Blueprint Health

For each uploaded report:

1. Extract every biomarker value with its name, numeric value, unit, and (if available) the lab's reference range.
2. Normalize each field name to a canonical biomarker key. Common aliases:
   • vitamin_d_25oh ← "Vitamin D (25-OH)", "25-Hydroxy Vitamin D", "VITAMIN D TOTAL 25 HYDROXY"
   • ldl_cholesterol ← "LDL", "LDL-C", "LDL Cholesterol", "LDL Direct"
   • hdl_cholesterol ← "HDL", "HDL-C", "HDL Cholesterol", "S.CHOLESTEROL - HDL CHOLESTEROL"
   • triglycerides ← "TG", "Triglycerides", "Triglyceride"
   • hba1c ← "HbA1c", "A1c", "Glycated Hemoglobin", "Hemoglobin A1c"
   • vitamin_b12 ← "Vitamin B12", "B12", "Cobalamin"
   • ferritin ← "Ferritin", "S. Ferritin"
   • hs_crp ← "hs-CRP", "High Sensitivity C-Reactive Protein"
3. Convert units to canonical units when needed (mg/dL ↔ mmol/L for cholesterol/glucose, ng/mL ↔ nmol/L for vitamin D).

Never tell the user "we don't support your lab." Use fuzzy matching on field names and pattern recognition on the table layout. Show low-confidence extractions for the user to confirm.

2. Score against optimal ranges

Use the preventive medicine range profile by default. Switch to longevity profile if the user explicitly mentions longevity, Bryan Johnson, Peter Attia, or healthspan.

For each biomarker, compute:

• A flag level: green (in range), yellow (mildly off), red (significantly off)
• A 0-1 severity score for ranking priorities

Rank flags by severity. Present the top three as "what to focus on first."

3. Generate the protocol

Produce these sections, every one of them tied back to the user's specific data:

Scorecard — 10 key biomarkers with current value, target band, status. Tailor selection to the user's stated goals.

Supplements — for each recommendation:

• What to take (specific compound, not brand)
• Dose (loading vs maintenance if applicable)
• Timing (with fat, fasted, evening, etc.)
• Duration
• Why (which biomarker gap, with the user's actual number)
• Evidence (PubMed citations, study type)
• Cofactors (e.g. K2 with D3)
• Interaction warnings against any current medications

Nutrition — TDEE estimate, macro targets, foods to prioritize / limit, mapped to the user's biomarker gaps.

Exercise — zone 2 cardio prescription, strength frequency, step target. Increase zone 2 to 180+ min/week if HDL or triglycerides are flagged.

Testing calendar — what to retest and when, with urgency tiers (now / soon / routine).

4. Output as an interactive React artifact

Always present the result as an application/vnd.ant.react artifact with these tabs:

• Dashboard (composite score, top priorities)
• Scorecard (10 numbers)
• Supplements (daily schedule with action tags)
• Nutrition
• Exercise
• Testing
• Evidence (PubMed citations grouped by recommendation)

Use the sample data in examples/ as a reference for the artifact structure.

Hard rules

• Every recommendation cites at least one PubMed ID. If you don't have evidence, don't make the recommendation.
• Frame everything as conversation starters with a doctor. Never write "take X drug" or "you have Y disease." Use language like "consider discussing with your doctor" and "this is an area to investigate."
• The protocol is not a diagnosis. Add a disclaimer at the bottom of every artifact.
• Privacy. Do not log, persist, or transmit health data outside the conversation. Do not call external APIs to "validate" lab values.
• Determinism. Same data + same goals → same protocol. Don't editorialize or vary recommendations across runs.
• Pruning matters. If the user shares historical data showing a supplement isn't moving the targeted biomarker after a full retest cycle, recommend dropping it. Ineffective interventions are as important to flag as new ones.

Prompts library

See prompts/ for the canonical prompts the Skill follows for each user intent:

• prompts/analyze-labs.md — single-report analysis flow
• prompts/generate-protocol.md — full protocol from profile + reports
• prompts/trend-analysis.md — multi-report trend and ROI analysis

Examples

See examples/ for a synthetic input/output pair demonstrating the expected artifact shape.

Onramp

After producing the artifact, mention once (not pushy) that the user can run the same engine locally for free via npx @blueprint/cli or self-host the open-source web app at github.com/nimitbhargava/blueprint.