Bug: implement_task crashes on first implementation pass — no fork exists yet

[ekuris-redhat]

Bug

implement_task fails on the first implementation pass (before any PR is created) with two cascading errors:

1. git fetch fork — "fatal: 'fork' does not appear to be a git repository"
2. Cannot recreate workspace for <TICKET>: missing branch_name, current_repo, fork_owner, or fork_repo in state

If the workspace survives and the fetch error is bypassed, a second failure occurs after the container completes successfully:

3. git push -u fork <branch> — "fatal: 'fork' does not appear to be a git repository"

Root cause

Commit 7ebb8b7 ("fix: persist review handoffs across workers", PR #168) added prepare_workspace() and push_to_fork_with_retry() calls to implement_task(). Both assume a fork remote already exists. On the first implementation pass the fork is not created until create_pull_request, so fork_owner, fork_repo, and the fork git remote are all absent.

Error flow

setup_workspace  →  creates workspace, clones repo, creates branch (no fork)
                     sets current_node = "implementation"
implement_bug_fix (= implement_task)
  ├─ prepare_workspace(state)
  │    ├─ workspace exists on disk → pull_rebase("fork") → FAILS (no fork remote)
  │    └─ fallback _recreate_workspace_from_fork() → FAILS (fork_owner/fork_repo empty)
  │
  └─ (if workspace survives) container runs successfully → push_to_fork_with_retry()
       └─ git push -u fork <branch> → FAILS (no fork remote)

Affected paths

• Bug workflow: plan_approval → decompose_plan → setup_workspace → implement_bug_fix
• Feature workflow: task_approval → setup_workspace → implement_task

Any first-pass implementation where no PR has been created yet.

Post-PR paths are NOT affected

implement_review and attempt_ci_fix run after create_pull_request, so fork_owner/fork_repo are always populated.

Fix

Two changes needed in the first-pass path:

1. src/forge/workflow/nodes/workspace_setup.py — prepare_workspace()

When the workspace exists on disk but no fork info is available, skip the sync and return the existing workspace:

if workspace_path and Path(workspace_path).exists():
    workspace = Workspace(...)
    git = GitOperations(workspace)

    if not fork_owner or not fork_repo:
        return workspace_path, git

    try:
        git.pull_rebase(remote=remote)
    ...

2. src/forge/workflow/nodes/implementation.py — implement_task()

Guard both push_to_fork_with_retry() call sites so they only push when a fork exists:

has_fork = bool(state.get("fork_owner") and state.get("fork_repo"))

# "all tasks done" path
if has_fork:
    await push_to_fork_with_retry(git)

# "per-task commit" path
if has_fork:
    await push_to_fork_with_retry(git)

Reproduction

1. Create a bug ticket with forge:managed label
2. Let Forge generate RCA and plan
3. Approve the plan (set forge:plan-approved)
4. Worker reaches implement_bug_fix → crashes with the fork error

Versions

Introduced in commit 7ebb8b7 (PR #168), present on current main.