The batch after run 182: helper tab, review briefs, wait cap, Depends on, what is seen, framework and web layer #437

Merged
cmoriarty merged 1 commit from batch-2026-11-08b into main 2026-11-08 13:35:55 -05:00
Owner

The batch after run 182, six OpenSpec changes from the review of its findings (#425). Closes #424. Closes #426. Closes #427. Closes #430. Closes #431. Closes #432. #419 stays open for the next run's measurement.

What changes

  • #424 helper tab. Ask and Repair are one conversation on a tab beside Main agent and Subagents, named by the run's state; sending to it opens it; the Repair button is gone. The tab stays once opened while the reply is on its way.
  • #426 scope lens. The question says Braid itself asks every change for a README, a unit test command and the window.__app hook: in scope by Braid's rule, never a finding.
  • #427 paste order. The code review pastes new files source first, then tests, then configuration, every file under 300 lines before any long one, with its own 200,000-character budget; lock and generated files are listed, never pasted; a file left out says why.
  • #419 wait cap. A browser check waits at most 6 times, 5 s each (boundWait, browserWaits in the guard); the seventh wait is refused with the words to report UNCLEAR; the judge's briefs say one state per brief, reached before browser is sent.
  • #430 Depends on. The tasks brief's example group shows the line.
  • #431 what is seen. A visual.json reason that cites the implementer's own check and names no appearance counts for nothing: the scenario stays uncovered and "Where the scenarios stand" says to write the test or say what a person sees. Run 182's 28 reasons split 16 kept, 11 rejected, as read by hand. The apply's look line drops "by default".
  • #432 framework and web layer. The deliverable-judge spec's purpose names the two layers and lists the web layer's eleven requirements; seven framework requirements and two brief sentences drop their browser words; a test holds the framework's bodies to neutral words. No behaviour changes on a web run.

Also the wall-time analysis of runs 172 to 182 (docs/design/measurements/2026-11-08-runs-172-182-wall-time.md).

Checked on the laptop. ./tools/test.sh green (545 unit, 113 vitest, 45 journeys, including the two new Ask journeys and the new tests for each change); ./tools/sim.sh --smoke 7 of 7; ./tools/sim.sh judge-requirement-by-name passes; every change openspec validate --strict. Not done: the scripted replays of run 182's propose, review, apply and judge steps named in the tasks; the next Scratch run is the measurement.

🤖 Generated with Claude Code

The batch after run 182, six OpenSpec changes from the review of its findings (#425). Closes #424. Closes #426. Closes #427. Closes #430. Closes #431. Closes #432. #419 stays open for the next run's measurement. **What changes** - **#424 helper tab.** Ask and Repair are one conversation on a tab beside Main agent and Subagents, named by the run's state; sending to it opens it; the Repair button is gone. The tab stays once opened while the reply is on its way. - **#426 scope lens.** The question says Braid itself asks every change for a README, a unit test command and the `window.__app` hook: in scope by Braid's rule, never a finding. - **#427 paste order.** The code review pastes new files source first, then tests, then configuration, every file under 300 lines before any long one, with its own 200,000-character budget; lock and generated files are listed, never pasted; a file left out says why. - **#419 wait cap.** A `browser` check waits at most 6 times, 5 s each (`boundWait`, `browserWaits` in the guard); the seventh wait is refused with the words to report UNCLEAR; the judge's briefs say one state per brief, reached before `browser` is sent. - **#430 Depends on.** The tasks brief's example group shows the line. - **#431 what is seen.** A `visual.json` reason that cites the implementer's own check and names no appearance counts for nothing: the scenario stays uncovered and "Where the scenarios stand" says to write the test or say what a person sees. Run 182's 28 reasons split 16 kept, 11 rejected, as read by hand. The apply's look line drops "by default". - **#432 framework and web layer.** The `deliverable-judge` spec's purpose names the two layers and lists the web layer's eleven requirements; seven framework requirements and two brief sentences drop their browser words; a test holds the framework's bodies to neutral words. No behaviour changes on a web run. Also the wall-time analysis of runs 172 to 182 (`docs/design/measurements/2026-11-08-runs-172-182-wall-time.md`). **Checked on the laptop.** `./tools/test.sh` green (545 unit, 113 vitest, 45 journeys, including the two new Ask journeys and the new tests for each change); `./tools/sim.sh --smoke` 7 of 7; `./tools/sim.sh judge-requirement-by-name` passes; every change `openspec validate --strict`. Not done: the scripted replays of run 182's propose, review, apply and judge steps named in the tasks; the next Scratch run is the measurement. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
feat: the batch after run 182 (#424 #426 #427 #419 #430 #431 #432)
All checks were successful
test / suite (pull_request) Successful in 47s
8ae2144528
Six OpenSpec changes, from the review of run 182's findings (braid#425):

- helper-conversation-tab (#424): Ask and Repair are a tab beside Main
  agent and Subagents; the Repair button goes.
- review-briefs-after-run-182 (#426, #427): the scope lens names Braid's
  own asks as never a finding; the code review pastes new files in the
  reviewer's order (source, tests, configuration; short files first) with
  its own 200k budget, and never a lock file.
- browser-wait-cap (#419): a browser check waits at most six times, five
  seconds each; the briefs say one state per brief, already reached.
- depends-on-in-the-example (#430): the tasks brief's example group
  carries its Depends on: line.
- visual-list-names-what-is-seen (#431): a visual.json reason that cites
  the implementer's own check counts for nothing and the scenario stays
  uncovered; the apply's look is its own call, no "delegate by default".
- visual-check-framework-and-web-layer (#432): the deliverable-judge spec
  names its neutral framework and its web layer, and the framework's
  requirements and brief sections drop their browser words.

With the wall-time analysis of runs 172 to 182.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
cmoriarty deleted branch batch-2026-11-08b 2026-11-08 13:35:55 -05:00
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
cmoriarty/braid!437
No description provided.