protocol — Claude Code skill: proposals and installation
The protocol Claude Code skill turns a physician’s request and patient data into plan options with doses and predicted outcomes. It runs on a request for a specific patient’s plan or when visitprep/notes suggests a change.
MIT ★ 93 updated 2026-11-15
npx skills add Insilico-org/plsi --skill protocol -a claude-codeWhat protocol does
The skill reads the patient’s state and context, then gathers evidence for each option. If the request lacks a target or exclusions, it asks one question. When the physician names a drug class rather than a specific medication, the skill leaves the choice open for the physician instead of selecting a drug.
Each proposal includes an action, dose, duration, population to avoid it, evidence grades and predicted changes to recorded measures; a no-action control is included too. The result goes to work/<thread>/proposal-clinev.json for PLSI Nostradamus to screen. The skill is marked draft. It does not check, record or adopt a plan; adoption follows the physician’s card.
What it helps with
- turn a physician’s patient-specific request into plan options
- attach doses, durations and outcome predictions to options
- record evidence grades and verified identifiers
- send proposals to PLSI Nostradamus for screening
Example prompts for the agent
- Prepare plan options to lower ApoB for patient patient-07.
- Use the patient’s state and context to draft plan options.
visitprep/notessuggests a change to the patient’s plan; prepare options for screening.
How to install protocol
1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill protocol -a claude-code2Manually: put the files in place
~/.claude/skills/protocol/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/protocol/
1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill protocol -a codex2Manually: put the files in place
.agents/skills/protocol/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill protocol -a cursor2Manually: put the files in place
.cursor/skills/protocol/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Insilico-org/plsi skills/clinev/protocol2Manually: put the files in place
.github/skills/protocol/SKILL.md1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill protocol -a gemini-cli2Platform CLI
gemini skills install https://github.com/Insilico-org/plsi --path skills/clinev/protocolManually: put the files in place
.devin/skills/protocol/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about protocol
How do I install protocol in Claude Code?
Run npx skills add Insilico-org/plsi --skill protocol -a claude-code. The skill file is at ~/.claude/skills/protocol/SKILL.md.
What does protocol do?
It turns a physician’s request and patient data into plan proposals with doses, durations, evidence grades and predicted outcomes. PLSI Nostradamus screens the proposals; the skill does not approve the plan.
Can protocol draft a plan for the physician’s own health?
No. It handles requests about an individual patient. The physician’s own plans go through specialists and Nostradamus as usual.
Source text
SKILL.md · MIT
---
name: protocol
description: Use as PLSI Clinical Evidence when the physician wants a plan option for one patient - turn the physician's intent and the evidence into dosed, predicted proposals for PLSI Nostradamus, which writes, screens and records the plan.
---
# PLSI protocol proposal
**Status: draft**.
## When to use
- The physician asks that patient's ward bot for a protocol ("a plan to lower patient-07's ApoB"), or
`visitprep/notes` returned a plan change.
- Not for the physician's own health: their own plans go through the specialists and Nostradamus as usual.
## Steps
1. **Read** the patient's state and context:
```bash
plsi maps select --instance <label>
plsi state show --instance <label> --map <map_id>
plsi state show --instance <label> --map care_context
plsi state show --instance <label> --map personal_context
```
The physician's intent (target, options they want, options they exclude) comes from the request; missing parts go
in ONE `question`.
2. **Evidence** for each option, as `clinev/review` steps 2-3; reuse a review on the same thread.
3. **Propose** one action per option, with its dose, duration and the population that should avoid it, each claim
with its grade and verified identifiers, and predictions on recorded coordinates (expected change, interval,
horizon, basis) plus a no-action control. For a medication, name the agent the physician named (so
screening can match its label); when they named only a class (`<lipid-lowering option>`), put the choice in
the step's ONE `question`; never pick one yourself. Nostradamus screens them.
4. **Return one `result`** for `nostradamus/plan` (PLSI Nostradamus): write the proposals to
`work/<thread>/proposal-clinev.json` (`shared/team` rule 3) and put its path in `files`. Nostradamus merges them,
runs `screen` and `plan check`, PLSI hands the plan to `challenger/challenge`, and it is recorded and
adopted only after the physician's card.
## Rules
- You propose; you never check, record or adopt a plan, and nothing you write is an order.
- No option beyond what the physician asked for or the evidence supports; say plainly when the evidence is thin.
© Insilico-org · MIT · Insilico-org/plsi