nemoclaw-maintainer-triage

par nvidia

Triage assisté par IA pour les problèmes et PRs NVIDIA/NemoClaw utilisant le type de problème natif, les champs de projet et la taxonomie canonique des étiquettes. Prend en charge les éléments uniques et par lots…

npx skills add https://github.com/nvidia/nemoclaw --skill nemoclaw-maintainer-triage

NemoClaw Maintainer — Triage

Use Issue Type for issue classification. Use Project fields for priority and lifecycle. Use labels for routing and action queues.

Step 1: Load Canonical Policy

Before evaluating an item, read these files in order:

  1. workflow-policy.md
  2. triage-instructions.md
  3. label-taxonomy.json
  4. examples.md

Do not use another label guide. The policy package is the source of truth for Issue Type, Project fields, labels, confidence, authorization, and output shape.

Step 2: Determine Mode

Single-item mode — the user provides an issue or PR number:

gh issue view <number> --repo NVIDIA/NemoClaw --json number,title,body,labels,url,author,projectItems
gh pr view <number> --repo NVIDIA/NemoClaw --json number,title,body,labels,url,author,files,isDraft,mergeStateStatus,projectItems,statusCheckRollup

Use the command for the item type. For issues, read Issue Type through the GitHub GraphQL API. Read Project Priority and Status from Project 199. Do not infer them from labels.

Batch mode — collect both normal inbox items and unlabeled items:

gh issue list --repo NVIDIA/NemoClaw --state open --label "needs: triage" --limit 50 --json number,title,body,labels,url,author
gh issue list --repo NVIDIA/NemoClaw --state open --limit 50 --json number,title,body,labels,url,author
gh pr list --repo NVIDIA/NemoClaw --state open --label "needs: triage" --limit 50 --json number,title,body,labels,url,author,isDraft,mergeStateStatus
gh pr list --repo NVIDIA/NemoClaw --state open --limit 50 --json number,title,body,labels,url,author,isDraft,mergeStateStatus

Keep unlabeled items from the unfiltered results. Combine them with the needs: triage results. Remove duplicates by item type and number. Process one item at a time.

Step 3: Present the Dry Run

Use the JSON-compatible payload defined by canonical triage-instructions.md. Include:

  • native Issue Type for issues
  • Project Priority and Status recommendations
  • only labels from label-taxonomy.json
  • labels to remove, including a completed needs: triage inbox marker
  • confidence, rationale, questions, and human_review_required
  • the proposed public comment, when one is useful.

Prefer no label over a guessed label. Do not use labels for Issue Type, Priority, Status, or resolution. Do not propose an unknown label. Do not propose PRR during triage.

In batch mode, present each dry run and wait for an explicit apply, skip, or edited write set before moving to the next item.

Step 4: Apply Only the Accepted Write Set

Acceptance authorizes only the fields, labels, and comment in the proposal. Before each write, re-read Issue Type, Project fields, and labels, then resolve all live IDs. Do not store mutable IDs in this skill. If the state differs from the accepted proposal's base state, stop and present an updated proposal for acceptance. Resume writes only after the user accepts the updated proposal.

Apply writes in this order:

  1. Set native Issue Type and accepted Project fields.
  2. Add and remove canonical labels.
  3. Remove needs: triage when the inbox action is complete.
  4. Post the accepted comment, if any.

Do not write if the accepted plan contains a low-confidence inference, an unknown label, or an unauthorized field. Return a corrected proposal.

Step 5: Report

For every applied item, report:

  • Issue Type before and after, when applicable
  • Project Priority and Status before and after
  • labels added and removed
  • whether a comment was posted
  • any proposed write that was skipped and why.

Do not write an external activity log unless the invoking maintainer explicitly asks for one.

Batch Ordering

Prioritize candidates using policy evidence, not labels that duplicate Project Priority:

  1. Security, outage, or data-loss reports that might need Project Priority Urgent or High.
  2. Items that need a response from an author or maintainer.
  3. Items waiting longest for triage.
  4. Remaining items by recency.

Plus de skills de nvidia

compileiq-debug
nvidia
Utilisez quand quelque chose ne va pas : Search() bloque, toutes les évaluations retournent INVALID_SCORE, les scores ne s'améliorent pas, chaque configuration retourne le même nombre, erreurs ptxas…
create-github-pr
nvidia
Créer des pull requests GitHub en utilisant l'interface en ligne de commande gh. Utiliser lorsque l'utilisateur souhaite créer une nouvelle PR, soumettre du code pour révision, ou ouvrir une pull request. Mots-clés de déclenchement -…
nemoclaw-maintainer-cross-issue-sweep
nvidia
Analyse les autres problèmes ouverts pour trouver ceux qu’une PR donnée pourrait également corriger ou casser accidentellement. Génère des opportunités de correctifs adjacents et des risques de contradiction avec fichier:ligne…
fhir-basics
nvidia
Apprend aux agents comment fonctionnent les API FHIR R4, quelles ressources sont disponibles, comment les interroger avec des paramètres de recherche, et comment analyser correctement tous les formats de réponse…
compileiq-validate-result
nvidia
Utiliser APRÈS qu'une recherche soit terminée et AVANT de réclamer un accélérateur ou d'expédier un ACF. Charge le CSV dump_results, extrait les K meilleurs candidats (mono-objectif)…
changelog-audit
nvidia
Auditer le CHANGELOG.md de Warp avant une publication : récupérer les entrées perdues, trier par impact utilisateur, affiner le langage des entrées, ajuster les retours à la ligne et (en mode branche de publication) mettre à jour la comparaison…
maintain-dynamic-plugins
nvidia
Maintenir les chargeurs de plugins dynamiques NeMo Relay, les manifestes, les SDK natifs Rust, le protocole worker gRPC, le SDK worker Python, la documentation, les tests et la couverture du workflow de publication
dgx-diagnose
nvidia
Diagnostiquer les problèmes courants du DGX Station GB300 — plantages CUDA, ciblage incorrect du GPU, bugs de conteneur vLLM/SGLang, problèmes d'état MIG, erreurs NVLink/Fabric Manager,…