| name | ai-collaboration-log |
| description | Organize and append AI collaboration records for coursework, reports, project retrospectives, Obsidian notes, and AI verification logs. Use when the user asks to record conversations with AI, summarize "my question + AI answer/action", create or update an AI校验记录 file, document how AI was used, capture iteration history across multiple agents/windows, or prepare evidence of AI-assisted learning and work. |
AI Collaboration Log
Maintain an incremental Obsidian-friendly log of user/AI collaboration. The log is not a raw transcript; it is a structured learning/work record that preserves the user's question, the inferred purpose, the AI's answer summary, and the AI's concrete actions/artifacts.
Core Workflow
-
Find or create the log
- If the user provides a target file, use it.
- If no target is provided, create an Obsidian Markdown note in the current vault/workspace.
- If the user's working language is Chinese, name the note
AI协作记录整理.md.
- If the user's working language is English, name the note
ai-collaboration-log.md.
- If a matching log already exists, append to it instead of creating a new file.
-
Preserve existing content
- Read the existing log before editing.
- Continue the existing heading style, numbering, tone, and person.
- Append new entries at the end unless the user explicitly requests insertion elsewhere.
- Do not rewrite or reorder prior entries unless asked.
-
Use one entry per Q&A unit
- Treat each user question plus the AI's answer/actions as one unit.
- If one user request produced several distinct work phases, split into subentries only when it improves readability.
- Keep records concise but concrete; include enough detail to prove how AI was used.
-
Include session metadata for multi-agent use
- At the top of each entry, include:
- Time
- Agent
- Model
- Conversation/window id or short label
- If exact metadata is unavailable, write
未提供 / unknown rather than inventing it.
- If the user provides a preferred label such as
Codex-1, ChatGPT网页, or 窗口A, use that label.
-
Record four required sections
目的推测: infer the user's goal behind the question.
用户提问原文: quote or lightly excerpt the user's original question.
AI回答摘要: summarize the useful answer, reasoning, recommendations, or corrections.
AI行为和产出文件: list concrete actions, file reads/writes, generated documents, modified paths, or verification steps.
Entry Template
Use this template for Chinese logs:
### N. 简短标题
**时间**:YYYY-MM-DD HH:mm(时区)
**Agent**:Codex / ChatGPT / 其他
**模型**:GPT-5 / 未提供
**对话窗口**:窗口A / thread-id / 未提供
**目的推测**:
用户希望……
**用户提问原文**:
“……”
**AI回答摘要**:
AI 主要说明……
**AI行为和产出文件**:
- 读取/分析了……
- 新建/修改了……
- 将结果同步到……
Use this template for English logs:
### N. Short title
**Time**: YYYY-MM-DD HH:mm (timezone)
**Agent**: Codex / ChatGPT / other
**Model**: GPT-5 / unknown
**Conversation**: Window A / thread-id / unknown
**Inferred purpose**:
The user wanted to...
**Original user prompt**:
"..."
**AI answer summary**:
The AI explained...
**AI actions and artifacts**:
- Read/analyzed...
- Created/modified...
- Synced results to...
Writing Rules
- Prefer first-person learning-retrospective style when the existing log uses it, e.g.
我要求 AI..., AI 帮我....
- Preserve key user wording in quotes, but do not paste entire long conversations unless requested.
- Summarize AI answers; do not dump full generated documents into the log.
- Include paths for important artifacts, especially files created or modified.
- Emphasize AI's role beyond answering: topic clarification, task decomposition, drafting, critique, risk review, misconception correction, final polishing, or record keeping.
- For coursework logs, explicitly connect the AI interaction to the assignment requirement or learning outcome.
- For multiple agents/windows, make entries distinguishable through metadata and avoid claiming one agent did work done by another.
Duplicate Avoidance
Before appending:
- Check the last few entries.
- If the same Q&A is already logged, update the existing entry only when the user requests correction; otherwise do not duplicate.
- If the new entry extends a previous topic, create a new numbered entry and mention the relationship.
File Editing Guidance
- Use
apply_patch for Markdown edits where possible.
- For files outside the writable workspace, request required approval or use an already approved command path if available.
- When updating a Git-tracked report before commit, follow any repository secret-scanning requirements from AGENTS.md or project instructions.
Output to User
After updating the log, report:
- Which log file was updated or created.
- Which entry number/title was added.
- Any related files created or modified.
- Whether verification was performed.