Root Cause Analysis Skill for Metric Anomalies

Where this fits

A growth analyst drops a dashboard screenshot in the team chat: revenue in the West region fell 22% last month, and nobody knows why. The agent is asked to “figure it out.” Without a protocol, it answers with generic possibilities. This skill gives the agent a diagnostic workflow for exactly this type of request—metric anomalies, KPI misses, revenue drops, and cost increases.

Why agents benefit

• Confirms the anomaly first. It uses run_python to load monthly revenue, profit, or order data and calculates change rates before forming any conclusion.
• Separates volume from price. The DuPont decomposition tells the agent whether the drop is driven by fewer orders, lower average item price, or smaller basket size.
• Ranks regions and channels. It aggregates target vs previous month, sorts by absolute change and rate, and surfaces the largest drags.
• Produces a standardized diagnosis. Every result ends in anomaly location, root cause, and recommended action—ready to hand to a human.
• Prevents speculation. The README constrains every conclusion to numbers produced by code, which reduces hallucinated business reasons.

Practical scenarios

Revenue drop after a pricing change

The agent finds order volume unchanged but average item price down. It attributes the drop to product-mix shift (more low-tier items sold), not demand loss, and recommends reviewing pricing architecture.

Regional KPI miss

The company misses its revenue target. The agent groups by region and finds one region contributes 90% of the gap. The diagnosis points to local competitive pressure and suggests a deeper regional drill-down.

Cost anomaly in operations

A cost line jumps 20% without a business explanation. The agent uses the same confirm-and-compare pattern: isolate the month, compare cost drivers by category, rank the largest increases, and identify whether the driver is volume, rate, or allocation changes.

Add it to your agent workflow

Install the skill into your agent runtime:

npx skills add zj-unicom-ai/UniEmployee --skill root-cause-analysis

Then invoke it from a prompt:

Use the root-cause-analysis skill to investigate why Q3 revenue in the West region declined 22% compared to Q2.

The skill's README contains the full protocol, so you don't have to re-engineer the diagnostic logic for every new agent.

Compare before adopting

Before adding this skill, compare it on four dimensions:

• Quality signals. Does the skill force real computation, or does it rely on the agent's internal guesses? This one requires run_python output as evidence.
• Maintenance freshness. Check the repository's commit activity and issue response rate. A stagnant skill can drift from current tool conventions.
• Alternatives. There are other root-cause and data-analysis skills. Compare whether this one is more specific to business metrics than generic debugging skills.
• Workflow fit. Is your agent already using run_python? If not, you will need to add that tool alongside the skill.

Why it is worth tracking

The project has 21 stars and a tight README—small but focused. The real signal is the skill's restraint: it does one job, defines strict steps, and builds in a “prove it with numbers” constraint. For agents that handle finance, operations, or product analytics, this is a low-risk skill to trial. Track its community momentum to see if the author ships updates and if other developers contribute scenarios. You can also follow the skill's OpenAgentSkill page at https://www.openagentskill.com/skills/zj-unicom-ai-root-cause-analysis for a maintained view of its metadata, alternatives, and recent changes.