v0.1.10 — SATA 2.1 and leaner guidance

v0.1.10 — SATA 2.1 and leaner guidance

Changes

• Ship the SATA 2.1 license in a versioned release; v0.1.9 packages still carry SATA 2.0. Fix the README license badges.
• Mention Less is more in the agent interface metadata.
• State shared evidence principles once in the entrypoint and remove their repetitions from references. Condense the entrypoint's Less is more rule; details stay in its reference.
• Move maintainer guidance out of the runtime entrypoint into the README.
• Mark claims about why models over-elaborate as a plausible mechanism rather than established fact.

更新内容

• 以正式版本发布 SATA 2.1 许可证（v0.1.9 安装包仍为 SATA 2.0），修正 README 许可证徽章。
• 智能体界面元数据补充 Less is more 模式。
• 共享的证据原则只在入口说明一次，删除参考文档中的重复表述；精简入口中的 Less is more 规则，细则保留在其参考文档。
• 维护者指引从运行时入口移到 README。
• 关于模型为何过度扩写的说法改为合理推测，不再当作既定事实。

Validation

Checked skill structure, UTF-8 encoding, line endings, local links and anchors,
paired README guidance, license consistency, and the installable ZIP against the
committed skill files. These are documentation and packaging checks, not a model
behavior benchmark.

v0.1.9 — Less is more for frontend copy

v0.1.9 — Less is more for frontend copy

Changes

• Add Less is more guidance for reducing redundant interface text while preserving clear actions, essential information, and accessibility.
• Include before/after examples, state-specific copy guidance, and research sources that distinguish measured verbosity bias from hypotheses about generated interfaces.
• Update English and Chinese usage instructions with the new frontend mode.

更新内容

• 新增 Less is more 前端指令，减少冗余界面文字，保留清楚的操作、必要信息和可访问性。
• 补充精简前后示例、不同界面状态的文案原则与研究来源，区分模型冗长偏差的实验证据和对生成界面的推测。
• 更新中英文使用说明，加入前端模式示例。

Validation

Checked skill structure, UTF-8 encoding, local links and anchors, paired README
guidance, license consistency, and the installable ZIP against the committed skill
files. These are documentation and packaging checks, not a model behavior benchmark.

v0.1.8 — clearer simplification decisions

v0.1.8 — clearer simplification decisions

Changes

• Clarify failure ownership and recovery boundaries while preserving security, data integrity, and cleanup guarantees.
• Judge abstraction and change scope by shared contracts and required outcomes, including necessary changes across callers, types, tests, and documentation.
• Make preparatory refactoring, parallel migration, characterization tests, and bounded experiments conditional on the actual dependencies and risks.
• Distinguish compatibility obligations from behavior the user has authorized changing, so simplification preserves required behavior without freezing a requested bug fix.
• Add examples of unnecessary agent-generated machinery and a reference bibliography. Consolidate repeated entrypoint guidance and clarify English and Chinese documentation.

更新内容

• 明确失败归属和恢复边界，保留安全、数据完整性及资源清理保证。
• 按共享契约和完整需求判断抽象与修改范围，保留调用方、类型、测试和文档中的必要配套修改。
• 根据实际依赖与风险选择准备性重构、分步迁移、特征测试和有限试验，避免把方法变成必经流程。
• 区分兼容性义务与用户已授权改变的行为，防止“保留行为”阻碍修复已确认的错误。
• 补充智能体过度实现的例子和参考来源，合并入口中的重复指引，并改善中英文说明。

Validation

Checked skill structure, local links and anchors, paired README links, license
consistency, and the installable package against the committed skill files.

v0.1.7 — leaner skill instructions

v0.1.7 — leaner skill instructions

Changes

• Narrow automatic selection to complexity reviews, scoped cleanup, and relevant implementation choices. Shorten the entrypoint and load detailed guidance only for the current question.
• Replace the numbered audit sequence with conditional guidance. Continue authorized implementation through relevant checks and repairs without another approval of intermediate findings.
• Preserve contract, security, compatibility, and lifecycle requirements. Update both READMEs with the OpenAI article behind this revision.

更新内容

• 收窄自动触发范围，聚焦复杂度审查、指定范围的清理及相关实现选择。精简入口，按当前问题读取详细指引。
• 将编号审查流程改为按需参考；已授权的实现持续推进到相关验证和修复完成，中间发现无需再次审批。
• 保留契约、安全、兼容性与生命周期要求，中英文 README 补充本次调整依据的 OpenAI 文章。

Validation

Skill structure, local links and anchors, paired README references, and package
contents were checked. No model behavior benchmark was run.

v0.1.6 - focused simplification guidance

v0.1.6 — focused simplification guidance

Changes

• Define observable success criteria before editing, and preserve project conventions and readable control flow.
• Add examples for behavior-changing shorthand, repeated I/O, and unnecessary update notifications. Recheck stale review findings instead of assuming another edit resolved them.
• Clarify public export removal, contract tests, and when existing CI results cover a change.
• Include the license in the installable skill directory and add whole-directory upgrade instructions. Record version 0.1.6 in skill metadata.
• Document related tools and adopted ideas in both READMEs. Use the existing SATA 2.0 (modified) license text and naming consistently.

更新内容

• 动手前明确可观察的成功条件，遵循项目约定，保持控制流易读。
• 补充简写改变行为、重复 I/O 和无变化更新通知的例子；重新核对过时审查结论。
• 明确移除公开导出、编写契约测试，以及复用 CI 结果的判断条件。
• 技能目录随附许可证，增加整体替换目录的升级说明，并在技能元数据中记录 0.1.6。
• 中英文 README 补充同类工具与借鉴；统一使用现有 SATA 2.0（本项目修改版）文本和名称。

Validation

Checked the release commit archive and the extracted
no-stop-gpt-v0.1.6.zip: all eight packaged files match the committed sources; skill
structure, UI metadata, license consistency, 31 repository Markdown links, two
README image paths and 21 isolated package links pass. Paired README source links
match. No new model behavior benchmark was run.

v0.1.5 - consolidate cleanup guidance

Changes

• Clarify Prevent, Audit, and Sweep scope: authorized cleanup continues without repeated approval of intermediate findings.
• Use contract and consumer evidence first. Treat ablation as an optional experiment and caller counts as design clues rather than deletion rules.
• Consolidate investigation, execution, decision-record cleanup, and imported findings into one Sweep reference. Reference files decrease from nine to five.
• Shorten both READMEs and remove historical validation summaries that did not cover the revised skill.

Validation

Checked the exact release commit archive: skill structure, UI metadata, and all 32 local links and anchors pass. No new agent behavior benchmark was run.

Changes since v0.1.4

count entities, not lines

What's changed

• SKILL.md gains two instruments that run through every mode: Occam's razor (count entities, not lines; the razor cuts entities, never requirements) and ablation (a suite that stays green because it never reached the mechanism is silence, not evidence)
• README.zh-CN: complete the truncated license section, align trigger words and typography with the English README
• references: unify H1 heading case across all nine files, clarify the "no safe candidate" line, drop a restated sentence in the audit checklist

v0.1.1 — fail at the unit that owns the work

Patch release: absorb Erlang's let-it-crash into the failure contract.

• Fail at the unit that owns the work: a request fails the request, a job fails the job, a process-level invariant fails the process
• Expected errors stay at the owned boundary (4xx, typed errors). Unexpected errors leave the dirty unit uncaught; recovery belongs to an outer owner with known-good state
• Do not scaffold a supervisor tree to make throwing safe — use the isolation the platform already provides
• New antipattern: Wrong failure radius (request-scoped process.exit, catch-and-patch in the dirty helper, HTTP 200 + { ok: false })
• Audit now names the unit that should fail

v0.1.0 — No, Stop! GPT!

First public release.

• One skill, three modes: Prevent (writing-time discipline), Audit (keep / remove / downgrade verdicts, read-only), Sweep (repo-scale evidence-backed simplification)
• Five decision tests, hard exceptions, and a domain portrait for correctness-critical systems
• Install: npx skills add klarkxy/no-stop-gpt, or copy skills/no-stop-gpt/ into your agent's skills directory
• Now with a mascot: ChatGPT-chan caught mid-over-engineering (assets/mascot.png)