[Feature Request] Fetch origin/<base> before creating a workspace so it branches from an up-to-date ref

[EvgeniiaVak]

Problem / use case

When I create a workspace, its branch is cut from my local origin/<base> remote-tracking ref — whatever it happened to be at my last manual fetch. If a teammate has pushed since (or I just haven't fetched in a while), the new workspace silently starts from a stale base. I usually only notice after the agent has done work, and then have to rebase/merge origin/main inside the workspace — exactly the manual step isolated workspaces are meant to avoid.

Concretely: I branch feature workspaces off origin/develop several times a day across a few repos. Any push by a colleague between my fetches means a stale start, a surprise merge, and occasionally conflicts the agent introduced against already-moved code.

Root cause (current behavior)

task_create in src-tauri/src/lib.rs branches directly off the local ref with no preceding fetch:

let base_full = args.base_branch.unwrap_or_else(|| proj.base_branch.clone()); // e.g. "origin/develop"
// ...
git(&["branch", "--no-track", &branch, &base_full], &repo)   // uses whatever origin/develop points at locally
// ...
git(&["worktree", "add", wt_arg, &branch], &repo)

The only git calls before the branch is cut are worktree prune/list and rev-parse — no fetch/pull/remote update anywhere in the create path (single- or multi-repo). So freshness depends entirely on the user having fetched recently.

Conductor already does this — it fetches from origin before creating a workspace, so the new branch is based on the latest remote commit even if the local checkout is behind. This is a straight parity gap.

Proposed solution

Before the git branch --no-track <branch> <base_full> call, do a best-effort, time-bounded fetch of just the base ref:

// base_full = "origin/develop" -> remote="origin", ref="develop"; skip if base is a local branch (no remote prefix)
// best-effort: never block or fail workspace creation on it
let _ = git_fetch_base(&repo, remote, base_branch); // git fetch --prune <remote> <base-branch>, bounded

• A single-ref fetch (git fetch <remote> <base>) updates refs/remotes/<remote>/<base> and is fast — no need to fetch all refs.
• Same insertion in the multi-repo path (host + each member around the worktree add sites), each honoring its own remote/base_branch.
• task_create already runs off the IPC handler thread (per the comment at its top), so the added network call won't block the UI.

Critical: it must never hang or block create

A naive fetch-before-create is dangerous precisely because of auth/network. Guard it so a dead remote can't wedge workspace creation:

• Env: GIT_TERMINAL_PROMPT=0 and GIT_SSH_COMMAND="ssh -oBatchMode=yes -oConnectTimeout=10" so a credential-less remote fails fast instead of prompting.
• A hard timeout on the fetch (e.g. 10–15s) with the child killed on expiry.
• Fetch failure is non-fatal — log it and proceed to branch off the existing local ref. Offline / auth-broken must still create the workspace, just from the stale ref (today's behavior), never an error.

Scope / config

• A toggle (global pref and/or .termic.yaml per-project), default on, e.g. fetch_before_create: true, so users on flaky networks can opt out.
• Optional nicety: surface a subtle "fetching base…" state during create, but the minimal version needs no new IPC surface — it's internal to the existing create command.

Note

I currently work around this with an external launchd LaunchAgent that runs git fetch --prune across all my Termic repos every 30s, which keeps the local origin/* refs current so workspaces start fresh. That works, but doing the fetch at create-time (targeted, bounded) is the right home for it and removes the polling.

[simion]

Implemented in #86.

Before a new task's branch is cut, termic now runs a best-effort, time-bounded single-ref fetch of the base (git fetch --no-tags <remote> <ref>), so the branch starts from the latest remote commit instead of a stale local origin/*, just as you described.

Guarded exactly as the issue asks so it can never wedge create: GIT_TERMINAL_PROMPT=0 + batch-mode SSH with a 10s connect timeout, plus a 15s hard deadline that kills the child. Offline / auth-broken / timeout is non-fatal, it logs and creates from the local ref (today's behavior). It's wired into both the single-repo and multi-repo (host + each member) create paths, each honoring its own remote/base, and only fires when cutting a new branch. There's a Settings > General > "Fetch base before creating a task" toggle, default on, for flaky networks.

It's a bare fetch (refs only, never pull/merge/checkout), so a dirty working copy is unaffected: it never touches tracked files. That also means you can retire the launchd LaunchAgent once this ships.

Planning to release this tonight.

[simion]

Shipped in #86 (merged to main, will be in the next release).

Thanks for the report @EvgeniiaVak, this was a genuinely good UX catch. New tasks now run a best-effort, time-bounded git fetch of the base ref right before the branch is cut, so a task starts from the latest remote commit instead of a stale local origin/<base>, no more surprise manual rebase later. It's on by default and fully non-fatal: if the remote is offline or unreachable the task still creates from your local ref, and there's a Settings > General > "Fetch base before creating a task" toggle to opt out on flaky networks. Appreciate you flagging it.