goal-mode-looplisted
Install: claude install-skill MLliu6/lMl_skills
# goal-mode-loop
**English** | [中文](#中文)
A small operating discipline for driving a long-running development conversation toward
a single, well-defined goal — without drifting, without declaring victory too early, and
without losing the thread when the context window fills up.
It is **generic project-management scaffolding**: define the goal, state measurable
success criteria, work in verifiable stages, checkpoint to a file, and stop deliberately.
No special review or analysis technique is implied — just plain, community-standard
"decide the target, make progress, prove it, write it down."
## When to use it
- A task is large enough that it needs **more than a few steps** and will span a long
conversation (or several).
- You want the assistant to **keep going on its own** between check-ins, but stay on
target and not overreach.
- The context window may fill up or the session may restart, and you don't want to lose
where things stand.
If the task is a quick one-off, you don't need this — just do it.
## How to run the loop
1. **Set the goal.** Write one sentence: the outcome, not the activity. "Ship X so that
Y works" — not "work on X".
2. **Write success criteria you can check.** Each criterion must be observable: a command
that exits 0, a test that passes, a number under a threshold, a file that exists with
the right content. If you cannot check it, rewrite it until you can. Vague criteria are
the main reason goal-mode runs drift.
3. **Create the chec