feat(projects): collaborator access guard and the two project reads (app#2048)
Implements the first api row of recoupable/app#2048. Contract: recoupable/docs#326.
Blocked on recoupable/database#68. The tables do not exist yet, so Tables<"project_collaborators"> is not in types/database.types.ts and tsc fails with Type '"project_collaborators"' does not satisfy the constraint. Merge that PR, run pnpm update-types, and I will push the regenerated types onto this branch to turn CI green. Nothing else about this PR is waiting.
What this adds
| Endpoint | Returns |
|---|---|
GET /api/projects/{projectId} | The project, its tasks oldest first each carrying comment_count, and its collaborators |
GET /api/projects/{projectId}/tasks/{taskId} | One task with its comment feed and the project's collaborators |
Plus requireProjectAccess, the gate the other two api PRs in this issue also use.
The decisions worth reviewing
404, never 403. A caller with no project_collaborators row gets 404, and so does an id that does not exist. A 403 tells the caller the project id is real, and these ids travel in links we send clients. getProjectHandler also answers 404 rather than 500 for a collaborator row whose project is missing — a broken invariant the caller can do nothing about either way.
hasProjectAccess denies on a lookup failure rather than propagating. Treating a database error as access would turn one bad minute into a disclosure. There is a test for it.
No account_id override. Access here is per project, so a caller acting for another account still needs that account's collaborator row — which is what the row already expresses. validateAuthContext is called with no override input.
Comment counts are one grouped read, not a count per task, so the project page does not scale calls with the length of the timeline.
Both ids are matched on the task read. selectProjectTask filters on project_id and id, so a task id belonging to another project cannot be read through a project the caller does collaborate on.
Verification
vitest lib/projects: 3 tests, passing. Written test-first, RED confirmed on "module not found" beforehasProjectAccessexisted.eslint: clean across every file in this diff.tsc: fails only on the missing table types described above. No other errors in these files.- Not preview-verified yet. Live checks against the deployment — 200 for a collaborator, 404 for a signed-in non-collaborator, 401 with no token, a deliberate non-UUID — run once the tables exist, and the results go in a comment before this merges.
🤖 Generated with Claude Code
https://claude.ai/code/session_01BTaK5sDuryYsQG5W56mR5n
Summary by cubic
Adds the two project reads and the collaborator access guard they share, implementing the first api row of app#2048. A malformed non-UUID path segment now answers 400 instead of 500.
GET /api/projects/{projectId}returns the project, its tasks oldest first each withcomment_count, and its collaborators.GET /api/projects/{projectId}/tasks/{taskId}returns one task with its comment feed and the project's collaborators.- A caller without a
project_collaboratorsrow gets 404, never 403. - A collaborator lookup failure denies access instead of throwing.
- One grouped comment select feeds the per-task counts and a task's comment feed, so counts don't scale with the timeline.
- The task read matches both
project_idandid, so a task id from another project can't be read through a project the caller does collaborate on. - Regenerated Supabase types for the four project tables are included, so
tscpasses.
Summary by CodeRabbit
- New Features
- Added project and project-task API endpoints for retrieving project details, tasks, collaborators, and comments.
- Project responses now include task comment counts and collaborator and comment author names.
- Added secure access checks for project data using supported API credentials.
- Added consistent handling for missing projects or tasks, unauthorized access, and server errors.