---
name: hr-case-intake-triage
source: https://app.decimal.ai/s/hr-case-intake-triage@1/SKILL.md
source_sha256: 6876a9b125f5
---

# HR Case Intake Triage

An employee concern arrives — a message, a hallway remark written down, a form submission. Turn it into **one category**, **one urgency level**, and **one handling path**, and name the fact that set the path. A capable model can summarize the complaint, but by default it triages the way the *complainant framed it*: if the person says "I don't want to make a big deal of this, just keep it between us," the model routes a harassment or discrimination allegation to a quiet coaching chat — which is exactly the failure that exposes the employer. It also treats "serious" and "urgent" as the same axis, and it swings between calling every unhappy employee a harassment case and dismissing a protected-class complaint as ordinary friction.

## When to activate

Activate when the input is **an inbound employee complaint or concern** and the ask is how to classify or route it — "an employee reported that…", "how should we handle this concern", "what category and how urgent is this".

Do **not** activate to write the contemporaneous incident record for the file, to draft a PIP or a warning letter, or to answer a "what does our policy say about X" lookup.

## The two things that decide the path

Score every intake on **two independent axes**, then pick the path.

**Axis 1 — Category.** The subject matter of the concern:

| Category | Examples |
|---|---|
| **Harassment** | Unwanted sexual attention, slurs, demeaning conduct tied to who a person is |
| **Discrimination** | Adverse treatment tied to a protected class — race, sex, age, religion, disability, pregnancy, national origin, and the like |
| **Retaliation** | Adverse action *because* the person reported something, took leave, or raised a concern |
| **Safety / violence** | Threats, physical danger, weapons, a hazardous condition, self-harm |
| **Pay / hours** | Missing wages, unpaid overtime, misclassification |
| **Management / interpersonal** | Rudeness, favoritism, poor communication, a personality clash, workload |
| **Policy / benefits / ops** | A question about a rule, PTO, enrollment, a form |

**Axis 2 — Urgency.** How fast someone must act, set by *risk of harm*, not by category:

| Urgency | Meaning |
|---|---|
| **P0 — Immediate** | Risk of physical harm *now* — a violence threat, a weapon, a hazard, self-harm intent. Act same hour. |
| **P1 — Same-day** | Ongoing conduct still happening, or retaliation in progress. Open within the day. |
| **P2 — Standard** | A specific past allegation with no active danger. Normal intake clock. |
| **P3 — Routine** | An informational or policy/ops question. Route and answer. |

## The path — and the rule the default gets wrong

Pick **one** path:

- **MANDATORY INVESTIGATION** — a trained investigator must look into it.
- **COACHING / MANAGEMENT** — a manager or HR partner addresses it as a performance or interpersonal matter.
- **ROUTE / REFER** — send to the right desk (payroll, benefits, EAP, policy) and answer.

The rule the model skips: **any allegation of harassment, discrimination, retaliation, or a safety/violence threat is MANDATORY INVESTIGATION — and the complainant cannot waive it.** "Please keep this confidential," "I don't want anyone to get in trouble," "just talk to them for me" changes *how* you communicate and *how* you protect the reporter — it does **not** downgrade the path. The duty is the employer's, not the employee's to release. Discretion about method never becomes discretion about whether to investigate.

## The line between mandatory and coaching

This is where the default over- and under-triggers. Test each complaint against a protected class and against a legal trigger:

- **Directed at everyone → coaching. Directed at a protected class → mandatory.** "My manager yells at the whole team" is a management problem. "My manager only yells at the women" is sex discrimination — mandatory, even though the raw behavior sounds the same.
- **Generalized unfairness is not a protected-class claim.** "The schedule feels unfair," "he plays favorites," "I feel micromanaged" are coaching/management, not investigations — unless the favoritism tracks a protected class.
- **Mild wording does not lower the category.** "He keeps making little comments about my age" is an age-discrimination allegation, not a personality note. Grade the underlying claim, not how politely it was raised.
- **Retaliation is about the *sequence*.** If an adverse action (cut hours, a bad review, exclusion) followed the person raising a concern or taking protected leave, that timing itself is the trigger — investigate it as retaliation even if the first concern is old.

## Output

State four things, plainly:

- **Category** — one from the list above.
- **Urgency** — P0–P3.
- **Path** — mandatory-investigation / coaching / route.
- **Trigger** — the specific fact that set the path (e.g., "unwanted sexual conduct alleged", "adverse review followed her complaint", "comments target her national origin", "no protected class — general management issue").

Example — *A report comes in that a colleague repeatedly touches an employee's shoulders and remarks on her body, and she adds, "please don't make this official, I just need it to end."*
→ **Category: harassment · Urgency: P1 · Path: mandatory-investigation · Trigger:** alleged unwanted sexual conduct — a mandatory category the reporter's confidentiality request cannot waive; honor the request in *method* (limit who knows), not in *whether to investigate*.

## Edge cases

- **"Keep it confidential."** Protect the reporter and limit who learns of it — but a mandatory category still gets investigated. Confidentiality shapes the how, not the whether.
- **Multiple issues in one message.** Set the path on the **most serious** trigger present. "My manager is disorganized *and* made a comment about my pregnancy" is mandatory on the pregnancy line, whatever the rest is.
- **Anonymous or thin report.** A vague protected-class or safety allegation still opens an inquiry — you gather facts *to* assess it; thinness lowers confidence, not the duty.
- **Complainant only wants advice, not action.** For a mandatory category you can still support them and go at their pace on outreach, but the investigation obligation stands.
- **Pure policy or benefits question.** "How much PTO do I have," "how do I add a dependent" is route/refer at P3 — not every HR message is a case.
- **Self-harm or a violence threat.** P0 immediate regardless of category label — loop in safety/security and EAP the same hour, before the slower intake steps.