FAQ
The objections that come up most, answered short. Each links to the page with the full story rather than repeating it.
Does it conflict with plan mode?
No — they do different jobs and compose. Plan mode helps you think inside one session; task-pipeline persists the result of that thinking to a file so it survives /clear and a fresh session tomorrow. Use plan mode to explore, then run /task:to-plan when the plan is worth keeping. See the comparison with default Claude Code.
What if I don't want it to commit?
Then it won't. The capture skills only ever write Markdown under .task/ — nothing is staged or committed until you explicitly tell a session to implement <path>. Even then it commits only the files it touched and never pushes, so nothing leaves your machine unless you push it yourself. Back out any working-tree change with plain git restore. Full detail: Why you can trust this.
Does it work in a monorepo or with git worktrees?
Yes. There's one .task/ per repo, and every worktree resolves the same one: through git config task.root, then an upward walk, then the main worktree root. With .task/ at the repo root — where setup puts it — nested and sibling worktrees find it with no setup at all. If yours lives elsewhere (a monorepo subdirectory, say), point the other worktrees at it by hand with git config --local task.root /path/that/contains/dot-task; running a capture skill from a worktree that can't see it would set up a second .task/ instead. See worktree can't find .task/.
What if I just read the file myself and ignore the Execution pointer?
That's fine — it's plain Markdown, not a runtime. The ## Execution pointer just sends a session to .task/CLAUDE.md → ## Executing a task, a standing instruction for a session that has no other context; nothing enforces it. You're free to read the ## Description and ## Plan and implement by hand, or hand the file to a session and let it follow the block. There's no hook and no hidden orchestration making you do either.
Does it work in languages other than English?
Yes. Descriptions and the dialogue follow .task/CLAUDE.md → Language, so you can write and discuss tasks in your own language. Only the format's fixed strings — section headers, commit trailers, and the ## Execution block — stay English, so the validator and any implementing session read them the same way. See Configuration.
How do I uninstall cleanly?
Two independent parts. Remove the artifacts with rm -rf .task — since .task/ was never tracked (it ignores itself through its own .task/.gitignore), the repo is left exactly as it was. Remove the plugin itself through /plugin (uninstall task@task-pipeline). If you want to erase the last trace, run git config --unset task.root. Projects set up by an earlier version may also carry a .task line in .git/info/exclude — harmless, and yours to delete by hand. See Why you can trust this.
→ Next: Troubleshooting — symptoms and fixes for the first run and the edge cases.