Planned projects finish right, Free runs finish, metrics and tidy-ups #301

Merged
cmoriarty merged 5 commits from claude/batch-301 into main 2026-10-29 03:21:04 -04:00
Owner

Three OpenSpec changes in one deploy. Issues close once each is verified on production.

planned-projects-finish-right (#263, #270, #268, #272, #273)

  • Project row and page (#263): state shown as components: a status dot, a progress bar with one segment per milestone (milestone 2 of 3 · 1 of 2 items), the running item, a time bar, and a stepper of the plan's items.
  • Gaps (#270): each run's Not done bullets are carried into PLAN.md's ## To check. The replan can take a gap into an item but can't drop one silently. A finished project's page lists the gaps first.
  • Tests (#268): the test-writing briefs list the project's tests with their counts. The new blocking lint rule B8 refuses a second browser spec file.
  • Judge (#272): the hook window.__app is asked for whenever the brief describes a web app. The judge's report now requires play, and unclear fails the round.
  • Node (#273): the lockfiles were complete. The Mac's Node 20.18 is older than Vite 8 supports, and npm skipped the native binding silently. New Node projects get .nvmrc and an .npmrc with engine-strict=true; runs install with the check off. ui/ gets the same.

free-runs-finish (#202, #204, #199 points 2–4)

  • Beam (#202): braid-reasoning.js gives each earlier turn that has no reasoning a space, so beam can continue another model's session. This finishes wip/202-beam-reasoning. The simulated run free-reasoning-required fails with the plugin off (a respawn, as in run 78) and passes with it on.
  • Zen (#204): the refusal move and zen_for skip a Zen model cooling as busy.
  • Inference page (#199): each running Free step names its model. Anthropic has no slot count unless Parallel sessions is set. The quota cell counts only lasting limits: lowest: reader tier 99% left today · refills 00:00 PT.

metrics-and-tidy-ups (#10, #191, #296)

  • Metrics (#10): /metrics adds osf_steps, osf_runs, osf_gates_open and osf_step_attempts_max, read on a read connection.
  • Settings (#191): the backend selector names a backend once when its name is its kind and model.
  • Simulated run (#296): the fixture's README quoted npx playwright test, and the documented-command rule picked it. It now names only its stand-in command.
  • Housekeeping: four changes that shipped on 10-18/19 are archived, and a stale copy of repair-agent is removed.

Checked

  • ./tools/test.sh: 223 backend tests, 63 unit tests and 18 browser journeys pass, in 14 s.
  • ./tools/sim.sh: all 69 scenarios pass.
  • openspec validate --specs --strict: all 72 specs pass.
  • I looked at the project page in a browser.

🤖 Generated with Claude Code

Three OpenSpec changes in one deploy. Issues close once each is verified on production. **`planned-projects-finish-right`** (#263, #270, #268, #272, #273) - **Project row and page (#263):** state shown as components: a status dot, a progress bar with one segment per milestone (`milestone 2 of 3 · 1 of 2 items`), the running item, a time bar, and a stepper of the plan's items. - **Gaps (#270):** each run's **Not done** bullets are carried into `PLAN.md`'s `## To check`. The replan can take a gap into an item but can't drop one silently. A finished project's page lists the gaps first. - **Tests (#268):** the test-writing briefs list the project's tests with their counts. The new blocking lint rule B8 refuses a second browser spec file. - **Judge (#272):** the hook `window.__app` is asked for whenever the brief describes a web app. The judge's report now requires `play`, and `unclear` fails the round. - **Node (#273):** the lockfiles were complete. The Mac's Node 20.18 is older than Vite 8 supports, and npm skipped the native binding silently. New Node projects get `.nvmrc` and an `.npmrc` with `engine-strict=true`; runs install with the check off. `ui/` gets the same. **`free-runs-finish`** (#202, #204, #199 points 2–4) - **Beam (#202):** `braid-reasoning.js` gives each earlier turn that has no reasoning a space, so beam can continue another model's session. This finishes `wip/202-beam-reasoning`. The simulated run `free-reasoning-required` fails with the plugin off (a respawn, as in run 78) and passes with it on. - **Zen (#204):** the refusal move and `zen_for` skip a Zen model cooling as busy. - **Inference page (#199):** each running Free step names its model. Anthropic has no slot count unless Parallel sessions is set. The quota cell counts only lasting limits: `lowest: reader tier 99% left today · refills 00:00 PT`. **`metrics-and-tidy-ups`** (#10, #191, #296) - **Metrics (#10):** `/metrics` adds `osf_steps`, `osf_runs`, `osf_gates_open` and `osf_step_attempts_max`, read on a read connection. - **Settings (#191):** the backend selector names a backend once when its name is its kind and model. - **Simulated run (#296):** the fixture's README quoted `npx playwright test`, and the documented-command rule picked it. It now names only its stand-in command. - **Housekeeping:** four changes that shipped on 10-18/19 are archived, and a stale copy of `repair-agent` is removed. **Checked** - `./tools/test.sh`: 223 backend tests, 63 unit tests and 18 browser journeys pass, in 14 s. - `./tools/sim.sh`: all 69 scenarios pass. - `openspec validate --specs --strict`: all 72 specs pass. - I looked at the project page in a browser. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Finishes the work on wip/202-beam-reasoning (d6c469c): the braid-reasoning plugin, the catalog's needs_reasoning, the fake router that refuses as beam does, and the simulated run free-reasoning-required, which fails with the plugin off (a respawn, as in run 78) and passes with it on.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The archives of writes-count-as-progress, fastllm-and-tabbyapi-kinds, malformed-tool-call-continues and slot-ceiling-and-throughput add to older requirement text only; context-budget's merged requirement is split to stay under the length rule. openspec/changes/repair-agent was a stale copy of 2026-10-19-repair-agent.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
cmoriarty deleted branch claude/batch-301 2026-10-29 03:21:05 -04:00
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!301
No description provided.