Guide · Agent skill
How to use the Clinical Decision-Support Research Evaluator agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
This guide explains the clinical-decision-support agent skill from the K-Dense-AI scientific-agent-skills collection and how to install it. It is for research teams, methodologists and privacy or governance reviewers who need structured documents about a clinical prediction or biomarker evaluation. By the end you will know what the agent can draft, how to trigger it, and where its limits are.
Clinical Decision-Support Research EvaluatorK-Dense-AI/scientific-agent-skills · MIT Download the skill Quick start
- 1Download and unzip. Download the skill ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
- 2Or upload in the Claude apps. In the Claude apps, open Settings → Capabilities → Skills and upload the skill.
- 3Confirm Python 3.11+. Check that Python 3.11 or newer is installed. The scripts need no extra packages.
- 4Confirm data is synthetic or aggregate. Have a responsible human reviewer confirm and document this before any script runs.
- 5Make a natural request. Ask for a named artifact, such as a survival analysis plan or GRADE evidence profile, and state the research purpose.
- 6Run a helper script locally. Run the matching script from the skill directory, using --help first, and write output only to a reviewed local folder.
- 7Have humans review. Route the draft to the methodologist, domain expert, privacy officer or governance owner as relevant before any use.
What does the skill make the agent do?
Once installed, the agent can help prepare structured research and governance artifacts about clinical decision-support tools. The source lists intended-use and limitation statements, GRADE evidence-profile checklists, aggregate cohort table shells, survival analysis plan review, aggregate model or biomarker performance evaluation, evidence and decision-logic traceability matrices, and de-identification process checklists.
The skill is built around a hard boundary. It is for research, evaluation, documentation and governance only. It does not diagnose or classify a person, recommend or modify treatment, calculate a patient-specific dose, triage, or support live clinical operation. Every output is a draft that qualified humans must review and approve.
- Intended-use statements with limitations and prohibited uses
- GRADE evidence-profile checklists (human panel judgments, not automated scores)
- Aggregate cohort table shells with disclosure controls
- Survival analysis plans, reviewed against pre-specified elements
- Aggregate model or biomarker summaries from confusion counts and calibration bins
- Traceability matrices and de-identification process checklists
Install the skill (ZIP download or Claude apps)
Download the skill ZIP from this page. To install it for your user account, unzip it into ~/.claude/skills/. To install it for a single project, unzip it into that project's .claude/skills/ folder.
Alternatively, in the Claude apps you can upload the skill under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload
Check the requirements
The bundled helper scripts need Python 3.11 or newer. They use only the Python standard library. The compatibility note says no API keys, accounts, network access, credentials or LLM services are needed to run them.
The helper scripts run locally. Run the help commands from the skill directory to confirm they start. The skill's own metadata lists the license as MIT.
- Python 3.11+
- Standard library only (no packages to install)
- No API keys, accounts or network access required
python3 scripts/validate_cds_artifact.py --helpTrigger the skill with natural requests
You do not need special syntax. Describe the research document you need and name the artifact type. Ask for a draft, not a clinical answer. Good requests name the research purpose, the population level (aggregate or synthetic) and the artifact.
Before any script runs, the skill requires a responsible human reviewer to confirm and document that input is synthetic or aggregate. The agent will reject patient rows, identifiers, free text, dates tied to people, images, waveforms or genomic sequences.
- "Draft an intended-use statement for our aggregate sepsis risk model evaluation, marked as not for patient care."
- "Build a GRADE evidence profile checklist for these three studies' outcome of mortality, with source IDs for each judgment."
- "Set up a survival analysis plan template for a synthetic cohort study with pre-specified time origin and horizon."
- "Create an aggregate cohort table shell with disclosure thresholds set before counts are shown."
Run the helper scripts locally
The skill maps each artifact type to a template and a script. Templates live in the assets folder and the scripts check that declared fields and internal consistency rules are present. Write outputs only to a reviewed local directory.
A script passing means only that its declared fields and consistency checks passed. It is not a privacy, quality or compliance determination.
- Intended-use review: assets/artifact_intended_use_template.json with scripts/validate_cds_artifact.py
- GRADE evidence profile: assets/evidence_profile_template.json with scripts/evidence_profile_check.py
- Model or biomarker evaluation: scripts/model_biomarker_evaluation.py
- Cohort table: scripts/cohort_table_generator.py
- Survival plan: scripts/survival_plan_validator.py
- Traceability: scripts/decision_logic_traceability.py
- De-identification review: scripts/deidentification_checklist.py
python3 scripts/survival_plan_validator.py --helpSafety notes on scripts and network use
The scripts run locally and the source states they do not fetch URLs, call APIs, read environment variables or send data to a model. The scripts cap file size, groups, rows and text length, and they reject URL-like paths and common row-level keys. The source describes these as controls that reduce accidental misuse, not as a privacy determination.
Do not place generated reports in an EHR, alerting system, clinical portal or device workflow. Do not use the skill to claim FDA authorization, HIPAA compliance or legal compliance.
- Scripts run locally; network use: no
- Keep source files local and aggregate or synthetic
- Route privacy, regulatory and legal decisions to the named qualified reviewers
What people use it for
- Documenting an aggregate clinical prediction model evaluation before peer review or internal governance review
- Drafting a survival analysis plan with pre-specified outcomes, time origin and horizon
- Preparing a GRADE evidence-profile checklist with source IDs for each judgment
- Writing a de-identification process checklist for a privacy reviewer
- Building traceability matrices that link research and governance rules to their sources and owners
Frequently asked questions
How do I install the clinical decision-support agent skill?
Download the skill ZIP from this page and unzip it into ~/.claude/skills/ for your user account, or into a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
What do I need to run it?
Python 3.11 or newer. The bundled helper scripts use only the standard library, and the compatibility note says no API keys, accounts, network access or LLM services are needed.
Is the skill free and what is its licence?
The skill's metadata lists the licence as MIT. The source does not state pricing, so check the listing on this page for any cost information.
Can I use it to make decisions about a patient?
No. The skill explicitly prohibits patient-specific diagnosis, treatment, dosing, triage and live clinical use, and it routes such matters to a licensed healthcare professional.
Does the skill check that my data is de-identified or HIPAA compliant?
No. The scripts check declared fields and internal consistency only. The skill states its outputs are not privacy, HIPAA, regulatory or legal compliance determinations, and a qualified privacy expert must make disclosure decisions.
How does it compare with other approaches to clinical evidence documentation?
The source does not compare it with other tools. Its distinguishing features are the research-only boundary, the local standard-library scripts, and templates for GRADE profiles, survival plans and traceability matrices.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits research teams, methodologists and privacy reviewers who want structured, human-reviewed drafts of aggregate or synthetic clinical evaluation documents. It is not suitable for raw patient data, patient-specific decisions, or any claim of regulatory or legal compliance, so choose a validated clinical system or a qualified expert for those needs.
Download the skill