| name | zoom-out-restate |
|---|---|
| description | Restate your immediately preceding answer from the top — for a reader who knows what the project is for but has not seen its files, jargon, or the last few turns. Use ONLY when the user explicitly asks by typing or mentioning /zoom-out-restate, or says something like "I feel like I missed a memo", "what exactly is X", "describe this as you would to someone who hadn't looked at the on-disk structure". |
I've lost the thread. Your last answer used names and steps I don't have a picture of. Restate it from the top, as you would to someone who knows what the project is for but has not looked at any file, tool, or earlier message.
Shape it like this:
- One sentence on the goal. What the whole project is trying to produce.
- The stages, in order, one bullet each. For every stage: what it is in plain words, whether it is done, and — if it has a name I have been using — the name in bold so I can map it back. Define each name the first time it appears; never lean on a name I coined earlier in the session.
- Where the current question sits. Which stage it belongs to, and why it came up, in two or three sentences.
- The decision, restated. The options and what I recommend, now that the reader has the map.
Rules:
- Every sentence should survive being read alone. No "as above", no "the delta", no "the fix from before".
- Expand every acronym and every project-specific noun ("writer", "assignment", "pass B") the first time it appears.
- Prefer a short concrete example over an abstract description ("one file per post" beats "the corpus layer").
- Cut anything the reader does not need to make the decision, even if it was in the original answer. They can ask.
This is a sibling of what-shorter: that one shrinks the answer; this one rebuilds the context the answer assumed. If both are needed, do this one first.
The exchange which prompted this skill:
Claud:
- The 34 other Wayback-heavy LOD2 pages stay un-re-run, with recovered posts assigned directly in the inventory. OK?
User:
re 6 hmm, why not re-run them? aren't they effectively stale? "with recovered posts assigned directly in the inventory" I don't really understand the significance of that. what does it mean to assign a post in an inventory?
Claude:
6, explained. "Assigning a post in the inventory" means this: each inventory entry has a "Raw posts to read" list, and I put the recovered posts' ids into the lists of the notes they belong to, so the writer reads those posts directly even though the LOD2 summary never mentions them. It works because writers read raw posts anyway; it does not fix the LOD2 entries, which stay stale. Stale by how much: after pruning the duplicate twins, 53 pages carry 210 recovered posts that their LOD2 does not cite. The heavy ones are p021 (15), p028 (11), p029, p030, p034 (10 each), p014, p015, p061 (8).
User:
re 6 I feel like I missed a memo on this one. what exactly is the inventory? what is a writer? what is an assignment? I think maybe I don't really understand what is being proposed. can you describe this as you would to someone who was conceptually familiar with the project idea, but hadn't looked at any of the on-disk structure, etc?
Claude:
6, from the top. The project turns a 143-page forum thread into a wiki of short, evergreen articles where every statement cites the post it came from. It runs in stages:
- Corpus. The thread itself, one file per post, plus the attachments. Done.
- LOD1 and LOD2. Two levels of summary, one per forum page. LOD2 is about 800 words per page: what happened, every claim with its msg id, every number, which disputes moved. This is the index a synthesis step can read instead of the whole thread. Done, but before a later step restored 288 deleted posts, so 53 of the 143 summaries do not mention posts that now exist on their pages.
- The inventory. That is what I wrote today. It is the wiki's build plan: a list of the 119 articles the wiki will have, and for each its filename, its thesis in a sentence, the claims it must make with their sources, the posts its author must read, and the other articles it links to.
- The writers. Each article is written by a separate subagent given its inventory entry. "Writer" just means one of those agents.
- Verification. A mechanical check of every citation, then a review.
Where the question comes from: writers get their material from the summaries and from the posts listed in their entry. The restored posts are missing from the summaries, so I had proposed listing them in the entries instead. Re-running the 53 stale summaries fixes the index properly, and I now recommend that.