and validation commands with their results; any replacement run; blocked checks; and the final status of NVX and openvmm/. Confirm that nothing was committed or pushed.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at fef56ac. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 7 hours ago.

Activeupdated 4 days ago
argument-hint
<workflow run URL or ID, or #pull-request>

README badge

README badge for microsoft/nvx/nvx-triage-ci-run