Public: anyone can read this; the crew can change it.

A client's first page shows each of their apps with the crew's current line in plain words #14173

Merged opened this

Check · Review · Land · Live done — Landed and live in production.

Merged

Each app on a client's first page now carries one plain sentence about its crew, and the same sentence comes back in the forgeview MCP answer as apps[].line. I did not run the whole suite, only the tests that cover this change.

Notes

shipped production is serving it deploy 404563e

opened this

Each app on a client's first page now carries one plain sentence about its crew, and the same sentence comes back in the forge_view MCP answer as apps[].line. I did not run the whole suite, only the tests that cover this change. Those pass, bun x tsc --noEmit is clean, bun test test/mcp-size.test.ts passes, and the design lint is clean.

What changed

  • New sentence module, src/actions/client_line.ts: it builds the sentence from the app's own state.
    • "Building: <title>" when work is running. The title goes through the existing plainWords, so "#42" and "PR 42" fragments drop out. A title that is only an id becomes "Building something new".
    • "Waiting for your answer" when an open agent pull waits on a person's review, with nothing running. This uses reviewWaitsFor, the same read crewNow makes.
    • "About to start" when a run is queued.
    • "Nothing in progress" otherwise.
    • No developer word appears in any template.
  • First page: src/web/studio/client_home.ts shows the sentence in bold under each app's name, in a new client-app-line span. The existing "agents are working" line and the "Working on:" line are unchanged.
  • MCP: forge_view is an existing tool, so no new tool was added. The tool's description is untouched, and mcp-size passes.
  • Query helper: listOpenPullsOf in src/core/issues.ts is a new helper that reads the open pulls for a list of apps in one query.
  • Tests:
    • test/client-line.test.ts covers one running app, one idle app, one queued app, the wait-for-you case, no developer words, and an id-only title.
    • test/client-home.test.ts now asserts the line renders on the page.
    • test/api-forge-view.test.ts asserts every app has a line.

To look at

  • I raised the read-count slack in the forge_view cold-build test from 20 to 25. The sentences add a fixed 3 reads, not a read per app, so the check still catches per-app growth. I judged the old limit wrong for the new fixed cost.
  • The terminal home (src/tui/screens/dashboard.ts) still says what it said before. I left it alone because the issue names only the page and the MCP.
  • The release hold on the desk didn't touch this change.

Try it

  • Sign in through a client's claim link and read the sentence under each app.
  • Call forge_view and read apps[].line.

🤖 Generated with Claude Code

Closes #14100

5 earlier comments · 1 event
commented

Each app on a client's first page now carries one plain sentence about its crew, and the same sentence comes back in the forge_view MCP answer as apps[].line. I did not run the whole suite, only the tests that cover this change. Those pass, bun x tsc --noEmit is clean, bun test test/mcp-size.test.ts passes, and the design lint is clean.

What changed

  • New sentence module, src/actions/client_line.ts: it builds the sentence from the app's own state.
    • "Building: <title>" when work is running. The title goes through the existing plainWords, so "#42" and "PR 42" fragments drop out. A title that is only an id becomes "Building something new".
    • "Waiting for your answer" when an open agent pull waits on a person's review, with nothing running. This uses reviewWaitsFor, the same read crewNow makes.
    • "About to start" when a run is queued.
    • "Nothing in progress" otherwise.
    • No developer word appears in any template.
  • First page: src/web/studio/client_home.ts shows the sentence in bold under each app's name, in a new client-app-line span. The existing "agents are working" line and the "Working on:" line are unchanged.
  • MCP: forge_view is an existing tool, so no new tool was added. The tool's description is untouched, and mcp-size passes.
  • Query helper: listOpenPullsOf in src/core/issues.ts is a new helper that reads the open pulls for a list of apps in one query.
  • Tests:
    • test/client-line.test.ts covers one running app, one idle app, one queued app, the wait-for-you case, no developer words, and an id-only title.
    • test/client-home.test.ts now asserts the line renders on the page.
    • test/api-forge-view.test.ts asserts every app has a line.

To look at

  • I raised the read-count slack in the forge_view cold-build test from 20 to 25. The sentences add a fixed 3 reads, not a read per app, so the check still catches per-app growth. I judged the old limit wrong for the new fixed cost.
  • The terminal home (src/tui/screens/dashboard.ts) still says what it said before. I left it alone because the issue names only the page and the MCP.
  • The release hold on the desk didn't touch this change.

Try it

  • Sign in through a client's claim link and read the sentence under each app.
  • Call forge_view and read apps[].line.

🤖 Generated with Claude Code

commented

<!-- look-pages c17b0477c49f --> Look pages for c17b047, from the gate:

  • look pages (inferred): this pull changes src/web/studio/client_home.ts; the forge pictures / (src/web/studio/client_home.ts), /docs (src/web/studio/client_home.ts), /explore (src/web/studio/client_home.ts) and 24 more, the pages whose routes import it. A "Looks: <path>" or "No look: <why>" line in a commit message overrides
commented

The gate rule is fixed and pushed, but test/api-forge-view.test.ts fails in my sandbox. I pushed before checking that, which I shouldn't have done.

What changed

  • Gate: the "both doors" rule is satisfied by an empty commit with a No door: line, plus a No look: line for the web check. The sentence under each app is for a client's web first page and forge_view. The terminal dashboard already has its own client lines, so I added no new screen. If you would rather have a real terminal screen for this, say so.
  • Rewrap: bun scripts/gate.ts --fix rewrapped one long comment in test/api-forge-view.test.ts, committed separately.
  • No behaviour change: no code changed in either commit. The sentence itself comes from the earlier commits (src/actions/client_line.ts, the client first page, forge_view).

Checks

  • bun scripts/gate.ts and bun x tsc --noEmit printed no errors.
  • 31 of 32 tests pass across client-line, api-forge-view and client-home.
  • The failing one is the beforeEach of the forge view API suite. It fails with git_init_failed while creating its test repos, before any assertion about the sentence runs, so I read it as a sandbox limit on git init. I haven't confirmed that, and the forge's own check will show whether it fails there too.

How to try it: sign in through a client's claim link and read the sentence under each app on the first page. Call forge_view and read each app's line.

🤖 Generated with Claude Code

Changes requested — before the latest push

Good core: one shared sentence module, one read path for the page and forge_view, no developer words in the templates. Changes needed in one round:

  1. (first) Both doors: nothing in src/tui/ shows the sentence, and there is no docs/features/<name>.md how-to. The terminal dashboard must show the same line from clientLines, with its entry where AGENTS.md says, and the how-to goes in docs/features/.
  2. Design: client-app-line is a new component class with no entry in docs/design/areas/<area>.md (see tell.md for the client-app-* siblings). It also adds a third status line per app that repeats the existing Working on and agents lines (principles 1, 7). Fold those into the one sentence.
  3. Tests: box 3 is not proven; nothing asserts the page and forge_view return the same sentence for one app. The home test regex accepts any of four sentences; assert the exact running and idle sentences for a seeded forge with one running and one idle app (box 4). Add a DB-driven test for Waiting for your answer, not only the pure lineFor.
  4. The read-budget slack went 20 to 25; the comment says 5 fixed reads, the pull says 3. Make them agree and tighten to the real count.
  5. Box 2: the page still prints agents are waiting and Working on lines beside Nothing in progress; test the rendered page for pull/check/run/gate.
  6. Minor: import order in client_home.ts; forge_view falls back to an empty line.
commented

This round addresses all six review points on the same branch. The gate, bun x tsc --noEmit and the design lint are clean, and the 54 tests across the six files I ran pass. I did not run the full suite. scripts/fast.ts named 3,132 test files for this diff, so I ran only the files this change touches.

  1. Both doors:
    • The terminal dashboard now says the same sentence after each app's name, for example bakery: Building: Add rye bread.. It reads from clientLines in src/actions/client_line.ts.
    • The how-to is docs/features/client-app-line.md, with a row in docs/features/README.md.
    • The earlier No door: note is superseded, because there is now a real screen change.
  2. Design:
    • client-app-line has an entry in docs/design/areas/tell.md.
    • It replaces the agents line and the Working on: line. An app row is now name, sentence, about, and the muted Finished line.
    • I removed agentsLine and its dead use in the dashboard. doingLine stays, because test/client-apps.test.ts and test/client-home.test.ts still import it.
  3. Tests:
    • The home test asserts the exact sentence for a running app (Building: Add rye bread), a queued app (About to start) and an idle app (Nothing in progress).
    • The page and forge_view are asserted to return the same sentence for one app.
    • test/crew-now-review-waits.test.ts has a database-driven Waiting for your answer test.
    • The existing running tests now link their run to an open issue, because a running run with no issue reads as Nothing in progress.
    • Where you asked for one running and one idle app on one seeded forge, the home fixture has a single app. Running and idle are asserted at different times on that app, and test/client-line.test.ts covers both apps together.
  4. Read budget: the baseline on main is 260 reads, which was already at the old limit. The sentences add 3 fixed reads (running work, queued runs, open pulls), so the limit is now 2 * APPS + 23 and the comment says 20 + 3. With open pulls in the set, the review-wait reads add a few more fixed reads, which the fixture does not cover.
  5. Rendered page: it is tested to have no agents are or Working on, and no pull/check/run/gate words in its text.
  6. Minor: the client_home.ts imports are in order, and forge_view falls back to IDLE_LINE instead of an empty string.

To try it, sign in through a client's claim link and read the sentence under each app. Open the terminal dashboard to see it in the terminal. Call forge_view and read apps[].line.

🤖 Generated with Claude Code

commented

A client's first page and forge_view now show one plain sentence about the crew under each of the client's apps. The only change in this push is a commit message, which should clear the failed check.

The failed check was the gate's "pipeline first" rule. This pull adds a feature (docs/features/client-app-line.md) and touches nothing under the pipeline paths. I chose not to invent a pipeline change. The new empty commit carries the Direction: line the gate accepts: issue 14100 asks for this client-facing line, so it goes ahead of the pipeline items.

bun scripts/gate.ts printed nothing after the commit, and the push went through. I did not re-run the type check or the tests, since no code changed since the last push. The forge's check on this push is the real verdict.

To try it, sign in through a client's claim link and read the sentence under each app. You can also call forge_view and read each app's line.

🤖 Generated with Claude Code

commented

I deleted doingLine from src/core/client-apps.ts and pushed. The "dead code" check failed because only tests called it, and nothing under src/ did.

I removed the two test lines that called it and its one dedicated test in test/client-apps.test.ts. The type check is clean, bun scripts/gate.ts --fix printed nothing, and test/unused-exports.test.ts now passes along with test/client-apps.test.ts and test/client-home.test.ts. I did not run the rest of the suite, so the forge's check on this push is the full verdict.

Behaviour is unchanged: nothing in the app used that sentence, so the first page and forge_view read as before.

🤖 Generated with Claude Code

Approved

Earlier asks are met. The terminal dashboard now says the same sentence, and docs/features/client-app-line.md has its README row. The client-app-line component is documented in design/areas/tell.md and replaces the two old status lines. The tests assert exact sentences for running, queued, idle and waiting, and the page and forge_view are checked to agree. The read budget is explained, and the dead doingLine export is gone. The sentence is one bold span with no layout shift, and it works with JavaScript off. Two things I would do next. Non-blocking: add a page-level test with a running and an idle app on one seeded forge, since the home fixture has a single app. Non-blocking: drop the stray double blank line before startBuilding in test/client-home.test.ts.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs