next-steps
Identify remaining completion requirements, meaningful learning to review, and ready successor work.
- License 许可
- MIT
- 所属仓库
- BetoFrega/skills
- 最近更新
- 2026-11-14
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- next-steps
- description
- Identify remaining completion requirements, meaningful learning to review, and ready successor work.
02 资源文件 1
- 参考文档agents/openai.yaml163 B
03 SKILL.md 正文
Next Steps
Start from the latest goal, plan, results, and user corrections. Read the relevant work records, canonical references, and configured completion and delivery rules, and verify the actual state of artifacts. Apply artifact links to every cited document or ticket.
The user may be switching tasks, so make each consequential next step understandable on its own: name the project or topic when ambiguous, the components or domain entities involved, the action or pending choice, and why it matters or what it unlocks. Explain references such as a phase, contract, or issue instead of relying on their names. Reuse context that is already clear. When a detail is missing, inspect the sources or name the gap, and skip a recap of the whole task.
Current work incomplete
List the required actions in execution order, immediate action first, with blockers and decisions next to the actions they affect. Take the requirements from the goal, acceptance, and project workflow: evidence, a rehearsed demo script (demo-day) and the user's review of interaction changes, canonical updates, checks, review, integration, deployment, operational verification, and tracker closure, each only where required. Discovery work is complete when it produces its expected evidence or output, even without a feature.
Each required condition needs verified evidence before you call the work complete,
per Evidence Before Claims. Apply
local document delivery to
separate saved documentation from pending origin/main integration. For product
observations, use consolidate to read implementation,
deployment, authorized release, exposure, and flags as separate dimensions.
Current work complete
Recommend review-learnings when results materially contradict hypotheses, change scope, direction, or rules, or affect related or pending work, taking configured review points into account. Keep required knowledge maintenance apart from learning review. Skip a review per item, and skip reviewing learning already incorporated unless there is new evidence or a review point.
Inspect follow-ups, the selected increment or spec references, parent and dependency relations, remaining work, readiness, and documented priority. Recommend one open successor whose prerequisites you verified, with its link and title or identifier and its value toward the goal. Resolve consequential findings that invalidate its scope or readiness before it starts. A future idea is not yet selected or ready, and having no known blockers does not make work ready. Without tickets, the successor needs support from the goal and verified relationships. Leave out unsupported successors, report unavailable sources, and keep current requirements apart from later work.
Invocation boundary
An explicit invocation only reports: return the guidance and end. During execution, finish the remaining authorized work before the final report, or name its blocker. Use execution skills for actions and consolidate for canonical records. A recommendation grants no new permission for work, decisions, publication, release, or scheduling, and existing authorization still applies. If no goal is known, ask which task the user means.
04 同仓库更多 Skills
writing-for-humans
Load before writing any reader-facing text, including messages, and follow its rules as you write. Also use to rewrite or audit existing prose for plain language, clarity, readability, or actionability.
writing-for-agents
Write agent-facing documents. Use when creating or editing skills, AGENTS.md, or CLAUDE.md.
write-adr
Write, revise, or supersede architecture decision records (ADRs) documenting an architectural choice and its rationale.
wizard
Wizard scripts for human-only provisioning, credential/CI setup, unfamiliar dashboards, migrations, or cutovers; use only for steps the agent cannot perform.
verify
Check whether the current task's claimed outcome is confirmed.
verb-tense
Choose verb tense and mood by when each event happens and how certain it is. Use when writing or reviewing any text, and when a reader asks for a tense review.