checkpoint
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
Chinesecheckpoint
checkpoint
Write-side twin of . Captures what this session taught the user, promoting each item as far up the ladder as it can go: code first, repo artifacts second, the personal store last.
lets-worklets-workPersonal state (zero-trace contract)
个人状态(零痕迹约定)
Same contract as — keep the two in sync:
lets-work- Root:
${XDG_DATA_HOME:-$HOME/.local/share}/identity16-skills/projects/<repo-id>/ - : normalized
repo-idURL slug (strip protocol/credentials/origin;.gitand/→:); no remote → absolute path slugged the same way.__ - Files: (last-visit marker),
state.json(mental model; fallback, never a mirror),dossier.md(append-only session log).journal.md
Never write any of this — or any reference to this skill pack — into the target project's files, commits, PRs, or issues.
The ritual's weight must be proportional to what the session taught. A routine session that produced nothing worth keeping gets no interview: update , confirm in one line, done. The full routing below is for sessions that actually moved the user's understanding.
state.json与 遵循相同的约定——保持两者同步:
lets-work- 根目录:
${XDG_DATA_HOME:-$HOME/.local/share}/identity16-skills/projects/<repo-id>/ - :规范化后的
repo-idURL slug(去除协议和凭据以及origin;将.git和/转换为:);没有远程仓库时,使用以相同方式转换后的绝对路径。__ - 文件:(上次访问标记)、
state.json(心智模型;仅作为兜底,绝不作为镜像)、dossier.md(仅追加的会话日志)。journal.md
绝不要将这些内容——或任何对此 skill 包的引用——写入目标项目的文件、commit、PR 或 issue。
收尾流程的详尽程度必须与本次会话带来的认知相匹配。 如果例行会话没有产生值得保留的内容,则无需访谈:更新 ,用一行确认,然后结束。只有在会话确实推进了用户的理解时,才需要执行下面的完整分流流程。
state.jsonSteps
步骤
- Collect candidates: walk back through the session and list what was learned — decisions made, gotchas hit, invariants discovered, assumptions refuted, questions still open, WIP state.
- Route each item up the ladder, proposing promotions rather than silently applying them:
- Code: a non-obvious constraint or "why" the code cannot express → propose a concise comment at the load-bearing spot, following the target project's comment norms.
- Repo artifacts: the rationale for the approach taken, alternatives rejected → propose adding it to the PR description or the commit message being shipped.
- Personal journal: project process facts (where discussion happens, how to run things locally), the user's open questions, WIP notes — anything with no home in the repo → append to under today's date.
journal.md
- Update with deltas to the mental model: new or changed invariants, corrected beliefs. Delete entries the codebase now makes obvious — a shrinking dossier means the codebase is improving.
dossier.md - Write with the current commit, date, and branch.
state.json - Confirm to the user what went where.
- 收集候选项:回顾本次会话,列出所获得的信息——做出的决策、遇到的陷阱、发现的约束条件、推翻的假设、仍未解决的问题以及 WIP 状态。
- 将每一项分流到阶梯中合适的位置,提出晋升建议,而不是静默应用:
- 代码:代码无法表达的隐晦约束条件或“为什么”——建议在关键位置添加简洁注释,并遵循目标项目的注释规范。
- 仓库产物:所采用方案背后的理由以及被否决的替代方案——建议将其添加到正在提交的 PR 描述或 commit 信息中。
- 个人日志:项目流程方面的信息(讨论发生在哪里、如何在本地运行)、用户的开放问题和 WIP 笔记——所有在仓库中没有归宿的内容——追加到 中今天的日期下。
journal.md
- 使用心智模型的变化更新 :新增或修改约束条件,修正已有认知。删除代码库现在已经能够直接说明的条目——dossier 不断缩减意味着代码库正在改善。
dossier.md - 将当前 commit、日期和分支写入 。
state.json - 向用户确认各项内容分别被分流到了哪里。
Guardrails
防护规则
- Promotions into the repo are normal work product (comments, PR text, commit messages) — but they must never mention this skill pack, the personal store, or its paths.
- Do not journal what the repo already records (code structure, git history, existing docs).
- 将内容晋升到仓库中是正常的工作产物(注释、PR 文本、commit 信息)——但绝不能提及此 skill 包、个人存储或其路径。
- 不要在日志中记录仓库已经包含的信息(代码结构、git 历史、现有文档)。