Project Identity — Display Name vs Slug

​

What it enables

• Project names are display-only. Spaces, mixed case, accents, emoji-adjacent characters are all fine. Users type “My new landing page” and that string is what shows up everywhere user-facing.
• Slug is the identity. A collision-free slug (auto-suffixed -2, -3, … on name collision) drives every external resource: GitHub repo path, Vercel project name, production URL, and the worker’s local <runtime>/<instance>/projects/<slug> dir.
• No more silent shared state. Before this change, two “test” projects upserted into a single Postgres row, shared one GitHub repo (gdi-labs/test.git), one Vercel project, and one production URL. Now each one gets its own row, its own <gdi-labs/test, test-2, test-3> repo, and its own Vercel project / production URL.
• Workflows list shows the project name. /workflows now reads metadata.project.name per workflow and renders it as the row title; non-project workflows still fall back to the objective text.

​

What’s changed

• alter table projects add column if not exists slug text
• update projects set slug = name where slug is null (backfill — existing names were already unique)
• alter table projects drop constraint if exists projects_name_key (drop old name UNIQUE)
• create unique index if not exists idx_projects_slug on projects (slug)
• alter table projects alter column slug set not null

• ProjectRecord gains slug: String; CreateProjectRequest stays as {name, kind, repo_url, base_branch, created_by} — slug is server-side derived.
• validate_project_name relaxed to 1–80 characters; rejects control chars only. Accepts spaces, mixed case, Unicode.
• slugify(name) mirrors the worker’s _sanitize so the slug is safe as both a GitHub repo name and a filesystem directory (alphanumeric + ._-, collapsed dashes, lowercased, unnamed fallback).
• insert_project_in_postgres allocates a unique slug with bounded retry: scan slug = $base OR slug LIKE '$base%', pick the first free -N suffix (starting at -2), INSERT, retry on the idx_projects_slug unique-violation if another worker won the race (8 attempts max).
• list_projects_from_postgres and the create response both serialize slug.

• ProjectMetadata gains slug: str. from_job reads metadata.project.slug; for workflows queued before this rollout (no slug field), it falls back to _sanitize(name) so the existing identity chain keeps working.
• Three identity sites switched from _sanitize(metadata.name) to _sanitize(metadata.slug):
  • _derive_public_remote — GitHub repo path component
  • _ensure_remote_repository — repo name passed to ensure_github_repository
  • prepare() — local project_dir = projects_dir / slug
• VercelDeploymentManager.ensure_project — name=project_metadata.slug (Vercel project names share a global team namespace, so the slug is what guarantees one Cerebrum project ↔ one Vercel project ↔ one stable production URL).

• ProjectJobMetadata and Project types gain slug. The picker / project list reads it back from the API.
• The command widget no longer slugifies the name client-side. It cap-limits (1–80 chars) and strips control characters but otherwise sends the user’s literal input as name. After createProject returns the canonical slug, the widget threads it into the workflow’s metadata.project.slug so the worker uses the right identity.
• The workflows list panel prefers metadata.project.name for the row title; falls back to workflow.objective for non-project workflows (intent jobs, prompt-only tasks).

​

Impact scope

• Additive on the wire. slug is optional in ProjectJobMetadata for backwards compatibility — workflows queued before this rollout omit it and the worker derives one from the sanitized name on the fly (same behavior as before this change).
• Migration is idempotent. Re-running ensure_schema() after the column already exists is a no-op. Existing project rows keep their id, name, and (now-populated) slug (== old name).
• No data loss. Existing names map 1:1 to their slugs on first boot; nothing is renamed, no GitHub repo gets a different slug retroactively.
• Two display names “test” are no longer the same project. Previously the second creation upserted into the first row; now it gets its own row with slug = "test-2" and an isolated GitHub repo / Vercel project. This is the intentional behavior change. If an operator wants two users to share a workspace, they pick the same project from the existing-project list (matched by id/slug, not by typing the same name).

​

Tests

/Users/azat/labs/cerebrum/worker/.venv/bin/python -m pytest tests/
cd /Users/azat/labs/cerebrum/mother-ai && cargo check && cargo test
cd /Users/azat/labs/cerebrum/apps/frontend
npx --no-install tsc --noEmit
pnpm run lint

1. Create a project named “test” in the FE; confirm slug test in the projects list.
2. Have another user create “test” — confirm a new row with slug test-2.
3. Check Postgres: select name, slug from projects order by created_at desc limit 5;.
4. Fire a project task on each — confirm the worker logs show two distinct project_dir paths (projects/test, projects/test-2), the Vercel vercel_deployed event shows two distinct project_name values, and the production URLs differ.
5. Open /workflows — confirm the row titles show the display names (“test” for both), not the slug or the objective.