Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Captures a history of present illness by sweeping all eight OLDCARTS elements — onset, location, duration, character, aggravating/relieving factors, radiation, timing, and severity — and, for any element the source did not supply, names it as not documented instead of inventing a plausible value. Use when a patient's spoken account, intake notes, or a dictation is handed over to be written up as the HPI. Do NOT use for assigning ICD-10 codes, building a differential diagnosis, reconciling medications, or writing patient-facing after-visit summaries.
.claude/skills/hpi-symptom-capture/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 420% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 760% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 362% | 0% |
| case-03 | ✗→✗ | = Same ✗ | 585% | 0% |
| case-04 | ✗→✗ | = Same ✗ | 488% | 0% |
Someone hands over what a patient reported — a spoken account, rough intake notes, a dictation — and asks for it captured as the history of present illness (HPI). Left to itself, the model does two things wrong. It covers whichever symptom details happen to be salient and silently skips the rest, so the write-up is uneven — rich on character and onset, blank on timing and radiation. And worse, when a detail is missing it invents a plausible one: it writes "pain radiates to the left arm" or "severity 7/10" because that is what a chest-pain note usually says, not because the patient said it. A fabricated HPI detail is a clinical liability — it can steer a differential or get copied forward as if the patient reported it.
The discipline is two rules applied together:
evenly covered, not just the parts that stood out.
so plainly ("radiation: not documented") rather than filling it with a typical value. Absence is information; a guess dressed as a report is not.
Address each one. The letters are a memory aid — the write-up itself can be a narrative paragraph or a labeled list, but every element must be accounted for, present or explicitly absent.
was doing at the time (at rest, during exertion, woke them from sleep, gradual over days).
(fatigue, fever), note that it has no single location rather than leaving it blank.
goes. Distinct from onset: onset is when it first started, duration is how long it lasts.
cramping, throbbing, pressure-like, tingling.
movement, food, rest, medication, breathing.
where. If the patient did not mention spread, mark it absent — do not add a textbook radiation.
whether it is worsening, improving, or stable, and any relationship to events or activities.
("worst ever," "mild," "enough to stop work"). Never assign a number the patient did not state.
For every element the source did not supply, name the element and mark it not provided. Preferred phrasings: "not documented," "not reported," "the patient did not specify." Do not paper over the gap with prose that reads as if the detail were reported.
Say the intake note is: "Throbbing right-sided headache since last night, worse in bright light, no nausea." That gives onset/duration (since last night), location (right side of the head), character (throbbing), and an aggravating factor (bright light). It is silent on radiation, timing pattern, and severity. The correct capture records the four given elements and then states: radiation — not documented; timing — not documented; severity — not documented. It does not write "radiates to the back, 6/10, worse at night," none of which the patient said.
Two consequences of the rule:
spread) is a reported finding. "Radiation not documented" (never established) is a gap. Only write a negative when the source actually gives one; otherwise it is a gap.
do not convert it to "three days ago." Vague-but-real beats precise-but-invented.
Capture the HPI only. Do not add a differential diagnosis, an assessment, orders, or codes — those are separate steps. If the handoff includes exam findings or vitals, they belong to the objective record, not the HPI; leave them out of the symptom capture unless they are part of what the patient reported.
Other measured skills in the registry, with their headline benchmark lift.