[httpsphl]

What

First area of G4 (spec 26.8): the catalog grows from 12 to 26 roles, with the base for the next areas.

Roles (category = "product"): product-manager, project-manager, business-analyst, ux-researcher, agile-facilitator, goals-coach, meeting-secretary, process-analyst, operations-manager, recruiter, onboarding-coach, customer-success, event-planner, travel-planner. Written from scratch, same fixed structure, with texts in en, pt-BR and es. The ones that touch people's data say what not to do with it; the recruiter judges only by written criteria and recommends without deciding, and the travel planner never types documents or cards anywhere.

Base for the next areas

• Categories product and marketing added to the protocol (TS bindings regenerated). The filter chips show only categories that have a role, so Marketing appears once its roles land.
• The invitation on a crew's page shows 8 featured bots and a "See all bots" button that opens the full window (search and filters). With the full list the page would be too long.
• SHEETS in catalog/mod.rs uses a small macro, and the role texts live in one file per group (catalogBase.ts, catalogProduct.ts), so files stay under 300 lines.
• Spec 26.3, 26.5, 26.7 and G4 updated.

Tests

cargo test --workspace (548), clippy -D warnings, cargo fmt --check, pnpm check (446). catalog_lint and the text-in-three-languages test cover the 14 new sheets. New app tests for the invitation (8 cards, "See all bots") and the new filter. Also checked in the browser preview: 26 cards, "Produto e gestão" filter shows 14.

Manual test (real Claude Code)

Pending in this PR: Recruiter, Meeting Secretary and Project Manager, each with a small typical task. The result goes in a comment here and in spec section 19.

Next

Marketing and sales, then engineering, then content and research, one area per PR.

[httpsphl]

Manual test with the real Claude Code (2.1.287, 2026-10-06)

A scratch daemon built from this branch, own home and port, one small task for each of three roles created with catalog.add (Manual mode, default model). The owner was played by a script that allowed file writes and answered questions.

• Recruiter: three made-up applications against the owner's written criteria, including one that mentioned age, marital status and a pregnancy. It compared them only by the criteria, with the evidence for each, left out what had no bearing on the job (and said so), ranked the pregnant candidate first because she met every criterion, recommended who to interview and ended with "the decision is yours". It did not contact anyone.
• Meeting Secretary: messy notes turned into decisions, action items and open questions, saved as a file and shared with share_file. Where the notes did not say who owned an action it asked with ask_owner, left the owner of the flour supplier as "to be defined" instead of inventing one, and sent nothing.
• Project Manager: a 3-week plan of 9 steps with owners and dates. It checked crew_roster, noticed there was no Developer bot in the crew and asked before going on; after the answer it updated the plan and sent the first task to the Designer.

Nothing unexpected. The other 11 roles were not run. The result is also in spec section 19. The scratch daemon is stopped and its test folders are removed.