oncokb-github-issues

Manage OncoKB GitHub issues end-to-end.

SKILL.md

GitHub

Scope

  • Use for issue work only: triage, intake, creation, metadata updates, status comments, duplicate handling, and closure rationale.
  • Default issue repository is https://github.com/oncokb/oncokb-pipeline unless the user explicitly requests a different OncoKB repo.
  • Use repository conventions already in place for labels, milestones, priorities, and ownership.
  • Prefer gh CLI for all GitHub operations.
  • For new issues, have the user pick a ticket kind first: feature, fix, documentation, or chore.
  • For existing issues, do not force ticket-kind selection unless the user asks to reframe the issue.

Ticket templates

Use the matching template after the user picks a ticket kind.

Feature

  • Title: Feature: <concise capability>
  • Body sections:
    • Problem or opportunity
    • Proposed solution
    • Scope and non-goals
    • Acceptance criteria
    • Dependencies or rollout notes

Fix

  • Title: Fix: <concise defect>
  • Body sections:
    • Problem summary
    • Steps to reproduce
    • Observed behavior
    • Expected behavior
    • Proposed fix direction
    • Acceptance criteria

Documentation

  • Title: Docs: <topic to improve>
  • Body sections:
    • Documentation gap
    • Audience and impact
    • Proposed updates
    • Files or pages affected
    • Acceptance criteria

Chore

  • Title: Chore: <maintenance task>
  • Body sections:
    • Maintenance task summary
    • Why now
    • Scope
    • Risks or coordination notes
    • Acceptance criteria

OncoKB required metadata for new issues

  • Always add the ai-generated label on created issues.
  • Project placement depends on near-term execution intent:
    • If the user is about to work on the issue now, add it to https://github.com/orgs/oncokb/projects/3/views/3 (sprint board) and set both sprint board fields: 2024-week and Status.
    • If the user is not about to work on it or this issue is a part of a larger project, ask which project(s) under https://github.com/orgs/oncokb/projects/ should receive the issue and add it there.
    • For non-sprint-board projects, ask the user for start date and target date and set those project fields when available.
    • If the user wants a new project, create it using https://github.com/orgs/oncokb/projects/22 as the template, then place the issue in that new project.
  • Never invent Status values: status selections must come from the destination project's configured status options.
  • Always ask the user which additional tags (labels) they want before finalizing issue metadata.
  • When asking for tags, fetch available labels from oncokb/oncokb-pipeline and present a short suggested subset from that live label list.

Required prompts

For new issue creation, ask for missing inputs in one grouped prompt whenever possible:

  • Ticket kind: feature, fix, documentation, chore
  • Project intent: sprint board now vs non-sprint-board project
  • Non-sprint-board dates (if applicable): start date, target date
  • Additional tags from the live oncokb/oncokb-pipeline label list

Workflow

  1. Confirm target and intent.

    • Identify the repository, issue number (if existing), and requested operation.
    • Unless the user says otherwise, set target repo to oncokb/oncokb-pipeline.
    • For new issues, ask the user to pick one ticket kind: feature, fix, documentation, chore.
    • Confirm project placement intent: being worked now (sprint board) vs queued/planned (another OncoKB project).
    • For non-sprint-board placement, collect start date and target date from the user.
    • If the user request is broad, infer the smallest safe issue action first (for example: draft issue text before opening).
    • Completion criterion: target repo, requested issue action, and all required creation inputs are explicit.
  2. Gather project context before changing issue state.

    • Read the relevant local code/docs and recent issue history as needed.
    • Check for duplicates or related issues with gh issue list/gh issue view/search.
    • Fetch current labels from oncokb/oncokb-pipeline (for example with gh label list --repo oncokb/oncokb-pipeline) and use them for tag suggestions.
    • Fetch allowed project status options from the destination project and only use those values when setting Status.
    • Completion criterion: enough evidence exists to write or update the issue with concrete context and valid metadata choices.
  3. Produce or refine actionable issue content.

    • For new issues, use the selected ticket template (feature, fix, documentation, chore) and keep the section structure.
    • For issue updates, keep existing issue context intact and add only the new decisions, status, or scope deltas.
    • Ensure title is specific and searchable.
    • Ensure body includes clear problem framing, desired outcome, scope boundaries, and concrete acceptance signals.
    • Completion criterion: another contributor can start work without guessing core intent.
  4. Apply metadata using existing conventions.

    • Set labels, assignee(s), milestone, and linked issues/PRs according to repository norms.
    • For new issues, apply required defaults: ai-generated label and project placement based on intent (sprint board + 2024-week + Status when active now; otherwise user-selected OncoKB project).
    • For any project Status field update, use only values listed in that project's status options.
    • For non-sprint-board projects, apply start date and target date provided by the user.
    • If requested, create a new project from template https://github.com/orgs/oncokb/projects/22 and then add the issue to it.
    • Ask the user for any additional tags and apply their selections from the live label list.
    • If a needed label/milestone does not exist, call it out clearly instead of inventing ad-hoc taxonomy unless asked.
    • Completion criterion: issue classification matches current repo conventions.
  5. Execute with gh and report outcome.

    • Create/update/comment/close using gh issue create, gh issue edit, gh issue comment, gh issue close, etc.
    • When closing as duplicate or done, leave a short traceable rationale with links.
    • Report back with issue URL, key metadata changes, and any follow-up actions.
    • Completion criterion: requested issue operation is complete and auditable.

Quality bar

  • Be precise, not generic: avoid placeholder issue text.
  • Keep history legible: prefer one strong issue update over many fragmented comments.
  • Preserve user intent; do not silently re-scope priorities or ownership.
  • Flag missing permissions, missing repo access, or unclear ownership immediately with the exact blocker.
  • If required project fields cannot be set automatically, report the exact missing permission or field mapping and provide the follow-up command(s).