把一次开发/协作过程沉淀成可复用的流程日志:回填项目进展、记录关键决策与转折点、保留紧凑历史,并在需要时提取成文章素材。
适合:项目历史复盘、关键决策留痕、bug 修复记录、复盘笔记、从开发过程提炼文章素材。
触发示例:"Backfill the process so far" / "Record this decision" / "把这个迭代转成复盘"
输出:time / context / problem / decision / change / result / lesson / article angle
零依赖,clone 后放进本地 skills 目录即可。英文正文见下文。
process-keeper is a lightweight skill for turning development work into a reusable process log.
It helps you:
- backfill what happened so far in a project
- record important decisions, fixes, and turning points
- keep a compact history for future review
- extract article-ready notes from development work
Use this skill when the work is about:
- project history
- key decisions
- bug fix records
- retrospective notes
- narrative material for posts or articles
Clone this repository and place it in your local skills directory, or point your agent to the folder if your setup supports custom skill paths.
There is no build step.
Do not use it for:
- feature implementation
- task management
- daily chat logs
- tiny edits that do not change the direction
- private project details in public output
Use a short instruction like:
- "Backfill the process so far"
- "Record this decision"
- "Turn this into a retrospective"
- "Keep a process log for this project"
- "Extract reusable notes from this iteration"
The default output should be compact and structured.
Preferred fields:
- time
- context
- problem
- decision
- change
- result
- lesson
- article angle
Short version:
- what happened
- why it changed
- what changed after
- how to move faster next time
- Find the active project root.
- Reconstruct the timeline from commits, files, and context.
- Keep only meaningful nodes.
- Write the process note in a compact format.
- Extract reusable material only when asked.
If the repo has no process notes yet, create docs/process/.
The supporting files in this repository are:
SKILL.mdPROCESS_GUIDE.mdreferences/process-template.mdreferences/examples.mdagents/openai.yaml
User: "We changed the result page structure. Record why."
Expected output: one compact process entry with the problem, decision, change, result, and lesson.
This skill is designed to work with local coding agents and repo-aware assistants.
It should be usable with:
- Codex-style local agents
- Claude-style local agents
- editor-integrated agents
- other repo-aware coding assistants
This skill keeps process notes lightweight.
It is not meant to:
- flood the repo with transcripts
- mix process notes with product code
- record every tiny edit
- expose private paths or internal-only notes in public output