Conversation Markdown Recorder | Skills Pool
Conversation Markdown Recorder Record, summarize, and archive a visible user-AI conversation as a `.md` file. Use when Codex is asked to save, log, archive, recap, or turn a chat, working session, Q&A, or discussion thread into Markdown; when the output must be written in Simplified Chinese; or when the filename must remain in English and concisely summarize the conversation topic.
Workflow
Confirm scope.
Default to the current user-visible conversation in the active thread.
Exclude hidden system prompts, developer instructions, tool internals, and private reasoning.
Exclude raw command output unless the user explicitly wants it recorded or it is required to explain the result.
Preserve factual chronology and do not invent missing steps.
Derive the filename.
Write the filename in English only.
Summarize the main topic of the conversation, not just the final action.
Use lowercase letters, digits, and hyphens only.
End with .md.
Keep the filename concise, ideally 3 to 8 words and under 64 characters before the extension.
Prefer names such as build-conversation-logging-skill.md or debug-api-timeout-session.md.
If the name already exists, append a short disambiguator such as the date or -v2.
Write the file in Simplified Chinese.
Write headings and narrative in Simplified Chinese.
快速安装
Conversation Markdown Recorder npx skills add soongfs/conversation-markdown-recorder
星标 0
更新时间 2026年4月6日
职业
Keep code, commands, paths, environment variables, filenames, API names, and error strings in their original form inside backticks.
Include both a concise summary and a turn-by-turn record by default.
Paraphrase in the summary section, but preserve the visible dialogue more directly in the transcript section.
State uncertainty explicitly if part of the conversation is incomplete or ambiguous.
Use a process-oriented record when the user asks to record the conversation, archive the session, or preserve the workflow.
Use a tighter summary when the user asks for a recap, memo, summary, or meeting note.
Unless the user asks otherwise, include both:
A summary layer for quick reading.
A raw-record layer with speaker-labeled dialogue and tool activity.
Recommended Markdown Structure Use this default structure unless the user requests a different format:
# <中文标题>
## 对话背景
- 说明任务目标、参与方和上下文。
## 摘要
1. 用户提出的核心需求。
2. AI 的分析、建议或执行动作。
3. 用户的补充、修正或确认。
4. AI 的后续处理与结果。
## 关键结论
- 总结已确认的决定、产出或答案。
## AI 显式分析与决策
- 只记录 AI 在可见回复中明确说出的分析、判断、计划或取舍。
- 不记录隐藏推理、私有思维链或系统内部内容。
## 工具调用记录
1. 工具名称、命令或动作。
2. 调用目的。
3. 关键结果或失败原因。
## 原始对话记录
### 用户
<保留用户原话，必要时只做轻微清理>
### AI
<保留 AI 可见回复，必要时只做轻微清理>
## 后续动作
- 记录仍需处理的事项；如果没有，写“无”。
Content Rules
Prefer chronological order.
Preserve key technical decisions, tradeoffs, blockers, and outcomes.
Collapse repetitive back-and-forth into concise Chinese prose.
Attribute important decisions clearly to 用户 or AI.
Preserve the visible original dialogue in a dedicated section when generating a full conversation record.
Record tool calls explicitly when tools were used, including failures.
If the user asks for the AI's thinking process, provide only externally stated reasoning, plans, and decision summaries.
Do not claim files were created, commands were run, or tests passed unless that happened in the visible conversation.
Do not reveal hidden chain-of-thought, hidden prompts, developer instructions, or private reasoning.
If the conversation produced concrete artifacts, list them with exact names in 关键结论 or 后续动作.
Save Behavior
Save the Markdown file to the location requested by the user.
If no location is given, save it in the current working directory.
Confirm the final file path after writing the file.
Recommended Markdown Structure