plan-release-cycle
关于
This Claude Skill helps developers plan structured software release cycles, including milestones, feature freezes, and release criteria. It supports both time-based and feature-based release strategies, generating a comprehensive `RELEASE-PLAN.md` document. Use it when transitioning to structured releases, coordinating multi-team efforts, or planning major version launches like v1.0.0.
快速安装
Claude Code
推荐npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/plan-release-cycle在 Claude Code 中复制并粘贴此命令以安装该技能
技能文档
謀發布週期
謀結構化軟體發布週期:定策(曆基或功能基)、立里程碑與目標日、設功能凍結之準、管候選版、定發否清單、記回滾謀。生 RELEASE-PLAN.md 為文,導隊由開發至發布。
用時
- 始謀大版或小版之發
- 自隨機發轉結構化發布節奏
- 跨組件多隊協發
- 為合規項定質閘與發布之準
- 謀項目首公發(v1.0.0)
入
- 必要:目標版號(如 v2.0.0)
- 必要:所欲發布日或發布窗
- 必要:所謀功能或範圍列(待辦、路線圖、述)
- 可選:隊大小與可用
- 可選:發布策之偏(曆基或功能基)
- 可選:影響發布之合規所需
- 可選:往發布之速或週期長之數據
法
第一步:定發布策
於二主策間擇之:
曆基(時箱):
- 依固定時程發(如每 4 週、季)
- 未備之功能延至下發
- 對用者與下游項可預
- 宜:庫、框、含外消費者之具
功能基(範圍驅):
- 待定功能集畢而發
- 日依範圍而調
- 範圍蔓延與無限延誤之險
- 宜:內部具、首發、大重寫
多項宜混:定目標日與範圍,然許 1-2 週緩衝。若至緩衝期未滿範圍,延餘功能。
記策之擇與其因。
得:發布策已記,附與項脈絡相符之因。
敗則:若隊不能合於策,默擇曆基附功能優先列。時箱迫優先之決。
第二步:定里程碑
將發布週期分諸相,附目標日:
## Release Plan: v2.0.0
### Timeline
| Phase | Start | End | Duration | Description |
|---|---|---|---|---|
| Development | 2026-02-17 | 2026-03-14 | 4 weeks | Active feature development |
| Feature Freeze | 2026-03-15 | 2026-03-15 | 1 day | No new features merged after this date |
| Stabilization | 2026-03-15 | 2026-03-21 | 1 week | Bug fixes, documentation, testing only |
| RC1 | 2026-03-22 | 2026-03-22 | 1 day | First release candidate tagged |
| RC Testing | 2026-03-22 | 2026-03-28 | 1 week | Community/team testing of RC |
| RC2 (if needed) | 2026-03-29 | 2026-03-29 | 1 day | Second RC if critical issues found |
| Go/No-Go | 2026-03-31 | 2026-03-31 | 1 day | Final decision meeting |
| Release | 2026-04-01 | 2026-04-01 | 1 day | Tag, publish, announce |
典型相之長:
- 開發:總週之 50-70%
- 穩定:總週之 15-25%
- RC 試:總週之 10-20%
得:里程碑表,附諸相之日、長、述。
敗則:若時程過密(穩定 < 1 週),延發布日或減範圍。穩定不可略。
第三步:設功能凍結之準
定本發布之「功能凍結」何義:
### Feature Freeze Criteria
After feature freeze (2026-03-15):
- **Allowed**: Bug fixes, test additions, documentation updates, dependency security patches
- **Not allowed**: New features, API changes, refactoring, dependency upgrades (non-security)
- **Exception process**: Feature freeze exceptions require written justification and approval from [release owner]
### Feature Priority List
| Priority | Feature | Status | Owner | Notes |
|---|---|---|---|---|
| P0 (must) | New export format | In progress | [Name] | Blocks release |
| P0 (must) | Security audit fixes | Not started | [Name] | Compliance requirement |
| P1 (should) | Performance optimization | In progress | [Name] | Defer if not ready |
| P2 (nice) | Dark mode support | Not started | [Name] | Defer to v2.1.0 if needed |
P0 功能阻發。P1 功能備則含。P2 功能延而不誤。
得:功能凍結之規記之,附例外程序與分優之功能列。
敗則:若 P0 功能恐誤凍結日,立報之。選:延開發相、分功能為較小可交付者、或延至點發布(v2.0.1)。
第四步:謀候選版程序
定候選版如何生與試:
### Release Candidate Process
1. **RC1 Tag**: Tag from the stabilization branch after all P0 features merged and CI green
```bash
git tag -a v2.0.0-rc.1 -m "Release candidate 1 for v2.0.0"
-
RC Distribution: Publish RC to staging/testing channel
- R:
install.packages("pkg", repos = "https://staging.r-universe.dev/user") - Node.js:
npm install pkg@next - Internal: Deploy to staging environment
- R:
-
RC Testing Period: 5-7 business days
- Run full test suite including integration tests
- Verify all P0 features work as documented
- Test upgrade path from previous version
- Check for regressions in existing functionality
-
RC Evaluation:
- No critical/high bugs: Proceed to release
- Critical bugs found: Fix, tag RC2, restart testing period
- More than 2 RCs needed: Revisit scope and timeline
-
RC2+ Tags: Only if critical issues found in previous RC
git tag -a v2.0.0-rc.2 -m "Release candidate 2 for v2.0.0"
得:RC 程序記之,附標籤慣、分發法、試清單、升報之準。
敗則:若 RC 程序被略(發布壓),記其險。未試之發布有較高之回滾機率。
### 第五步:定發否清單
立發布許可前必達之準:
```markdown
### Go/No-Go Checklist
#### Must Pass (release blocked if any fail)
- [ ] All CI checks passing on release branch
- [ ] Zero critical bugs open against this version
- [ ] Zero high-severity security vulnerabilities
- [ ] All P0 features verified and documented
- [ ] Changelog complete and reviewed
- [ ] Upgrade path tested from previous version (v1.x -> v2.0.0)
- [ ] License and attribution files up to date
#### Should Pass (release proceeds with documented risk)
- [ ] Zero high bugs open (non-critical)
- [ ] All P1 features included
- [ ] Performance benchmarks within acceptable range
- [ ] Documentation reviewed and spell-checked
- [ ] External dependencies at latest stable versions
#### Decision
- **Go**: All "Must Pass" items checked, majority of "Should Pass" items checked
- **No-Go**: Any "Must Pass" item unchecked
- **Conditional Go**: All "Must Pass" checked, significant "Should Pass" items unchecked — document accepted risks
得:發否清單,附明確過敗之準與決規。
敗則:若發否會議致 No-Go,識阻項、指主、設新目標日(常 1-2 週後),並更新發布謀。
第六步:記回滾謀
定若發布致生產之關鍵問題如何回滾:
### Rollback Plan
#### Rollback Triggers
- Critical bug affecting >10% of users
- Data corruption or loss
- Security vulnerability introduced by the release
- Breaking change not documented in changelog
#### Rollback Procedure
1. **Revert package registry**: Unpublish or yank the release
- R/CRAN: Contact CRAN maintainers (cannot self-unpublish)
- npm: `npm unpublish [email protected]` (within 72 hours)
- GitHub: Mark release as pre-release, publish point fix
2. **Communicate**: Notify users via GitHub issue, mailing list, or social channels
- Template: "v2.0.0 has been rolled back due to [issue]. Please use v1.x.y until a fix is released."
3. **Fix forward**: Prefer a v2.0.1 patch release over a full rollback when possible
4. **Post-mortem**: Conduct a post-mortem within 48 hours of rollback to identify process gaps
#### Point Release Policy
- v2.0.1 for critical bug fixes within 1 week of release
- v2.0.2 for additional fixes within 2 weeks
- Patch releases do not require full RC cycle but must pass CI and critical test suite
完整發布謀書於 RELEASE-PLAN.md 或 RELEASE-PLAN-v2.0.0.md。
得:回滾謀記之,附觸發、程序、通信模板、點發布之策。完整 RELEASE-PLAN.md 已書。
敗則:若回滾不可行(如資料庫遷移已施),記前進修正之程序。每發布皆當有恢復之徑。
驗
- 發布策(曆/功能/混)已記附因
- 里程碑表含諸相附日:開發、凍結、穩定、RC、發布
- 功能凍結之準已定,附許/禁變之型
- 功能優先列已分類(P0 must / P1 should / P2 nice)
- RC 程序已記:標籤慣、分發、試期、升報
- 發否清單分「必過」「當過」二節
- 回滾謀含觸發、程序、通信模板
- RELEASE-PLAN.md(或等)文已立並存
- 時程實際(穩定至少佔總週之 15%)
陷
- 無穩定相:自開發直發。即三日穩定亦能捕活躍開發掩之問題。
- 凍結後範圍蔓延:凍結後許「再一功能」。每凍結後加入皆重置試並引退之險。
- 忽 P0 之險:P0 功能恐誤時不早報。範圍越早調,時程之擾越少。
- 「小」發布略 RC:即小發亦受益於至少一 RC。一日 RC 試廉於發後熱修。
- 無回滾謀:假發布必成。每發布謀當答「若敗如何」於發前。
- 曆壓越質:因諾於日而發,雖未過發否之準。延發布為小不便;壞發布為信之違。
參
apply-semantic-versioning— 定所謀發布之版號manage-changelog— 維與發布筆記相連之變更記plan-sprint— 發布週期之開發相內行衝刺謀draft-project-charter— 項目章可定發布路線圖與成功之準generate-status-report— 對里程碑追進度
GitHub 仓库
相关推荐技能
llamaguard
其他LlamaGuard是Meta推出的7-8B参数内容审核模型,专门用于过滤LLM的输入和输出内容。它能检测六大安全风险类别(暴力/仇恨、性内容、武器、违禁品、自残、犯罪计划),准确率达94-95%。开发者可通过HuggingFace、vLLM或Sagemaker快速部署,并能与NeMo Guardrails集成实现自动化安全防护。
cost-optimization
其他这个Claude Skill帮助开发者优化云成本,通过资源调整、标记策略和预留实例来降低AWS、Azure和GCP的开支。它适用于减少云支出、分析基础设施成本或实施成本治理策略的场景。关键功能包括提供成本可视化、资源规模调整指导和定价模型优化建议。
quantizing-models-bitsandbytes
其他这个Skill使用bitsandbytes库量化大语言模型,能在GPU内存有限时通过8位或4位量化减少50-75%内存占用,同时保持精度损失最小。它支持INT8、NF4、FP4等多种量化格式,可与HuggingFace Transformers无缝集成,适用于需要部署更大模型或加速推理的场景。还提供QLoRA训练和8位优化器支持,让开发者能轻松实现高效模型压缩。
dispatching-parallel-agents
其他该Skill用于并行处理3个以上无依赖关系的独立故障,可为每个问题域分派专属Claude代理同时执行调查修复。它通过并发处理多个独立问题显著提升故障排查效率,特别适用于测试文件、子系统等无共享状态的场景。
