New app's name-from-sentence loop is one exported action, newAppFromSentence in src/actions/apps.ts, and the web form calls it #8405

The New app name-from-sentence loop is now one exported action, newAppFromSentence(app, actor, { sentence, owner, … }, onPhase?) in src/actions/apps.ts. The web form calls it.

I ran the gate, tsc, design-lint, and actions-newapp, apps-new, apps and mcp-size tests. All pass, with none of the flaky suites going red.

• The action: it derives the name from the sentence, then tries base-2, base-3, … up to 30 on name_taken. It returns { ok, repo, app, name, issue?, plan? } or { ok:false, code }. An empty sentence returns no_name. The sentence is also passed on as the first issue's description, as the web form did before. It takes the same optional visibility, agent and pickAgent as newApp, so the web form's other fields still pass through.
• The words: NEW_APP_ERRORS and NO_NAME_ERROR now live beside it and are exported, with a new no_name entry.
• The web form: newAppSubmit calls newAppFromSentence when the name field is blank and newApp directly when a name is typed. Messages, status codes and what newApp scaffolds are unchanged, and the existing apps-new tests pass unchanged.
• New tests: test/actions-newapp.test.ts covers a sentence making a named app, the same sentence again making <name>-2, an empty or punctuation-only sentence returning no_name, an unverified account being refused, and a non-admin at the app limit being refused.
• Layering: src/actions/apps.ts now imports deriveAppName from src/web/pages/apps-name.ts, as the issue asked. That is the one place actions reach into web code.
• MCP: I added no tool and did not touch the ceilings.

To try it, open /apps/new, type a sentence and leave the name blank. Submit the same sentence twice and the second app gets -2.

Closes #8384

The New app name-from-sentence loop is now one exported action, newAppFromSentence(app, actor, { sentence, owner, … }, onPhase?) in src/actions/apps.ts. The web form calls it.

I ran the gate, tsc, design-lint, and actions-newapp, apps-new, apps and mcp-size tests. All pass, with none of the flaky suites going red.

• The action: it derives the name from the sentence, then tries base-2, base-3, … up to 30 on name_taken. It returns { ok, repo, app, name, issue?, plan? } or { ok:false, code }. An empty sentence returns no_name. The sentence is also passed on as the first issue's description, as the web form did before. It takes the same optional visibility, agent and pickAgent as newApp, so the web form's other fields still pass through.
• The words: NEW_APP_ERRORS and NO_NAME_ERROR now live beside it and are exported, with a new no_name entry.
• The web form: newAppSubmit calls newAppFromSentence when the name field is blank and newApp directly when a name is typed. Messages, status codes and what newApp scaffolds are unchanged, and the existing apps-new tests pass unchanged.
• New tests: test/actions-newapp.test.ts covers a sentence making a named app, the same sentence again making <name>-2, an empty or punctuation-only sentence returning no_name, an unverified account being refused, and a non-admin at the app limit being refused.
• Layering: src/actions/apps.ts now imports deriveAppName from src/web/pages/apps-name.ts, as the issue asked. That is the one place actions reach into web code.
• MCP: I added no tool and did not touch the ceilings.

To try it, open /apps/new, type a sentence and leave the name blank. Submit the same sentence twice and the second app gets -2.

The name-from-sentence loop is now one exported action, newAppFromSentence in src/actions/apps.ts, and newAppSubmit calls it when the name field is blank. I pushed a fix for the two failing guard tests and ran only those two files, plus test/actions-newapp.test.ts and bun x tsc --noEmit, all green. I haven't seen the forge's check on this push yet.

• Cause of the red check: two tests, test/crew-limit-words.test.ts and test/repo-words.test.ts, require that src/web/pages/apps.ts imports the refusal words (NO_SUCH_AGENT, REPO_NAME_RULES, REPO_INIT_FAILED). Those imports moved to src/actions/apps.ts along with NEW_APP_ERRORS, so the guards were looking at the wrong file.
• The fix: both tests now name src/actions/apps.ts as the reader. The guards were stale, not wrong: no copy of the words is typed anywhere, and the files that read them still import them rather than retyping them.

What changed

• newAppFromSentence(app, actor, { sentence, owner, … }, onPhase?) derives the name, then tries base-2, base-3, … up to 30 times when the name is taken.
• It returns { ok, repo, app, name, issue?, plan? } or { ok:false, code }. A sentence with nothing to name from returns no_name.
• NEW_APP_ERRORS and NO_NAME_ERROR now sit beside it and are exported, so the terminal can show the same words.
• A typed name still goes straight to newApp, and the existing New app tests pass unchanged.
• test/actions-newapp.test.ts is new. It covers a sentence making a named app, the same sentence again making <name>-2, an empty or punctuation-only sentence returning no_name, an unverified account being refused, and a non-admin at the app limit being refused.

src/actions/apps.ts now imports deriveAppName from src/web/pages/apps-name.ts, as the issue asked. I added no MCP tool and did not touch the ceilings.

To try it, open /apps/new, type a sentence and leave the name blank. Submit the same sentence twice and the second app gets -2.