Skip to content

review.md

Latest commit

 

History

History
62 lines (40 loc) · 3.08 KB

review.md

File metadata and controls

62 lines (40 loc) · 3.08 KB

Review Notes

Account Model

chatgpt_user_id is the primary user identity for account-name sync.

  • One chatgpt_user_id represents one user.
  • A user can have multiple workspace chatgpt_account_id values.
  • Those workspace records can cover personal and Team workspaces under the same email.
  • We do not treat "same email but different chatgpt_user_id" as the grouping rule for this flow.

P2

Rejected for the reported downgrade scenario.

account_id is a workspace identifier. Personal plans (free, plus, pro) can upgrade or downgrade while keeping the same personal workspace, so their account_id stays stable. Team workspaces are different: each team has its own account_id, and one user may belong to multiple teams.

The registry identity is chatgpt_user_id::chatgpt_account_id, so a team workspace and a personal workspace are different records by construction. Because of that, the reported "Team account downgraded into plus/pro/free and reused the old Team record" path does not match the account model here.

In practice:

  • Personal account transitions such as free -> plus -> pro keep the same personal account_id.
  • Team membership is represented by separate workspace account_id values.
  • A Team workspace record does not become a personal workspace record in place just because the user's personal plan changed.
  • Personal accounts do not receive synced workspace account_name values in this flow, so the claimed stale Team workspace name does not transfer through the personal upgrade/downgrade path described in the review.

P3

Accepted.

The race is not that names are written onto the wrong record by account_id matching. The problem is earlier: the detached background refresh is scheduled by one switch, but when the child process starts it re-reads the latest auth.json, so it may refresh the later active workspace instead of the workspace that triggered the job.

Current effect:

  • switch to workspace A schedules a refresh for A
  • before the child starts, another switch updates auth.json to workspace B
  • the first child reads B and refreshes B
  • workspace A is left without the expected name backfill

Simpler direction:

  • let both list and switch trigger the same detached background refresh
  • make that background refresh scan registry snapshots instead of re-reading the current auth.json
  • for each chatgpt_user_id scope that still has grouped Team accounts missing account_name, load a stored ChatGPT snapshot token and call the account API once
  • apply returned names by account_id against the latest registry state

Multiple workspace records under the same chatgpt_user_id are allowed to resolve to the same account_name. In that case, duplicate child labels are acceptable, and we do not need to preserve the old grouped fallback labels such as team #1 and team #2 once a synced account_name is available.

Example:

  • user@example.com / plan=team / account_name="Acme"
  • user@example.com / plan=team / account_name="Acme"

Rendered output:

user@example.com
  Acme
  Acme

This is acceptable for the new behavior.