Projects keep their backend, plans hold the brief, limits follow progress, review reads the code #310

Merged
cmoriarty merged 6 commits from claude/batch-305 into main 2026-10-29 16:19:44 -04:00
Owner

Four OpenSpec changes in one deploy. Issues close once each is confirmed on production.

projects-keep-their-backend (#305)

  • Fixed at creation: a run records its profile when it's created, and runs on it from its first agent step, whatever profile is in use by then.
  • Projects: a project's item runs take the project's profile. The project page shows it and can change it (POST /api/projects/{id}/backend).
  • New Run: a Backend choice starts a run on another profile without switching the one runs use. POST /api/runs takes profile.

plans-hold-what-the-brief-asks (#308)

  • Now, Next and Later hold only what the description asks for.
  • The planner's own ideas go in a new ## Could add list (at most five), which no run builds. The replan keeps the list as it is, and the project page lists it apart from the milestones.

limits-follow-progress (#303)

  • Near and at its deadline, a step whose last ten minutes show progress (a file written, three different tool calls finished, a subagent's report) gets more time: a quarter of its limit, at least 15 min, never past three times its limit.
  • A step with no progress is asked what's left rather than ordered to finish, and is stopped at its deadline as before.
  • The same tool call repeated, with nothing written, doesn't count as progress.

review-briefs-and-rows (#307, #268, #309, #199)

  • Code review (#307): it runs as a review agent with no browser subagent and no browser tools, and its brief leaves looking at the app to the judge.
  • Scenario-test brief (#268): it now says one thing about where its tests go.
  • Inference page (#199, #309): on hosted backends, running steps get their rows back. The model is named only on Free.

Also

  • planned-projects-finish-right is archived: #263, #270, #272 and #273 were confirmed on production, and the rest of #268 moves here.

Checked

  • ./tools/test.sh: 238 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 specs pass.

🤖 Generated with Claude Code

Four OpenSpec changes in one deploy. Issues close once each is confirmed on production. **`projects-keep-their-backend`** (#305) - **Fixed at creation:** a run records its profile when it's created, and runs on it from its first agent step, whatever profile is in use by then. - **Projects:** a project's item runs take the project's profile. The project page shows it and can change it (`POST /api/projects/{id}/backend`). - **New Run:** a Backend choice starts a run on another profile without switching the one runs use. `POST /api/runs` takes `profile`. **`plans-hold-what-the-brief-asks`** (#308) - Now, Next and Later hold only what the description asks for. - The planner's own ideas go in a new `## Could add` list (at most five), which no run builds. The replan keeps the list as it is, and the project page lists it apart from the milestones. **`limits-follow-progress`** (#303) - Near and at its deadline, a step whose last ten minutes show progress (a file written, three different tool calls finished, a subagent's report) gets more time: a quarter of its limit, at least 15 min, never past three times its limit. - A step with no progress is asked what's left rather than ordered to finish, and is stopped at its deadline as before. - The same tool call repeated, with nothing written, doesn't count as progress. **`review-briefs-and-rows`** (#307, #268, #309, #199) - **Code review (#307):** it runs as a `review` agent with no `browser` subagent and no browser tools, and its brief leaves looking at the app to the judge. - **Scenario-test brief (#268):** it now says one thing about where its tests go. - **Inference page (#199, #309):** on hosted backends, running steps get their rows back. The model is named only on Free. **Also** - `planned-projects-finish-right` is archived: #263, #270, #272 and #273 were confirmed on production, and the rest of #268 moves here. **Checked** - `./tools/test.sh`: 238 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 specs pass. 🤖 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>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Near and at its deadline, an agent step whose last ten minutes show progress (a file written, three different tool calls finished, a subagent's report) has its deadline moved out by a quarter of its limit, at least 15 minutes, never past three times its limit. A step with no progress is asked what is left instead of ordered to finish, and stopped at its deadline as before.

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-305 2026-10-29 16:19:44 -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!310
No description provided.