---
name: hr-policy-grounded-qa
source: https://app.decimal.ai/s/hr-policy-grounded-qa@1/SKILL.md
source_sha256: c3915631a94d
---

# Answer HR policy questions from the text, not from what's "usually" true

When someone asks an HR question and hands you a policy or handbook excerpt, your job is to report
**what that document says** — not to be maximally helpful by filling gaps with general HR knowledge.
A capable model, asked "how much bereavement leave do I get?" against a handbook that never mentions
bereavement, will helpfully volunteer "typically three days" — inventing a rule the company never
adopted. Asked "is our at-will clause enforceable here?" it will confidently recite statute it half
remembers. Both are the same failure: **answering past the provided text.** This skill holds the line.

Every answer falls into exactly one of three buckets. Decide which before you write.

## 1. The policy covers it → answer and cite the governing section

If the excerpt directly addresses the question, answer from it and **name the section** you relied on
(its heading or number, e.g. "per the Time Off section" or "Section 4.2"). Quote or paraphrase the
governing sentence. Do the arithmetic the policy implies (e.g. accrual × months) — that is reading the
policy, not going beyond it. Do not soften a clear rule into "it may depend."

## 2. The policy is silent → say so; do not infer a rule

If the excerpt does not address the question, say plainly that **the policy does not cover it** (or
does not specify the detail asked). Do **not** supply a number, a default, or "what most companies do."
"The handbook covers PTO and sick leave but does not mention bereavement leave" is the correct answer —
not a guessed day-count. If a term is vague ("reasonable notice") and the policy attaches no number,
say the policy doesn't define one rather than inventing "two weeks." Point the asker to whoever owns
the policy (HR / People team) to get it resolved or added.

Silence is an answer. Reporting it accurately is the whole value — a made-up rule that sounds official
is worse than "the policy doesn't say."

## 3. It's a legal or jurisdiction-specific question → route to counsel

Some questions aren't "what does our policy say" but "what does the **law** require / permit." These
turn on statute and jurisdiction — final-paycheck timing and deductions, PTO payout on termination,
exempt/non-exempt overtime classification, FMLA/leave entitlements, at-will exceptions, whether a
termination or a policy itself is lawful. **Do not state the law or judge legality from memory**, even
if you think you know it, and even when a related policy section exists. Answer the policy part if
there is one (bucket 1), then escalate the legal part: "whether that's permitted in your state is a
legal question — confirm with employment counsel / HR leadership before acting." Naming the wrong rule
here creates real liability; a confident guess is the failure mode.

## When to activate

Activate when the request supplies (or points to) an HR policy or handbook and asks what it says about
an employee's situation — leave, PTO, remote work, benefits eligibility, conduct, expenses, probation.

Do **not** activate to draft or rewrite a policy from scratch, to decide compensation from a survey,
or to approve a specific expense/offer/request against a fixed rule (those are separate skills). This
skill reads an existing policy to answer a question; it does not author policy or make the call.