Agent Skills
› mohitagw15856/pm-claude-skills
› personal-wip-limits
personal-wip-limits
GitHub通过设定3个活跃结果上限、建立停车区及启动门禁，强制完成优先于开始。适用于任务堆积、多线并行却无进展的场景，利用两周吞吐量实验验证效率提升。
Trigger Scenarios
我有十二件事只做了一半，为什么什么都没完成
为我的工作设置WIP限制
我什么都开始做但什么都不做完
Install
npx skills add mohitagw15856/pm-claude-skills --skill personal-wip-limits -g -y
SKILL.md
Frontmatter
{
"name": "personal-wip-limits",
"description": "Cap your work-in-progress so things finish — the personal WIP limit (3 active outcomes, defended), the finish-before-start rule with its exceptions named, the parking lot for the overflow, and the throughput evidence that converts the skeptic. Use when asked I have twelve things half-done, why does nothing ever finish, set a WIP limit for my work, or I start everything and complete nothing. Produces the active-list cap, the parking protocol, the start-gate, and the two-week throughput experiment."
}
Personal WIP Limits Skill
Twelve half-done things is not twelve things in progress — it's twelve contexts paying switch-tax (context-switch-budget arithmetic), twelve stale re-entries, and a finish rate near zero, because starting is cheap and finishing is expensive and unlimited WIP always buys the cheap thing. The limit — three active outcomes, hard — forces the trade unlimited-WIP hides: starting the new means parking something, said out loud. Everything else supports the cap: the parking lot (where the other nine wait, honorably), the start-gate (the new enters only when something finishes or gets consciously demoted), and the two-week experiment that converts the internal skeptic with their own throughput data.
What This Skill Produces
- The active three — chosen from the current pile (task-triage-matrix verdicts feed this), each with its next action and finish line
- The parking lot — the demoted work, parked with a one-line state note so re-entry is cheap when its turn comes
- The start-gate — the rule at the door: finish, park-consciously, or decline (saying-no-kindly handles the outward face)
- The throughput experiment — two weeks at WIP-3, finishes counted, vs. the historical baseline
Required Inputs
Ask for these if not provided:
- The full in-progress pile — everything started-and-unfinished, honestly (the count is the diagnosis; most people find 8–15)
- The finish lines — per item, what done means (delegation-brief done-test grain); half the pile usually has no finish line, which is why it can't finish
- The involuntary WIP — the boss-assigned and the blocked-on-others; the cap counts what the user controls, and the blocked get their own lane (waiting ≠ active)
- The historical finish rate — roughly, what actually completed in the last month (the baseline the experiment beats)
Framework: The Limit Rules
- Three active, by outcome not task: the cap counts outcomes in progress ("the vendor decision," not its seven subtasks) — three is the level where each gets real weekly momentum (weekly-review-ritual big-three alignment is intentional: they're the same three). Four is the compromise that becomes seven by Friday.
- Parking is honorable, and stated: demotion to the lot is a decision, made out loud ("the newsletter redesign parks until the audit ships") with a one-line state note ("drafts in folder X, blocked on nothing, next: pick template") — so parked work re-enters cheaply instead of restarting from archaeology. The lot is reviewed at the weekly; it is not a graveyard with better PR (task-triage-matrix someday-file logic).
- The start-gate forces the trade: every new commitment faces the question "which of the three does this displace?" — finish one first (the default answer), park one consciously, or decline the new. The gate converts scope creep from ambient to visible; the displacement question is the entire mechanism.
- Blocked isn't active: work waiting on others moves to the waiting lane with its chase date (follow-up-chaser owns the cadence) — it doesn't occupy a WIP slot, but it doesn't vanish either. The slot opens; the chase continues.
- The experiment converts the skeptic: two weeks at WIP-3, finishes counted against the baseline — the standard result (more completions, not fewer, despite "doing less") is the counterintuitive evidence that makes the limit stick. The decision-journal pre-registration applies: predict the finish count before, review after.
Output Format
WIP Limit: [name] — active cap: 3
The Active Three
| Outcome | Finish line | Next action |
|---|
The Parking Lot
[Each: the one-line state note · reviewed at the weekly · the conscious-demotion statements made]
The Lanes
[Waiting-on-others: item × chase date · the involuntary (boss-assigned) noted honestly]
The Start-Gate + Experiment
[The displacement question, posted · the two-week finish-count vs. baseline: predicted [N], actual [ ]]
Quality Checks
- Active items are outcomes with finish lines, capped at three
- Every parked item has its state note and demotion was said aloud
- Blocked work lives in the waiting lane with chase dates, not in slots
- The start-gate question is installed at the door
- The experiment has a predicted number and a review date
Anti-Patterns
- Do not cap tasks instead of outcomes — thirty subtasks of one outcome is one slot, not thirty
- Do not park silently — unconscious demotion is just the old drift wearing a system's badge
- Do not let blocked work squat in active slots — waiting is a lane, not a job
- Do not negotiate the cap upward mid-crunch — the crunch is the argument for the cap
- Do not skip the experiment — the limit survives on its own throughput evidence or not at all
Version History
- 471c606 Current 2026-07-22 15:02
