docs: rewrite the README around a real demo - #7
Conversation
The old README opened by calling the TUI, durable subagents, and sandboxing "on the roadmap" and then described all three as working further down. All three have shipped; the status paragraph had just gone stale. Rewrites it as something someone can actually adopt from: a recorded demo of `bullpen agents` up top, the three differentiators stated plainly (durable by default, no supervisor, sandbox as a feature), install and quickstart that work without reading further, and a status table that separates shipped from next from planned instead of implying more than exists. The demo is a real run, not a mockup — VHS drives the installed binary against an isolated store, dispatching three agents, peeking a real answer, and showing the sessions persist after the dashboard closes. The tape is committed next to the GIF so it can be re-recorded. Also documents the WAL immutable-flag trick for reading the store while sessions are running, which is otherwise a surprise. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PfAfAujueuZ3rDTiL9apx3
|
Warning Review limit reached
Next review available in: 17 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: ⛔ Files ignored due to path filters (1)
📒 Files selected for processing (2)
Comment |
1 participant
Summary
The README's status paragraph called the TUI, durable subagents, and OS sandboxing "on the roadmap" — then the body described all three as working. They shipped; the opening had gone stale and was actively underselling the project.
This rewrites it as something a stranger can adopt from, and puts a recorded demo at the top.
What changed
A real demo, not a mockup. VHS drives the installed binary against an isolated store: three agents dispatched from the dashboard input line, watched through Working → Completed, one peeked to show its actual answer and the
bullpen run -rresume hint, thenbullpen sessionsproving the record survives the dashboard closing. 49s, 279KB. The tape is committed beside the GIF so anyone can re-record it.The three differentiators stated up front rather than buried: durable by default, no daemon or supervisor, and sandboxing as a first-class feature.
A status table that separates shipped / next / planned. The previous framing made it hard to tell which of five providers were live-verified versus config-only, or that Landlock and the workflow engine are still unbuilt. Both are now explicit, including the honest note that outside
--sandboxtools run with full process authority.One genuine gotcha documented: reading
bullpen.dbwhile sessions run needs?immutable=1, because a WAL database can't be opened read-only without its shared-memory file. Plainsqlite3 -readonlyfails with a misleading "unable to open database file".Verification
Every factual claim was checked against the code rather than carried over from the old text:
crates/tools/src/exactly.github/workflows/ci.yml) existsrust-toolchain.toml(1.97.1)ARCHITECTURE.md) resolvesdocs/media/assets are committedPost-Deploy Monitoring & Validation
Check the rendered README on GitHub after merge: the centered header block, the badges (CI should read passing), and the GIF loading at full width. If the demo drifts from reality as the CLI changes, re-record with
vhs docs/media/bullpen-agents.taperather than editing the GIF.No runtime or production impact — documentation and a media asset only.