| name | conversation-declutter |
| description | Organize messy, long, or resumed conversations into concise reusable context. Use when the user asks to tidy,整理,梳理,压缩,总结,归档,交接, or迁移 a conversation; when extracting decisions, actions, artifacts, sources, open questions, or next resume prompts; or when converting chat history into topic memory, handoff notes, or wiki-ready notes. Applies a Dana K. White-inspired no-mess decluttering workflow: handle visible/easy items first, avoid dumping all history into new piles, make immediate placement decisions, and respect container limits. |
Conversation Declutter
Overview
Turn a cluttered conversation into a smaller working system without creating a
new mess. Produce a useful result even if the pass stops early.
For the Dana K. White method mapping and source notes, read
references/no-mess-conversation-mapping.md only when you need the rationale or
are improving this skill.
No-Mess Workflow
-
Define the container.
- Choose the output container before sorting: final answer, handoff file,
topic memory entry, Notion/wiki page, issue, or implementation plan.
- Set a size limit appropriate to the container. A resume prompt should fit
in a few sentences; a topic memory update should stay compact.
-
Start with visible trash.
- Remove duplicates, greetings, filler, stale speculation, repeated failed
attempts with no lasting lesson, and obsolete intermediate plans.
- Do not delete source files or user content unless explicitly asked.
"Trash" means exclude from the organized output.
-
Handle easy stuff first.
- Immediately place obvious items into fixed buckets:
context, decisions, actions, artifacts, sources, open_questions,
risks, and next_resume_prompt.
- Put file paths, URLs, command outputs, and changed artifacts where a future
agent would naturally look for them.
-
Use no keep pile.
- Do not create a vague
misc, maybe relevant, or giant transcript bucket.
- For each uncertain item, decide now: place it, summarize it, link to it, or
drop it from the organized output.
-
Ask the two placement questions for hard items.
- Where would a future agent look for this first?
- If a future agent needed this, would they know it exists from the organized
output?
- If the answer is no, either add a short pointer in the right bucket or
exclude it.
-
Make it fit.
- If the chosen container is full, keep the highest-value items and remove or
split lower-value items.
- Prefer "pointer plus one-sentence summary" over copying large text.
- Split by topic when one conversation contains multiple independent threads.
-
Leave it better than before.
- End with a complete output that can be used immediately.
- Include a next resume prompt when the conversation is not finished.
Output Patterns
Use the smallest pattern that satisfies the request.
Quick Conversation Declutter
Context:
- ...
Decisions:
- ...
Actions:
- ...
Artifacts:
- ...
Open Questions:
- ...
Next Resume Prompt:
...
Handoff
Use when another agent or future session must continue the work.
# Handoff
## Current Goal
...
## What Changed
...
## Key Decisions
...
## Files and Artifacts
...
## Risks / Unknowns
...
## Suggested Next Steps
...
If the user explicitly asks for a handoff document, prefer the existing
handoff skill when available.
Topic Memory Update
Use when a workspace has persistent topic files.
### YYYY-MM-DD
- User asked ...
- Key conclusion ...
- Files/artifacts changed ...
- Next resume prompt: ...
Never archive secrets, raw private chat histories, API keys, activation codes,
or sensitive personal details.
Wiki / Notion Capture
Use when the user wants durable documentation rather than session continuity.
If Notion is requested and the Notion skill/tools are available, use
notion-knowledge-capture.
Quality Rules
- Preserve decisions, obligations, constraints, and exact artifact pointers.
- Prefer concrete dates over relative dates when preserving timelines.
- Keep quoted copyrighted material minimal; summarize instead.
- Do not over-summarize away blockers, failed validations, or unresolved risks.
- Keep personal or private material at the highest useful abstraction.
- When the user asks to apply this to the current conversation, produce the
organized result directly; do not merely describe the workflow.
