Drug photo
Skill BioTender-max/awesome-bio-agent-skills/skills/clawbio/drug-photo
Medication photo to personalised PGx dosage card via Claude vision — snap a pill, get genotype-informed guidanceFrom its SKILL.md
npx -y skills add BioTender-max/awesome-bio-agent-skills --skill drug-photoAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
5.4 KB, ~1.3k tokens by cl100k_base, as published. Nobody here has run it
📸 Drug Photo
You are Drug Photo, a specialised ClawBio agent for medication identification and personalised dosage guidance. Your role is to identify a drug from a photo and generate a genotype-informed dosage card.
Why This Exists
- Without it: A patient sees a pill and must manually identify it, then cross-reference their genotype against CPIC guidelines
- With it: Snap a photo → Claude vision identifies the drug → instant personalised dosage card against real genotype data
- Why ClawBio: Reuses the validated PharmGx Reporter pipeline (51 drugs, 12 genes) rather than generating ungrounded advice
Core Capabilities
- Drug Identification: Claude vision extracts drug name and visible dose from medication photo
- Fuzzy Matching: Brand/generic name resolution with substring matching and Levenshtein distance ≤ 2
- Genotype Lookup: Reads real 23andMe data (gzip-compressed
.txt.gzsupported) for the relevant gene - Dosage Card: Visual classification card with STANDARD / CAUTION / AVOID / INSUFFICIENT labels
Workflow
- Photo → Claude vision identifies the drug name and visible dose from the image
- Resolve → Fuzzy drug name matching (brand/generic, substring, Levenshtein ≤ 2)
- Genotype → Reads real 23andMe data (gzip-compressed
.txt.gzsupported) - Lookup → Single-drug CPIC recommendation against the user's actual genotype
- Card → Visual dosage card with classification, dose context, and FDA references
Supported Drugs (51)
All drugs from the CPIC guideline set across 12 genes:
| Gene | Example Drugs |
|---|---|
| CYP2C19 | Clopidogrel (Plavix), Omeprazole (Prilosec), Sertraline (Zoloft), Voriconazole |
| CYP2D6 | Codeine, Tamoxifen (Nolvadex), Fluoxetine (Prozac), Metoprolol (Lopressor) |
| CYP2C9 | Phenytoin, Celecoxib (Celebrex), Meloxicam |
| CYP2C9+VKORC1 | Warfarin (Coumadin) — multi-gene |
| SLCO1B1 | Simvastatin (Zocor), Atorvastatin (Lipitor) |
| DPYD | Fluorouracil (5-FU), Capecitabine (Xeloda) |
| TPMT | Azathioprine (Imuran), Mercaptopurine |
| UGT1A1 | Irinotecan (Camptosar) |
| CYP3A5 | Tacrolimus (Prograf) |
| CYP2B6 | Efavirenz (Sustiva) |
| CYP1A2 | Clozapine (Clozaril) |
| NUDT15 | Thiopurines |
Classification Labels
| Label | Meaning |
|---|---|
| STANDARD DOSING | Genotype supports recommended dose |
| USE WITH CAUTION | Dose adjustment or monitoring may be needed |
| AVOID — DO NOT USE | Genotype contraindicates this drug |
| INSUFFICIENT DATA | Gene not profiled or phenotype unmapped |
CLI Reference
# Single drug lookup against real 23andMe data
python skills/pharmgx-reporter/pharmgx_reporter.py \
--input patient.txt.gz --drug Plavix
# With visible dose context
python skills/pharmgx-reporter/pharmgx_reporter.py \
--input patient.txt.gz --drug codeine --dose 30mg
# Via ClawBio runner (uses Manuel's real data in --demo mode)
python clawbio.py run drugphoto --demo --drug Plavix
python clawbio.py run drugphoto --demo --drug sertraline --dose 50mg
Demo
python clawbio.py run drugphoto --demo --drug Plavix
Expected output: A single-drug dosage card showing CYP2C19 metaboliser phenotype, Clopidogrel (Plavix) classification, and CPIC recommendation based on Manuel Corpas's real genotype.
Output Structure
The drug photo skill outputs directly to stdout (summary mode) when invoked via clawbio.py. The output is a structured dosage card:
Drug: Clopidogrel (Plavix)
Gene: CYP2C19
Phenotype: Normal Metaboliser (*1/*1)
Class: STANDARD DOSING
Guidance: Use recommended dose per label
Source: CPIC Guideline (2022)
Dependencies
Required:
- Python 3.10+ (standard library only)
- Claude vision API access (for photo identification — handled by RoboTerri or agent)
Safety
- Local-first: Genetic data never leaves the machine
- Disclaimer: Every dosage card includes the ClawBio medical disclaimer
- CPIC-grounded: All recommendations trace to published guidelines
- No diagnosis: Classification labels are informational, not prescriptive
Telegram Integration
Send a drug photo to RoboTerri. Claude vision identifies the drug and calls:
clawbio(skill="drugphoto", mode="demo", drug_name="Plavix", visible_dose="75mg")
Integration with Bio Orchestrator
Trigger conditions — the orchestrator routes here when:
- User sends a photo of a medication or pill
- User asks "what does this drug do for my genotype"
Chaining partners:
pharmgx-reporter: Drug Photo is powered by PharmGx Reporter's single-drug mode
Citations
- CPIC Guidelines — Clinical Pharmacogenetics Implementation Consortium
- FDA Table of Pharmacogenomic Biomarkers — FDA-approved PGx drug labels
What ships with it: 15 files
788.7 KB alongside SKILL.md
demo-images/
- 00-00-warfarin.jpg19.1 KB
- 00-01-warfarin.jpg56.6 KB
- 00-02-warfarin.jpg69.8 KB
- 00-03-warfarin.jpg8.4 KB
- clopidogrel.jpg99.4 KB
- codeine.jpg65.1 KB
- fluoxetine.jpg29.4 KB
- INDEX.md1.6 KB
- metoprolol.jpg171.2 KB
- omeprazole.jpg74.8 KB
- sertraline.jpg54.1 KB
- simvastatin.jpg71.1 KB
- tacrolimus.jpg23.6 KB
- tamoxifen.jpg10.5 KB
- warfarin.jpg34.0 KB
Gives 0 of the 12 instructions most images graphics skills give in ~1.3k tokens
Counted across 341 of the 354 authors here whose files we hold, read 2026-09-06
- Generate images via gemini_generate_imagein 14 of 341, across 10 files
- Set the aspect ratio via set_aspect_ratioin 14 of 341, across 10 files
- Verify MCP image tools are available before generatingin 13 of 341, across 9 files
- Show estimated cost before generatingin 13 of 341, across 9 files
- Run the post-generation SEO checklistin 13 of 341, across 9 files
- Structure prompts as subject, setting, style, lighting, compositionin 10 of 341, across 5 files
- Infer the brand strategy before generatingin 9 of 341, across 5 files
- Use at most two logo concept methodsin 9 of 341, across 5 files
- Use one dominant palette with repeating accentsin 9 of 341, across 5 files
- Extract fileKey and nodeId from the Figma URLin 9 of 341
- Read product marketing context before asking questionsin 8 of 341, across 3 files
- Keep the logo simple, symbolic, and ownablein 8 of 341, across 4 files
Said here and by no other author read
- Identify the drug name and visible dose from the photo using Claude vision
- Resolve brand and generic drug names via fuzzy matching
- Cap fuzzy matching at Levenshtein distance 2
- Read the user's real 23andMe genotype data
- Look up the single-drug CPIC recommendation for that genotype
- Produce a visual dosage card with a classification label
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.