fix(process): pin process-state writes to canonical BotRoot, not worktree

[emrahzunicaplab]

Linked issue

Closes #612

Summary of changes

Write-ProcessFile and Test-ProcessStopSignal resolve BotRoot via $PWD
whenever no -BotRoot is passed. During task execution,
Invoke-WorkflowProcess.ps1 Push-Locations into the task's worktree, whose
.bot/.control is a junction back to the canonical .control dir. Worktree
teardown (Complete-TaskWorktree, Reset-TaskWorktree, Remove-OrphanWorktrees)
removes that junction before removing the worktree itself, so a heartbeat write
racing the teardown resolves to a path that no longer exists and silently drops
after exhausting retries (Could not find a part of the path '…\worktrees\1\task-…\.bot\.control\processes\proc-….json.tmp').

Changes:

• Pin every process-registry write inside the worktree-scoped execution window
  to $botRoot — the canonical root captured once at process start and never
  reassigned. Covers all Write-ProcessFile / Test-ProcessStopSignal calls
  between the worktree Push-Location and its Pop-Location.
• Thread the same canonical root through Invoke-TaskClarificationLoopIfPresent
  via a new -ProcessBotRoot parameter, kept deliberately separate from its
  existing worktree-scoped -BotRoot (which is still used for the answers-file
  path).
• Add a processes/ directory guard to Write-ProcessFile, mirroring
  Write-ProcessActivity, so a missing directory self-heals instead of failing
  outright.

Process writes outside the worktree window (CWD = project root) are intentionally
left unpinned — they already resolve to the canonical root. Sibling runners
(Invoke-PromptProcess.ps1, Invoke-DotbotProcess.ps1) never Push-Location
into a worktree, so they are unaffected. Activity-log writes go through
Write-BotLog, which is init-pinned to the canonical control dir at startup, so
they were never vulnerable to this race.

Testing notes

• pwsh parse check on both changed files — clean.
• tests/Test-ProcessRegistry.ps1 — 20/20 pass.
• tests/Test-Compilation.ps1 — 372 pass, 0 fail.
• Reproduced the exact Write-ProcessFile targets transient task worktree for global process state → fails during worktree churn #612 scenario in a sandbox with a real .control junction:
  a BotRoot-pinned write reaches the canonical processes/ dir and survives
  junction teardown, the directory guard recreates a deleted processes/, while
  an unpinned write (the old behaviour) resolves to the now-orphaned worktree path
  and never reaches canonical — confirming both the bug and the fix.

Checklist

• Linked issue exists
• Follows the contribution guide

[Copilot]

Pull request overview

This PR fixes a race where workflow process-state/heartbeat writes could resolve through a task worktree’s transient .bot/.control junction (when Write-ProcessFile/Test-ProcessStopSignal fall back to $PWD), causing state updates to be dropped during worktree teardown. It pins process-registry reads/writes to the canonical project bot root during the worktree-scoped execution window, and hardens Write-ProcessFile by ensuring the processes/ directory exists before writing.

Changes:

• Pin Write-ProcessFile / Test-ProcessStopSignal calls inside the worktree Push-Location→Pop-Location window to the canonical $botRoot.
• Thread the canonical root into the task clarification loop via a new -ProcessBotRoot parameter (kept separate from the existing worktree-scoped -BotRoot used for answers pathing).
• Add a processes/ directory existence guard to Write-ProcessFile to self-heal missing directories.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/runtime/Scripts/Invoke-WorkflowProcess.ps1	Pins process stop-signal checks and process-state writes to the canonical bot root during worktree execution; threads canonical root into the clarification loop to avoid resolving via the transient worktree junction.
src/runtime/Modules/Dotbot.Process/Dotbot.Process.psm1	Ensures processes/ exists before writing process-state files, preventing failures when the directory is missing.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[carlospedreira]

Looks good. I reviewed the process-registry/worktree-rooting changes and ran focused validation against the PR commit: parser check on the changed PowerShell files plus tests/Test-ProcessRegistry.ps1 (20/20 passing). I don’t see a merge blocker.