next
Decide the single next best thing to work on, with two alternates. Reads setup progress, work in progress, open PRs and failing CI, the backlog (GitHub Issues or Linear), overdue harness rituals, product decisions parked until now, and north-star analytics if PostHog is connected. Use when someone asks "what next?", "what should I work on?", at the start of a session, or after a PR merges.
- License 许可
- MIT
- 最近更新
- 2026-10-06
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- next
- description
- Decide the single next best thing to work on, with two alternates. Reads setup progress, work in progress, open PRs and failing CI, the backlog (GitHub Issues or Linear), overdue harness rituals, product decisions parked until now, and north-star analytics if PostHog is connected. Use when someone asks "what next?", "what should I work on?", at the start of a session, or after a PR merges.
02 资源文件 1
- 脚本signals.py10.6 KB
03 SKILL.md 正文
Next
The owner shouldn't need to know what to ask for. This skill looks at everything and proposes one action, plus two alternates, each with a one-line reason. It picks; it doesn't start. The owner says "go" (or picks an alternate), and then you run the skill it names.
1. Gather signals (deterministic first)
python3 .agents/skills/next/signals.py --remote
That returns JSON with setup progress, the current branch and work in progress, open
PRs with failing checks, failing CI on main, open issues, overdue rituals, and
decisions_due: product calls parked on day 0 in design/brief.json → decisions
whose phase has now arrived (the first feature, pre-launch, post-launch), and
stale_flags: feature flags past their expires date (scripts/check_flags.py). It also returns
risk_open: the items the risk gate (scripts/risk_gate.py, the same check ship runs)
says are open, such as an unanswered risk question, an accepted risk with no owner or
date, or a screen that touches a category the idea was never screened for. If
gh_available is false, say so in one line and carry on with local signals.
Then add what the script can't read:
- Linear (
tracker: linearinappbox.yaml): Linear MCPlist_issuesfor the team, status Todo/In Progress, top 20 by priority. - North star (if the PostHog MCP is connected): the funnel of the 5 events in
docs/product/BRIEF.mdover the last 7 days. Note the step with the biggest drop-off. If the MCP isn't connected, skip this and say "analytics not read". - In-flight specs:
ls docs/product/specs/for any spec whose ticket has no merged PR yet.
In Claude Code you may hand the CI-log and dependency reading to the triage
subagent (Haiku, cheap); in Codex, read them inline.
2. Rank (first match wins the top slot)
| Rank | Signal | Proposed action |
|---|---|---|
| 1 | Setup phase not done | Resume new-app at that phase |
| 2 | main CI failing |
Fix it: backlog to file a fix ticket, then fix on a branch |
| 3 | Your open PR has failing checks or requested changes | land it: fixes CI and threads, then re-reviews |
| 4 | Uncommitted or unpushed work on a feature branch | Finish it: gates, then PR |
| 5 | Open PR with green CI, not yet reviewed | land it (runs pr-review, then merges or asks you) |
| 6 | Ritual overdue by 2× its cadence | Run that ritual (reflect, harness-optimize, north-star-report, market-watch) |
| 7 | Security or major dependency PR | triage it, then pr-review |
| 8 | A p0 ticket, or the ticket at the funnel's worst step |
feature-discovery on it (or build-feature if it's specced) |
| 9 | A risk item is open (risk_open) |
Ask the open questions as one structured round and write each answer back to design/brief.json → risk; a new category runs python3 scripts/risk_gate.py rescreen first. ship refuses until they're answered or accepted |
| 10 | A decision is due (decisions_due) |
Ask it as one structured question, the docs/DEFAULTS.md answer recommended, and write the answer back to its ledger entry; pre-launch ones go to ship |
| 10b | A flag is past its expiry (stale_flags) |
Remove it (both registries and every flag()/useFlag() call, then delete it in PostHog), or extend expires if its owner says it must stay: a backlog chore ticket, then a branch |
| 11 | Next ticket by priority (p0 > p1 > p2, ready first) |
feature-discovery / build-feature |
| 12 | Empty backlog | north-star-report to find the biggest drop-off, then file tickets |
Tie-break: the thing closest to users (a broken build beats a new feature), then the cheapest to finish. Never propose starting new work while one of your PRs is red.
3. Answer (exactly this shape, under 12 lines)
The Next: line always names the skill in parentheses. An action without its skill
leaves the owner asking "how?".
Exactly two alternates. Nothing goes after the Signals read line except the
question: no "after that…" or "also pending" list, because a fourth option dilutes the pick.
Next: <action> (<skill to run>, e.g. `/feature-discovery #14` or `$feature-discovery #14`)
Why: <the signal, with its number: PR #, run, ticket, funnel step and %>
Or:
2. <alternate>: <one-line why>
3. <alternate>: <one-line why>
Signals read: setup ✓ · PRs ✓ · CI ✓ · backlog ✓ · rituals ✓ · decisions ✓ · flags ✓ · risk ✓ · analytics ✗ (PostHog not connected)
Use a structured question (AskUserQuestion in Claude Code, request_user_input in
Codex) with the three options if the tool exists, otherwise ask in chat. On "go", run
the named skill. If you find a bug or scope while gathering signals, file it with
backlog before answering. Don't keep a to-do list in chat.
Ask the owner
Follow .agents/rules/product-judgement.md: ask with a structured question (recommended option first), never decide these silently. In this skill that means: When two actions tie on rank, or the top one is a product call (a scope cut, launching, pricing), present the pick and alternates as one structured question instead of starting.
04 同仓库更多 Skills
ship
Release checklist for getting a build to testers or the stores. Runs the risk gate first (refuses while a risk item is open) and walks the owner through the compliance checklist. Covers the version bump, EAS build and submit, OTA updates and when they're safe, changelog, store listing, privacy nutrition label / data safety form, the Accessibility Nutrition Labels, and the account-deletion requirement. Guidance-level; it never handles signing secrets. Use when asked to ship, release, submit to TestFlight / App Store / Play, or push an OTA update.
routines
Put the project's recurring rituals on a schedule, with the owner's consent. Weekly north-star report, monthly market watch, dependency triage, Supabase advisor sweep, reflect and harness-optimize. Uses Claude Code Routines (cloud cron) where they fit, and documents the Codex / local cron equivalent. Use when asked to automate, schedule or "keep the project moving on its own", or after setup finishes.
reflect
Consolidate what recent sessions learned into the project's long-term memory (.agents/memory). Use at the end of a substantial session, after a correction from the user, after a painful debugging session, when the session-start healthcheck says reflect is overdue or a pending-reflection seed is waiting, or when MEMORY.md exceeds its budget.
pr-review
Review a pull request on the owner's behalf and fix what's safe to fix. Runs the deterministic gates first, then correctness/security, design-system, accessibility and analytics review passes, verifies each blocking finding, applies safe fixes, re-checks, and ends with a plain-English verdict. Use when a PR is opened, when asked to review a PR, a branch or a pasted diff, or before merging anything.
north-star-report
Weekly product-analytics and reliability readout. Pulls the funnel of the brief's 5 north-star events from PostHog, finds the biggest drop-offs and errors, checks each SLO in docs/slo.yaml against its error budget, compares with last week, and proposes (then, with a yes, files) tickets that target them. Use weekly, when the session-start line says it's overdue, when the backlog is empty, or when asked "how is the app doing?".
new-worktree
Create an isolated git worktree + branch off origin/main for one unit of work, so parallel agents or sessions never share a checkout. Use when starting any ticket, or when another session may be working in the main checkout.