Agent Skills
› mohitagw15856/pm-claude-skills
› kpi-tracker-design
kpi-tracker-design
GitHub设计精简高效的KPI追踪器，聚焦5-9个关键指标。通过‘所以呢’测试筛选指标，建立趋势优先布局和基线目标设定规范，并配套定期审查仪式，确保数据驱动决策而非装饰，解决仪表盘无人使用的痛点。
Trigger Scenarios
为团队设置KPI追踪
在表格中构建指标仪表盘
确定应追踪的关键数字
现有仪表盘无人采取行动
Install
npx skills add mohitagw15856/pm-claude-skills --skill kpi-tracker-design -g -y
SKILL.md
Frontmatter
{
"name": "kpi-tracker-design",
"description": "Design a KPI tracker that drives decisions instead of decorating them — the few-metrics discipline (5–9, each with an owner and a so-what), targets with honest baselines, the trend-first layout, and the review ritual where the tracker actually gets used. Use when asked set up KPI tracking for the team, build a metrics dashboard in sheets, which numbers should we track, or our dashboard exists but nobody acts on it. Produces the metric selection with kill-list, the tracker structure, the target-setting notes, and the review ritual."
}
KPI Tracker Design Skill
KPI trackers fail by addition: every meeting adds a metric, none subtracts, and within two quarters the tracker is a wall of numbers nobody reads — motion without instrumentation. The working tracker is small (5–9 metrics, each with an owner and a written so-what), trend-first (a number without its history is a Rorschach test), honestly baselined before targets exist, and — the part that decides everything — attached to a ritual where someone reads it aloud and decisions reference it. The tracker is the artifact; the ritual is the product.
What This Skill Produces
- The metric selection — 5–9 survivors of the so-what test, each: definition, source, owner, cadence — plus the kill-list of metrics deliberately not tracked, with reasons
- The tracker structure — metric × period grid, trend columns/sparklines, target and threshold lines
- The target notes — baseline-first discipline: track before targeting; targets with owners and reasons, not round numbers
- The review ritual — where, when, who reads it, and the metric-in-every-decision norm
Required Inputs
Ask for these if not provided:
- The decisions the tracker should feed — metrics follow decisions; "what would we do differently if this number moved?" is the selection filter, and it needs the team's real decision list
- The candidate metrics and their sources — what's measurable today vs. requiring new instrumentation (day-one tracker uses today's sources; the wishlist is a separate roadmap)
- The audience and cadence — team-weekly vs. leadership-monthly are different trackers (grain, commentary, tone) — pick one; hybrids serve neither
- Existing baselines — history where it exists; where it doesn't, the no-targets-for-a-month rule applies
Framework: The Design Rules
- The so-what test selects: every candidate metric answers "if this moved 20%, what would we do?" — no answer, no slot. The kill-list (tracked-nowhere metrics with their reasons) is half the design's value: it's the documented defense against the wall-of-numbers drift.
- 5–9, owned, defined: each metric has one owner (who explains its movement, not who's blamed), a written definition (the numerator/denominator argument happens once, at design time — not monthly), and a stated source. Undefined metrics generate the two-truths meeting.
- Trend-first layout: every metric shows its last 6–12 periods (sparkline or mini-column), not just current-vs-target — direction and volatility are the actual information; a red 87% that's been climbing for four months is a different fact than a red 87% in freefall.
- Baseline before target: new metrics run target-less for 3–4 periods (you can't set honest targets on unmeasured behavior); then targets get set with a reason and an owner, at attention-worthy thresholds — the budget-tracker-design signal discipline applies: red must mean something.
- The ritual is the product: a standing slot (the weekly's first ten minutes) where the owner-of-the-moment reads the tracker aloud — one sentence per moved metric — and any proposal in the meeting gets the "which metric does this move?" question. Trackers without rituals decay into decoration in six weeks, on schedule.
Output Format
KPI Tracker: [team] — feeds: [the decisions]
The Metrics (5–9)
| Metric | Definition (num/denom) | Source | Owner | Cadence |
|---|
The Kill-List
[Deliberately untracked: metric → reason — the wall-of-numbers defense]
Structure & Targets
[Grid + trend columns · baseline-first schedule for new metrics · targets with owner + reason + threshold]
The Ritual
[The slot · the read-aloud norm (one sentence per mover) · the which-metric-does-this-move question · quarterly metric review: add/kill with the so-what test]
Quality Checks
- Every metric passed the so-what test against a real decision
- Count is 5–9; the kill-list exists with reasons
- Definitions settle the numerator/denominator argument in writing
- Trends are visible per metric — no naked current-values
- The ritual has a slot, a reader, and the decision-linkage norm
Anti-Patterns
- Do not track what you can't act on — interesting is not a criterion; decidable is
- Do not exceed nine — the tenth metric costs attention from the first nine
- Do not set targets on unbaselined metrics — round-number targets on unknown behavior are theater
- Do not show numbers without trends — a value with no history is a mood
- Do not build the tracker without booking the ritual — undecorated walls beat decorated ones
Version History
- 471c606 Current 2026-07-22 14:04
