Feature request: persistent "thread brief" artifact as a default mode for long strategic conversations

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Some conversations with Claude Code are not coding tasks but multi-day
strategic threads: reviewing research, shaping product decisions, delegating
work, coming back after a day or two with follow-up questions. As a founder,
most of my highest-value Claude usage looks like this.

Re-entering such a thread is painful today. The conversation title is not
enough to reload context — and it is often misleading by the time the thread
has evolved past its original question. The only way back in is scrolling
through dozens of exchanges, reconstructing what the original intent was,
which conclusions were reached along the way, what got decided or delegated,
and what is still open. That takes many minutes of tedious re-reading on
every return — for a thread I myself have been driving. The cost repeats on
each comeback and grows with the length of the thread, so the most valuable
conversations are the most expensive ones to resume.

Existing mechanisms don't solve this: context compaction/summarization serves
the model's context window, not the human re-entering the thread; and the
conversation list gives me only a title. Related open requests (#82203
threads, #81695 conversation anchors, #55787 cross-product history) address
conversation mechanics, not human re-entry.

Proposed Solution

A built-in, ideally default, "thread brief" for long-running conversations:

• Claude detects that a thread has become long-lived and decision-heavy

(multi-day span, many substantive exchanges, evolving decisions) and starts
maintaining a compact brief automatically — or offers to.

• The brief is one short page at a stable URL/artifact per conversation,

updated in place as the thread evolves (not a one-off summary), with
three sections:

1. Where we started — the original question and intent behind the thread;
2. Conclusions so far — only the findings/decisions that still shape the

final outcome (deliberately selective, not a log);

1. Where we are now — done / needs-attention / waiting items, plus what

the last few exchanges imply.

• Hard constraint: readable in under a minute.
• Bonus: surface the brief from the conversation list (hover/preview), which

would fix the "title tells me nothing" problem at the source.

We have validated this manually: I asked Claude to maintain exactly such an
artifact and persisted the behavior as a CLAUDE.md rule, so it now happens in
every long session in our workspace. Returning after two days went from ~10
minutes of scrolling to ~30 seconds with one link. It works — but it's a
per-workspace hack that every user would have to reinvent. Any user running
multi-day threads (founders, PMs, researchers) pays the re-entry tax this
would remove; the cost is one small artifact per long thread, and the benefit
compounds with every comeback.

Alternative Solutions

Manual CLAUDE.md rule + per-thread artifact (works, but must be reinvented per workspace); context compaction (serves the model, not the returning human).

Priority

High - Significant impact on productivity

Feature Category

Performance and speed

Use Case Example

Use Case Example

Real scenario from last week (as a startup founder using Claude Code on the web):

Day 1. I open a thread and ask Claude to review our internal research site,
a repo of experiment results, and a slide deck, then summarize strategically
what our researcher achieved and brainstorm what to build on top of it. Over
the next hours the thread evolves far past that opening question: Claude
corrects one of my assumptions with data, I challenge one of its claims (it
verifies in code and concedes), we register a product hypothesis, it files
three tickets to my team's tracker, designs an A/B experiment, and builds a
demo. Dozens of substantive exchanges; several decisions.

Day 3. I come back with "just one follow-up question." But: the
conversation title still reflects the Day-1 opening question, not what the
thread became. I no longer remember which of my ideas survived Claude's
fact-checking, what exactly got delegated vs. parked, or why we rejected one
approach. Before I can ask my follow-up safely, I scroll for ~10 minutes
through the history, re-reading my own questions and guessing the intent
behind them. This repeats on Day 5, Day 7 — every return, a bit worse.

With this feature:

1. Around exchange ~15 on Day 1, Claude notices the thread has become

long-lived and decision-heavy, and starts maintaining a "thread brief"
artifact (or asks: "Want me to keep a brief for this thread?").

1. The brief is one page, stable URL: (a) where we started — the original

question and intent; (b) the conclusions so far that shape the outcome —
e.g. "idea X died after verification; hypothesis Y registered; tickets
A/B/C filed"; (c) where we are now — done / needs my attention / waiting
on whom.

1. On Day 3, I open the brief link, read for ~30 seconds, and ask my

follow-up with full context — no scrolling, no re-deriving my own past
reasoning.

1. Every notable change updates the same artifact in place, so the brief is

always the thread's current state, not a stale snapshot.

We simulated this manually (I asked Claude to maintain exactly such an
artifact, and we persisted the habit as a CLAUDE.md rule). Re-entry went from
~10 minutes to ~30 seconds. That's the feature.

Additional Context

_No response_