fix(server): brand generated branches and checkpoints as Aldo - #63
fix(server): brand generated branches and checkpoints as Aldo#63logancsack wants to merge 1 commit into
Conversation
Managed workspaces are branded Aldo everywhere the client renders, but the server still wrote T3 Code into the user's own repository: generated worktree branches were `t3code/<name>`, cross-repository pull request checkouts were `t3code/pr-<n>/<head>`, and every turn checkpoint was authored by `T3 Code <t3code@users.noreply.github.com>`. The generators now use `aldo`, while every matcher keeps reading the old prefix from `LEGACY_WORKTREE_BRANCH_PREFIXES`. Existing temporary branches stay regenerable, an existing pull request worktree is still found instead of duplicated under the new prefix, and a reused worktree reports the name it actually carries rather than the one this build would have generated. Verified: vp test run on packages/shared, apps/server/src/git, apps/server/src/vcs, apps/server/src/sourceControl, apps/server/src/orchestration and apps/server/src/server.test.ts; targeted lint, fmt, and typecheck. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
You have reached your Codex usage limits for security reviews. Please try again later. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit:5f1708c423
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| export const WORKTREE_BRANCH_PREFIX = "t3code"; | ||
| // Canonical form is `t3code/<8 hex>`. Older mobile builds generated `t3code/<uuid>` | ||
| export const WORKTREE_BRANCH_PREFIX = "aldo"; |
There was a problem hiding this comment.
When a normal npx t3 or desktop server runs with managedDevPc === false, this shared constant is still aldo, so every temporary worktree, generated feature branch, and cross-repository PR worktree writes aldo/... into an ordinary T3 Code repository. The web already limits Aldo branding to VITE_DEVPC_MANAGED in apps/web/src/branding.ts, so the server prefix should likewise come from the managed configuration rather than changing the global default.
AGENTS.md reference: AGENTS.md:L23-L31
Useful? React with 👍 / 👎.
| GIT_AUTHOR_NAME: "Aldo", | ||
| GIT_AUTHOR_EMAIL: "aldo@users.noreply.github.com", | ||
| GIT_COMMITTER_NAME: "Aldo", | ||
| GIT_COMMITTER_EMAIL: "aldo@users.noreply.github.com", |
There was a problem hiding this comment.
On a standalone T3 Code installation, this driver executes without checking ServerConfig.managedDevPc, so every turn's hidden checkpoint commit is now authored and committed by Aldo as well. Inspecting, exporting, or otherwise retaining those refs therefore records the managed product's identity in normal users' repositories; select the identity from the server branding and retain T3 Code for unmanaged builds.
AGENTS.md reference: AGENTS.md:L23-L31
Useful? React with 👍 / 👎.
| return pullRequestWorktreeBranchNameCandidates( | ||
| input.pullRequestId, | ||
| sanitizeBranchFragment(input.headBranch), | ||
| )[0]!; |
There was a problem hiding this comment.
When a fork PR was previously checked out locally by an older build, only a branch such as t3code/pr-42/main exists. This path always selects candidate zero, and the later listLocalBranchNames check consequently tests only the new Aldo name, so preparePullRequestThread in local mode fetches and switches to a duplicate aldo/... branch instead of reusing the legacy checkout. Select the first existing current-or-legacy candidate before fetching and configuring the upstream.
AGENTS.md reference: AGENTS.md:L67-L71
Useful? React with 👍 / 👎.
| const localBranch = result.refs.find( | ||
| (branch) => !branch.isRemote && branch.name === localPullRequestBranch, | ||
| (branch) => !branch.isRemote && localPullRequestBranchNames.includes(branch.name), | ||
| ); |
There was a problem hiding this comment.
When both branch names coexist—for example, a newer bare t3code/pr-… ref remains after its worktree was removed while the corresponding aldo/pr-… branch is currently checked out—this find selects the legacy ref because GitVcsDriver.listRefs sorts by commit recency rather than candidate order. It then materializes the Aldo branch instead of returning its existing worktree, and Git rejects the forced branch update because that branch is checked out; git worktree add -h likewise documents that checking out an already-used branch requires --force, which this path does not pass. Prefer a matching ref with a worktreePath before considering bare candidates.
Useful? React with 👍 / 👎.
Thread transfer impact✅ Thread transfer remains within every enforced ceiling.
Baseline: Scenario and decoded snapshot size10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.
Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed. |
Problem
Managed Aldo workspaces are branded everywhere the client renders, but the server still wrote T3 Code into the user's own repository:
t3code/<name>— visible in the composer footer,git branch, and any PR opened from onet3code/pr-<n>/<head>T3 Code <t3code@users.noreply.github.com>, so the rename never reached the user's git historyFix
WORKTREE_BRANCH_PREFIXis nowaldo, PR checkout names are built from the same prefix in both the GitHub and Bitbucket paths, and checkpoints are authored byAldo <aldo@users.noreply.github.com>.Only the generators moved.
LEGACY_WORKTREE_BRANCH_PREFIXESkeeps every matcher reading the old names, so existing state is not stranded:isTemporaryWorktreeBranchstill matchest3code/<hex>and the UUID-shaped names older mobile builds produced, so those threads stay eligible for branch regenerationstripWorktreeBranchPrefixremoves either prefix, so regenerating a name cannot stack one prefix on the othert3code/pr-<n>/<head>is still found rather than duplicated under a newaldo/branch, and the reused worktree now reports the branch it actually carries instead of the one this build would have generated.github/devpc-release-versionis bumped to0.0.31-devpc.2so merging cuts the artifact the managed gateway pins.Verification
vp test run packages/shared apps/server/src/git apps/server/src/vcs apps/server/src/sourceControl apps/server/src/orchestration apps/server/src/server.test.ts— all green, including a new regression test that prepares a PR thread against a legacy-prefixed worktree and asserts it is reusedvp lint,vp fmt --check, andvp run --filter t3 typecheck(0 errors)No UI surface changed, so there are no screenshots.
Model: Claude Opus 5 (1M context), harness: Claude Code.