Repository navigation
Releases
Factory 0.5.0
Factory 0.5.0 adds holdout scenarios: end-to-end checks written from your original request that only QA and you can read. QA's checks used to come from the same task and plan the Builder read, so a Builder could aim its code at the checks it expected. A Builder never sees holdout scenarios, so they show whether the work does what you asked, not just what its own tests check.
This is phase 4 of the factory's redesign.
How it works
- A Scenario writer starts with each Builder. It reads the task's description and your answers to the plan's questions, never the plan. It reads the code only to learn how a person uses the product. It writes 3 to 7 scenarios, each with a starting state, steps and what you should see, including at least one failure case. Its final message is only a count, and it doesn't wake the Lead.
- Only QA and you read them, on the task in the Factory page, or with
bb factory scenarios APP-12from your own terminal. They're kept in the plugin's database, never as a task comment, because the Builder reads every task comment. Status, the board, the merge card and Health show counts only, such as "Holdouts: 4/5 pass at abc1234". - QA runs each one at the commit it checks and records pass, fail or invalid. Invalid is for a scenario that misreads your request or can't be run here; QA gives the reason, and you see it on the task. QA reports a failure to the Builder as a symptom, never by copying the scenario.
- QA can't pass the task while a scenario has failed or hasn't been run at that commit, or while the writer is still at work. A fail is never held. If the writer ends without saving any scenarios, QA passes as before, and assigning QA again starts a new writer.
- You can change them on the task in the Factory page, or with
bb factory scenarios-setfrom your own terminal. Each save is a new version, and QA's results on the earlier one stop counting. If QA had already run them, the Lead is asked to run QA again. - Health shows how many scenarios passed on QA's first run of each task. The Builder never saw them, so they measure the work better than its own tests do.
Keeping them from agents
- The writer's thread isn't advertised. It isn't one of the Lead's child threads, and no task comment, status, board, run ledger or Lead message names it.
- QA reports through the factory. With holdouts on, QA's final message reaches the Lead only if it quotes no scenario. A message that quotes one is held back, and you're told. A QA task comment that quotes a scenario, which the Builder could read, is flagged to you too.
- Reads and saves from a terminal are reported. The plugin can't tell your terminal from an agent's, so each time scenarios are read or saved from a terminal outside an agent thread, you get a notification.
- Agents are told never to look for the scenarios, relay them, or use
bb plugin rpcon the factory.
Also in this release
- The Lead's standing rules are shorter. They're re-sent every turn, so they survive compaction. For Live Caption Studio and Alpha Learning website, with every option on and the Lead merging when green, they ran past bb's limit of 4,096 characters for them. They now fit for every factory name, with a line about holdouts added.
When you upgrade
- New factories have holdouts on.
- Existing factories keep them off until you turn them on:
bb factory holdouts APP on, or the Holdout scenarios switch in Settings on the Factory page. Turning them off is from your own terminal only. - A new seat in Agents: Scenario writer. It defaults to QA's provider, Codex, and falls back to QA's provider when its own isn't available. Presets treat it as a reading seat.
- With holdouts on, the Lead hears QA's result as a "[factory] QA … finished" message instead of the usual child-thread notification.
- Running agents pick up the new rules on their next turn.
Limits
- Detection, not a security boundary. An agent with file access can still read the plugin's database or search for the writer's thread. QA could describe a scenario in its own words. The Factory page can't tell your tap from an agent's call.
- Only the first build is fully blind. Reporting a failure, even as a symptom, reveals part of that scenario.
- Not tried for real yet: no factory has holdouts on. Turning them on for one factory first would show how the scenarios, QA's runs and the cost of the extra thread work out.
Full Changelog: v0.4.7...v0.5.0
Factory 0.4.7
Factory 0.4.7 lets a Planner mark its READY plan ready again when no design review round has read it, so the plan gets its round without a duplicate comment. It also says plainly why a plan has no round.
After 0.4.6, a plan stuck by the old round cap could only get its review by being posted again, word for word, as a new comment: bb factory plan-ready refused the comment that was already the READY plan. The same happened when a round's reviewers couldn't start. The Lead was told to have the Planner "mark it ready again", which was impossible.
What changes
- Marking the same plan again starts its round. This works when no round has read the plan and you haven't accepted it, for example when its reviewers couldn't start. A review still at the round cap starts no round, whatever is marked.
- Your answers are kept. Marking the same plan again keeps your answers to its questions, and the Lead isn't asked again. If you answered a question differently, the Planner still owes a new plan, and marking the old one again is refused.
- Clearer refusals. A plan a round has read is refused with "design review has read it". A plan you accepted is refused with "the person accepted it as it is".
- The Lead hears about a failed start once per plan. If the reviewers' provider is down, marking the plan again while it's still down tells the Planner the Lead already knows. A new plan is a new attempt.
- The Builder gate says why a plan has no round: "its design reviewers couldn't start" or "no design review round has read it", with the exact
plan-readycommand that starts one, ordesign-accept.
When you upgrade
- Nothing to do. A plan marked ready with no round under an earlier version can be marked again as it is.
Limits
- Not tried for real yet: no plan has been marked again on this version. The plan that prompted this was reposted as a new comment, and has since passed its review.
Full Changelog: v0.4.6...v0.4.7
Factory 0.4.6
Factory 0.4.6 fixes a plan that could never be reviewed again. The 5-round design review cap counted every round a task had ever had. A plan changed after its review passed, for example because your answers sent it back, would have been round 6, so no round started. The Builder stayed held behind a plan nobody had read, and the Planner was told the plan was ready.
On one factory, three plans in a row were posted after a passed review, and none of them got a round.
What changes
- The cap counts one review. A review ends when a round passes it, when its findings are all settled, or when you accept the plan. A plan changed after that starts a new review with every lens, and gets up to 5 rounds of its own. Rounds keep their numbers across the task, so the new review's first round can be round 6.
- Accepting at the cap ends the review. When you accept a plan that reached the cap, the next plan your answers send back gets a new review, instead of hitting the cap again.
- A new review says so, on its first round: "A new review: the last one ended after round 5, on an earlier plan."
- The Planner hears what happened.
bb factory plan-readynow adds "Design review round 6 started…", or "No design review round started: this review has had 5 rounds…". - At the cap, the factory says why. Assigning a Builder is refused with "this review has had 5 rounds, so no more start, and the person decides", and the
design-acceptcommand to use. It used to say the review "hasn't started". - One notice per capped review. The Lead and the task hear once that a review reached the cap, however many plans follow it.
When you upgrade
- A plan stuck this way gets its review once the Planner marks it ready again with
bb factory plan-ready. Its last review already ended, so the plan starts a new one. - Nothing else to do. Reviews that ended before the upgrade are recognized from their rounds.
Limits
- Within one review, the cap is unchanged: after 5 rounds, you accept the plan as it is or send it back to the Planner.
- Not tried for real yet: the stuck plan hasn't been re-marked on this version.
Full Changelog: v0.4.5...v0.4.6
Factory 0.4.5
Factory 0.4.5 trims design review: after the Planner fixes a plan, the next round re-checks only the lenses that found a Blocker or Major, instead of every lens.
Why
On the BusinessOS factory, design review cost about 12% of a planned task, weighted roughly by price, and caught 8 Blockers on four tasks before any code was written. Most of its reviewer threads, 37 of 57, ran in rounds after the first, and each of those rounds restarted every lens, including lenses that had found nothing.
How rounds work now
- Round 1: every lens of the plan's depth, as before.
- After a fix: the next round re-checks only the lenses that found a Blocker or Major, each with a fresh reviewer. This carries through any number of rounds. A lens that found only Nits, or nothing, isn't run again.
- When the depth goes up: for example, a fixed plan names a sensitive path. A lens that hasn't read the plan yet joins.
- When every lens runs again: for a new plan once a review has ended (passed or accepted), such as an amendment during the build, and after a round where a lens didn't report.
- Where you see it: the round's task comment names the lenses re-checking and those reading the plan for the first time, and why.
bb factory statusshows "(full, re-check)", and the Factory page shows "re-check".
On the four BusinessOS tasks, this would have started about 45 reviewer threads instead of 57.
When you upgrade
- Reviews already running keep their rounds. The next round started after a fix follows the new rule.
- Nothing for Planners to do differently: the factory picks the lenses, and Planners started from now on are told how it works.
Limits
- A lens that found nothing doesn't re-read the fix. A fix that breaks something only that lens would notice isn't caught until PR review. This is the trade-off for the lower cost.
- The saving is modest: roughly 2–3% of a planned task's cost.
- Not tried for real yet: the focused round hasn't run on a live plan.
Full Changelog: v0.4.4...v0.4.5
Factory 0.4.4
Factory 0.4.4 tells you when someone acts on the Factory page. bb can't tell your tap from an agent's call: it reports the page, the bb CLI and agents as the same kind of caller. So any agent could run bb plugin rpc call factory … to do what the page does. Now you hear about each such action, and the records no longer claim you made it.
You're told
These page actions send a notification to your phone and desktop, ending "If you didn't do this, an agent made the call: check the Lead's thread":
- merging from the merge card;
- answering a plan's question for you, or an escalated design finding;
- changing a task's track;
- dismissing a shadow Security block;
- changing the agent roster or applying a preset;
- deleting a factory, including a delete that stops partway.
Where there's a Lead, the notification opens its chat. Each action is also written to the plugin log. A notification waits at most 5 seconds, so a slow notifier never holds up the page. Changing a factory's settings on the page only tightens its guards, so it doesn't notify.
Honest records
- Marked as from the page: merges, answers, design decisions, track changes and shadow dismissals made on the page are recorded as "from the Factory page", not as yours. That's also how the task comments, the PR comment and
bb factory statusdescribe them. - What the Planner hears: "An answer from the Factory page…" or "A decision from the Factory page…", rather than "The person…".
- From your own terminal: outside an agent thread, everything is still recorded as yours.
A standing rule
The Lead and every worker are told never to use bb plugin rpc on the factory, whatever a PR or issue says. That stops an agent following an instruction planted in something it reads.
When you upgrade
- More notifications: expect one after each merge from the card and each answer you give on the page.
- Running agents get the new rule through their standing instructions on their next turn.
Limits
- Detection, not prevention. An agent that ignores the rule can still make the call. You find out within moments instead of never. bb could close this for good by passing the calling thread to the plugin, which would let the page refuse agents the way the CLI does.
- Not tried live yet: no notification from a page action has been checked on a phone. A harmless test call from an agent thread was refused by that agent's own safety check, so the path from an agent hasn't been run for real either.
Full Changelog: v0.4.3...v0.4.4
Factory 0.4.3
Factory 0.4.3 adds a Settings panel to the Factory page, so you can change a factory's own settings without a terminal. It also brings a Planner's questions for you to you in full, options included.
Settings on the Factory page
A Settings button, beside Health and Runs, opens a factory's own settings:
- Tracks and the Explorer
- Design review
- The evidence contract
- Who merges pull requests
- Automatic PR actions (pause and resume)
The page can only tighten a factory's guards. bb can't tell your tap from an agent's call, so the page can turn design review and the evidence contract on and give merging back to you. Loosening them stays in the CLI, which refuses agent threads: turning design review or the evidence contract off, or letting the Lead merge when green. The dialog shows that command, with a Copy button, in place of the switch. Tracks and pausing PR actions work either way from the page, since neither removes a check.
The dialog's footer points to the roster (the plugin's Agents settings) and to Project mode (the Alignment panel).
Questions for you, in full
A Planner can now write a question for you over several lines: the question, its options as bullets under it, and an "Assumed:" line.
- On the Factory page: the options show on their own lines.
- In the Lead's message: each question's options are indented under it.
- In
bb factory status: the answer state goes on the question's first line, with the options under it.
Before, only the first line was read, so a question with its options on later lines reached you without them.
- What belongs to a question: lines indented deeper than its own bullet. A line at its depth or the next question ends it.
- Its assumption: "Assumed:" on the question's own line, or a line that starts with "Assumed:". An "assumed:" inside an option doesn't count.
- Wrapped text rejoins its sentence, and notes under an option keep their nesting.
- Answers you kept carry over to the next READY plan only when the whole question, options included, and its assumption are unchanged. If the plan rewords an option or states a different assumption, you're asked again rather than having your earlier answer applied to something you didn't see.
When you upgrade
- The Settings button appears on every factory's page. Nothing changes until you use it.
- A question with options that you already answered is asked once more the next time its plan is marked ready, because its recorded wording lacks the options.
Limits
- Some per-task actions stay in the CLI: design depth, QA skip, task kind, and blocking or unblocking a task.
- Older page-only calls still can't tell you from an agent: roster changes and presets, deleting a factory, dismissing a shadow block, and answering a question on the page. They work as they did in 0.4.2.
- Tried for real: on the page, an agent's attempt to let the Lead merge was refused. The multi-line questions come from a real one: a plan's question whose options the earlier version dropped. That question, and one other, were answered through the Lead, and each Builder started. Not yet: the Settings dialog in its final form, and a multi-line question read by this version.
Full Changelog: v0.4.2...v0.4.3
Factory 0.4.2
Factory 0.4.2 shows where a factory's tokens go and keeps the Lead's context small. On the BusinessOS factory, the Lead went from 732K tokens of context to 112K, and every later Lead turn re-reads that much less.
Where tokens go
Health now has a token profile, read from the run ledger, for the turns that started in its date window:
- By role: tokens excluding cached, cached tokens, each one's share, failed and interrupted turns, and the median cached tokens per turn.
- By what became of the task: merged, closed without merging, PR still open, no PR, and turns not tied to a task (mostly the Lead's).
- Failed or interrupted turns, and their share.
- The Lead's cached tokens per turn, in the first and last quarter of each Lead thread's turns.
It's in bb factory health APP [--since …] [--before …] and below the PR numbers in the Health dialog. Tokens per merged task now use the same definition, so the two agree.
On BusinessOS since Nov 5, merged tasks took 94% of tokens excluding cached, and failed turns 2%. The Lead held 28% of cached tokens, and its context kept growing.
A smaller Lead context
- Early compaction. When a Lead goes idle and its context holds more than the new setting, "Compact the Lead's context above this many tokens" (300,000 by default; 0 is off), the factory asks bb to compact it.
- What it waits for: it waits while a merge card or a question waits for you, and checks the Lead is still idle just before compacting.
- How often: at most once every 30 minutes. After a failed attempt it waits 30 minutes, doubling with each failure in a row, up to a day.
- What stays: the Lead's standing rules come from the plugin, so they survive compaction.
bb factory lead-context APPshows how full the Lead's context is, the threshold, and the last compaction or failure.- The Lead reads evidence, not code. It's told not to read source files or diffs itself. It checks a claim against QA's verdict, CI, a task comment or
git diff --stat, and sends a specialist when code needs a look. - Short final messages. The Planner, Explorer, Builder, QA and Designer post their full report as a task comment, then end with three lines or fewer, since their final message goes into the Lead's context. PR reviewers, design reviewers and the shadow reviewer are unchanged.
When you upgrade
- Compaction is on for every factory at 300,000 tokens. Each Lead above that is compacted after its next turn, not while it's idle, so a Lead nobody is using costs nothing. Set it to 0 to turn it off.
- Running agents pick up the new rules through their standing instructions on their next turn.
Limits
- What compaction keeps: older conversation becomes the provider's summary, so details of older discussions with the Lead may be summarized.
- Cached tokens per turn grow with both the context and the number of steps in a turn. They're a sign of growth, not a measure of context size. Providers count cached input differently, so compare roles on the same provider.
- Tool overhead: the run ledger records totals per turn, so it can't show how much of a request is tool definitions.
- Tried on real work: the BusinessOS Lead was compacted from 732K to 112K, and then answered a status question correctly, at about 113K. Not yet: a compaction that fails and backs off, and compaction on a provider other than Claude Code.
Full Changelog: v0.4.1...v0.4.2
Factory 0.4.1
Factory 0.4.1 adds roster presets. Each one proposes a provider and reasoning level for every agent seat, with a reason for each. You tick the seats you want and apply them. Nothing changes until you do.
Presets
- Balanced keeps each seat as you set it, except that it moves Reviewer 1 to a provider other than the Builder's, design reviewers to one other than the Planner's, and any seat off a provider that isn't available.
- Lower cost is Balanced, plus the lowest reasoning level for the seats that read more than they write: Lead, QA, Explorer and Design reviewer.
- Maximum depth is Balanced, plus the highest reasoning level (up to
max) for the Planner, the design reviewers and the PR reviewers. Reviewer 2 also moves to a third provider when there is one.
What a preset keeps
- Your choices: a seat that stays on its provider keeps the model and reasoning level you picked, unless the preset targets that level. A seat that moves starts from the new provider's defaults.
- Levels never move the wrong way: Maximum depth doesn't lower one, such as
ultra, and Lower cost doesn't raise one, such asnone. A level the preset can't judge stays as it is, for example when a provider's models can't be read. - Providers: a preset proposes only providers available on this machine, and ones the roster already uses come first. With too few providers, a seat keeps running and says why it can't be independent.
- Reviewer seats (Reviewer 1, Reviewer 2 and Security) are shown but unticked. Tick them yourself to apply them.
- The trial seats: the shadow reviewer and the trial comparison are never touched.
Where
- Settings → Plugins → Factory → Agents: a preset picker above the seat rows, with a checkbox per seat, the reasons, and roster warnings that update as you tick seats. If the roster changes elsewhere, an open preview refreshes and says so.
- Command line:
bb factory agents preset balanced|lower-cost|max-depthpreviews a preset. Add--applyto apply the ticked seats, or--apply --seats review1,planfor only the seats you name. - Only you apply a preset. Applying is refused from an agent thread. It's also refused if the roster or the providers changed since the preview, so you never get a change you didn't see.
When you upgrade
- Nothing changes on its own: your roster stays as it is until you apply a preset.
- Leads can show you a preview through the skill, but can't apply one.
Limits
- Fixed rules, no data. bb's provider catalog has no cost or quality data, so the presets follow fixed rules about which seats need depth or independence. They make no claim about which model is better.
- Reviewer 2's third provider can mean a lower top level. At Maximum depth, Reviewer 2 can move to a provider whose best level is lower than its current one. The seat is unticked, so that trade is yours.
- Tried on this machine: previews of Balanced and Maximum depth on the command line, and
--seatswithout--applyrefused. The tests cover the rest. Not yet: applying a preset for real, from the settings page or the command line.
Full Changelog: v0.4.0...v0.4.1
Factory 0.4.0
Factory 0.4.0 adds fresh design review: reviewers who haven't seen the Planner's reasoning check each READY plan before any Builder starts. It's adapted from Google's Elephant & Goldfish Model (Apache-2.0). Questions a plan says only you can answer now reach you before work starts.
Design review
- Fresh reviewers, one per lens. When a Planner marks a plan READY, the factory starts a reviewer thread for each lens. Each one reads only the plan and the code it names, read-only.
- Core lenses: Zoom Out (simplicity, YAGNI, reuse of existing code), Comprehension (a naive reader, which starts with a paraphrase) and Codebase Details (what the plan says about the code, checked against the code).
- Full depth adds Design Critic (failure modes, scale, trade-offs) and Implementation Readiness (files, steps and tests). It runs when the plan names sensitive paths, the same ones that bring in Reviewer 2 on a PR, or when the Lead or you set
bb factory design-depth APP-12 full. - Findings and evidence. Each finding is a Blocker, Major or Nit, recorded with
bb factory design-findings. Every report ends with what it checked and what it didn't, so a clean review shows what it looked at. - The Planner resolves each Blocker and Major with
bb factory design-resolve:- fix: a new READY plan follows and starts the next round.
- contest: with evidence the reviewer could have read. A Comprehension finding can't be contested.
- escalate: for you to decide.
- decline: for Nits only.
- The gate. A Builder can't start until no Blocker or Major is unsettled. The same plan can't be marked ready again to shake off its findings, and a round's findings can't change once it has ended.
- What reaches you: only escalations, through the Lead (
bb factory design-decide) or as Keep the plan and Fix it buttons on the task on the Factory page. After 5 rounds, the Lead brings the plan to you, andbb factory design-accepttakes it as it is. - Clean-up. Once a reviewer finishes, the factory archives its thread and deletes the worktree and branch bb made for it.
Questions for you
A READY plan can list questions only you can answer, as - For the person: Q1. <question> Assumed: <what the plan builds on> under "## Open questions". The Lead gets them as soon as the plan is READY, while design review runs. No Builder starts until you've answered each one, through the Lead (bb factory plan-answer APP-12 keep|change --question Q1) or on the task. A different answer sends the plan back to the Planner.
READY plans also need a "## Success checks" heading.
When you upgrade
- New factories get design review. Existing ones keep working as before until you turn it on from outside an agent thread:
bb factory design-review APP on. It applies to plans marked ready from then on, and turning it off lets plans it was holding go ahead. - The Agents settings gain a Design reviewer seat, on Codex by default, a provider other than the Planner's. If it isn't available, reviewers run on the Planner's provider and the task says so.
- The host worker can now delete design reviewers' branches as well as Explorers'.
- Leads already running see design review through their standing rules and the skill, not their first brief.
Limits
- "Read only the plan" is a rule. The reviewers run where they could read the task's other comments.
- A contest's evidence isn't checked by the factory, only that a note is there.
- A fix that undoes an earlier one reaches you only if the Planner escalates it.
- Your answers are recorded by the Lead on your word, or by you on the page. bb can't tell a tap from an agent's call.
- A question written only in prose, without the "For the person:" line, isn't caught.
- Cost: each round starts 3 reviewer threads, or 5 at full depth.
- Tried on real tasks: on the sandbox, a five-lens review that passed in round 1, and a plan whose three Blockers were fixed and passed in round 2, both merged from the card. Not yet: an escalation from a finding, the Keep and Fix buttons, a question in the "For the person:" format, accepting after 5 rounds, and a reviewer that finishes without recording findings.
Full Changelog: v0.3.0...v0.4.0
Factory 0.3.0
Factory 0.3.0 adds tracks and an Explorer, so a task can be tried before anyone builds it. It also lets you compare Health before and after a change, and counts only the latest run of each CI check.
Tracks
Each task gets a track that decides its path before the production line:
- direct: a small, explicit change, straight to a Builder.
- plan: the Planner's READY plan comes first. No Builder starts until it's marked ready.
- explore: a time-boxed Explorer tries it first, with no PR. Then you decide what comes next, and the Lead records it by changing the track.
The Lead picks a task's track with bb factory track APP-12 direct|plan|explore when it takes the task, and you can change it, including from the task on the Factory page. Assigning the Planner or an Explorer moves a task to that track, and a Builder on a task with no track makes it direct.
Explorer
- Its own worktree, a 45-minute time box and no PR. Start it with
bb factory assign APP-12 --role explore. - A required report. It ends with a task comment recorded by
bb factory explore-report: hypothesis, what was tried, what was learned, a recommendation (build, plan, explore more or close) and what wasn't checked. - The time box. The factory reminds the Explorer near its end, tells the Lead once if it runs past, and gives up on one that stops without a report after twice its time box.
- Clean-up. Once it has finished, the factory archives its thread and deletes its worktree and local branch. The host worker deletes only a branch named as that Explorer's, never one still checked out.
- Several candidates. To compare approaches, ask for up to 3:
--candidates 2with one numbered approach each in--instructions. Once all have finished, the Lead gets them side by side, and the task on the Factory page shows them in columns.
Planner and Designer
- The Planner can end with NEEDS SPIKE, which moves the task to explore, or RECOMMEND CLOSE. That, like an Explorer's close recommendation, marks the card "Close advised". Only you close a task on an agent's advice.
- The Designer proposes two or more approaches, with their trade-offs, when the choice is open.
Health and CI
- Health for a date window.
bb factory health APP --since 2026-11-05and--before 2026-11-05count only PRs opened in that window, to compare before and after a change. The Health view on the Factory page has the same two date fields. - The latest run of each check. When a check ran more than once on a PR's head, only its latest run counts, as on GitHub. A run cancelled by a newer one no longer holds the merge card.
When you upgrade
- Tracks are on for new factories. Existing factories keep working as before, including the old plan gate, until you turn them on from outside an agent thread:
bb factory tracks APP on. - The Planner's gate stays. When you turn tracks on, a task that already has a Planner counts as plan, so no Builder starts before its plan is READY.
- The Agents settings gain an Explorer seat, on Claude Code by default.
- Leads already running see tracks through their standing rules and the skill, not their first brief.
- The host worker can now delete Explorer branches, besides running its
gh prcommands.
Limits
- The time box is a reminder and a notice to the Lead, not a hard stop.
- Closing a task on an agent's advice is yours by rule. The factory doesn't enforce it.
- The Factory page starts one Explorer. Several candidates need the CLI or the Lead.
- Pushed branches: an Explorer is told never to push. A branch it pushed anyway stays on GitHub.
- Tried on real tasks: each track on the sandbox, including a 2-candidate explore that the Lead got side by side, with both worktrees and branches cleaned up afterwards. Not yet: the time-box reminder, the past-the-time-box notice, NEEDS SPIKE, RECOMMEND CLOSE, and choosing a track on the page.
Full Changelog: v0.2.1...v0.3.0