artifacts
为用户“任务请求”创建和管理本地仓库工作成果,通过生成`.artifacts/artifacts/{request_id}/`目录下的`plan.md`、`todo.md`和`walkthrough.md`,然后将完成的工作索引到存储在`.artifacts/chroma_db/`的ChromaDB中,以支持7天回溯。当用户需要(1)跟踪/确认后执行的工作流程,(2)链接到工作流程的最简最终答案,或(3)询问“我们之前如何使用工作成果技能?”时应通过查询ChromaDB工作成果索引来回答。
📖 中文导读:为用户“任务请求”创建和管理本地仓库工作成果,通过生成`.artifacts/artifacts/{request_id}/`目录下的`plan.md`、`todo.md`和`walkthrough.md`,然后将完成的工作索引到存储在`.artifacts/chroma_db/`的ChromaDB中,以支持7天回溯。当用户需要(1)跟踪/确认后执行的工作流程,(2)链接到工作流程的最简最终答案,或(3)询问“我们之前如何使用工作成果技能?”时应通过查询ChromaDB工作成果索引来回答。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。
name: artifacts
description: Create and manage repo-local work artifacts for a user “task request” by generating .artifacts/artifacts/{request_id}/ with plan.md, todo.md, and walkthrough.md, then indexing completed work into ChromaDB stored at .artifacts/chroma_db/ for 7-day recall. Use when the user wants (1) a tracked/confirm-before-execute workflow, (2) a minimal final answer that links to a walkthrough, or (3) to ask “what did we do before with the artifacts skill?” and you should answer by querying the ChromaDB artifacts index.
Artifacts
Workflow
Goal: leave a durable, linkable trace of work in .artifacts/artifacts/{request_id}/, and answer future questions by querying the .artifacts/ ChromaDB index (retained for 7 days).
0) Bootstrap the required environment under .artifacts/ (always)
- Create
.artifacts/.venv/and installchromadbthere:python3 "$CODEX_HOME/skills/artifacts/scripts/artifacts_bootstrap_env.py"
- After bootstrap, run all other commands using
.artifacts/.venv/bin/python(required).
1) Initialize artifacts (always)
- Run the initializer script (from repo root or anywhere inside the repo):
.artifacts/.venv/bin/python "$CODEX_HOME/skills/artifacts/scripts/artifacts_init.py" --request "<user request>"
- The script prints paths for:
.artifacts/artifacts/{request_id}/plan.md.artifacts/artifacts/{request_id}/todo.md.artifacts/artifacts/{request_id}/walkthrough.md
- Reply to the user with:
- The
plan.md+todo.mdpaths - A single confirmation question: “이대로 진행할까?”
- The
- Do not start implementation until the user explicitly confirms.
2) Execute work (after user confirmation)
- Update
.artifacts/artifacts/{request_id}/todo.mdby checking items as they complete. - Do the requested work in the repo (code changes / analysis / commands).
- Write
.artifacts/artifacts/{request_id}/walkthrough.mdas the detailed record (plan snapshot, steps, commands, files, validation). Ensure it is sufficient for a reviewer to reconstruct the work without the chat.
3) Index + retention (end of task)
- Run indexing (this also prunes anything older than 7 days, both index + folders):
.artifacts/.venv/bin/python "$CODEX_HOME/skills/artifacts/scripts/artifacts_index.py"
- The index is stored at
.artifacts/chroma_db/(repo-local, derived). - If
.artifacts/chroma_db/is missing or chromadb is not installed, initialize by running the indexer once with a Python environment located under.artifacts/(required).
4) Answering “what did we do before?” (query mode)
- Query the artifacts index:
.artifacts/.venv/bin/python "$CODEX_HOME/skills/artifacts/scripts/artifacts_query.py" "<question>"
- Use the returned results as the basis for your answer.
- In your reply, include the
todo.md+walkthrough.mdpaths for the most relevant result(s).
Output rules (important)
When responding after work is done, keep the answer minimal:
- Include clickable paths to:
.artifacts/artifacts/{request_id}/todo.md.artifacts/artifacts/{request_id}/walkthrough.md
- Keep prose to 1–3 short sentences: what instruction you received + what you did.
- Do not paste the walkthrough into the chat; the walkthrough is the source of detail.
Reference
- File format + retention details:
references/artifacts_format.md