القائمة
Analyzes open issues, docs, and git history to recommend your next task. Interactive coaching flow.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
| name | whats-next |
| description | Analyzes open issues, docs, and git history to recommend your next task. Interactive coaching flow. |
| user-invocable | true |
| disable-model-invocation | true |
| allowed-tools | Bash, Read, Grep, Glob, AskUserQuestion, Agent |
Analyze open issues, project docs, and contribution history to recommend the user's next task. Interactive coaching flow.
Ask the user what they are looking for BEFORE fetching any data. Use AskUserQuestion with all four questions in a single call:
Q1 — Time availability:
Q2 — Work type:
Q3 — Work style:
Q4 — Area preference:
Store answers for Phase 2 filtering.
Run in parallel where possible:
gh issue list --state open --limit 30 --json number,title,labels,body,assignees,comments 2>/dev/null
If gh fails or returns nothing, fall back to docs scanning only. Do not stop or error out.
For each issue found, classify:
Look for project documentation that contains work items:
ls docs/ 2>/dev/null
If a docs/ directory exists, scan for files containing actionable items — look for TODO, FIXME, "future work", "next steps", "planned", "roadmap", or "open questions" sections. Read relevant docs and extract work items.
If no docs/ directory exists, skip gracefully.
git log --oneline -20
git log --all --oneline --since="2 weeks ago"
Understand what areas have been recently active to avoid recommending redundant work and to identify the user's recent focus areas.
git branch -a --list "origin/*" | grep -v main | grep -v master | grep -v HEAD
Flag branches that might represent in-progress or abandoned work.
Filter results based on Phase 1 answers. Show max 3 items per category. Omit empty categories entirely.
Tasks completable in under an hour. Small fixes, config tweaks, doc updates, closing known gaps.
| # | Source | Task | Area | Why it's quick |
|---|
Improvements that meaningfully enhance the project's reliability, capability, or quality.
| # | Source | Task | Area | Complexity | Impact |
|---|
Source column should reference the origin: GH #42, docs/roadmap.md, stale branch: feat/xyz, etc.
Security improvements, dependency updates, cleanup, monitoring gaps, tech debt.
| # | Source | Task | Area | Notes |
|---|
Based on git history, provide a brief coaching summary (4-6 sentences):
Before committing to a recommendation, dispatch to the engineering-lead subagent with the categorized candidates (Quick Wins / High Value / Maintenance) and the user's stated preferences from Phase 1. Ask it to pick the single highest-leverage next task and justify why it beats the runners-up, weighing value, readiness, and the user's context. Use its pick to inform the recommendation below — you may override it, but say why if you do.
Agent({
subagent_type: "engineering-lead",
description: "Pick the next task",
prompt: "Candidate work items: <Quick Wins, High Value, Maintenance lists with sources>. User preferences/context: <from the questionnaire>. Pick the ONE highest-leverage task to do next and justify why it beats the runners-up, weighing value, readiness, effort, and the user's stated context. Return the pick plus a one-line rationale for the top 2-3 alternatives."
})
Recommend ONE specific task. Match verbosity to complexity:
/start-work)End with: "Want to start? I can run /start-work to draft a plan."
If the user picks a task:
/start-work to create a branch and draft a plan."gh CLI fails, continue with docs-only mode — never stop or error outgh pr list if available)Research a work item, draft an implementation plan, and begin work after approval.
Guided one-shot install — discover installed plugins (or help install new ones), pick install scope, calibrate risk tolerance into a concrete allowlist + hooks, scaffold or merge CLAUDE.md, and sanity-check the result.
Finalize work — commit via /commit, push, and create PRs on feature branches.
Reviewer-side PR workflow — checks out the branch in a worktree, runs focus-area reviews plus a senior-engineering pass, optionally cross-checks against an autonomous reviewer, and posts inline findings only on explicit approval. Read-only — never edits, commits, or pushes.
Fetch PR review comments, classify by severity and confidence, and fix the selected subset.
Triage CVE and SBOM scanner output (Trivy, Grype, Snyk, Docker Scout, Dependabot) into a ranked, deduplicated action list.