Meta ml screener

Skill Vambrocop/EvidenceForge/skills/meta-ml-screener

Agent skills for systematic review, meta-analysis, umbrella review, and AI-assisted evidence synthesis.

Install

npx -y skills add Vambrocop/EvidenceForge --skill meta-ml-screener

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Designs machine-learning assisted systematic review workflows. Use for active-learning screening, deduplication, study classification, LLM-assisted extraction, risk-of-bias triage, topic modeling, moderator discovery, and audit logs, while preserving human verification and transparent evidence-synthesis decisions.

SKILL.md

2.7 KB, as published. Nobody here has run it

Meta ML Screener

Use this skill when machine learning will assist a systematic review or meta-analysis.

Core Principle

ML can reduce workload, prioritize records, extract candidates, and explore heterogeneity. It should not hide eligibility criteria, final inclusion decisions, or effect-size verification.

Intake

Identify the ML task:

deduplication;
title/abstract screening;
full-text triage;
study design classification;
outcome classification;
extraction assistance;
risk-of-bias triage;
topic modeling;
moderator discovery.

Identify:

labeled data available;
human verification plan;
recall requirement;
audit log format;
software or platform;
whether LLMs are used.

Load references/ml-assisted-review.md for task-specific guidance.

Workflow

Define ML role and what decisions remain human.
Create seed labels or validation set.
Define features, model, or prompt schema.
Run prioritization/classification/extraction.
Record scores, labels, and decisions.
Validate recall or extraction accuracy.
Escalate uncertain records to human review.
Export decisions, model scores, prompts/schemas, and human adjudication.
Report ML use transparently.

Use:

templates/screening-log.md for a human-readable log.
templates/screening-log-schema.csv for machine-readable logging.
templates/example-screening-log.csv for a minimal example.
scripts/validate_screening_log.py to check required fields, exclusion reasons, duplicate record IDs, and human follow-up decisions.

Output Modes

ML Screening Plan

ML task:
Human decision point:
Training/seed labels:
Validation metric:
Audit log:
Stopping rule:
Failure modes:
Reporting sentence:

Extraction Schema

Field:
Definition:
Source anchor:
Confidence:
Human verification:

Guardrails

Do not exclude records solely because the model is confident unless the protocol explicitly allows it and recall is validated.
Do not use LLM-extracted numbers without source anchors and verification.
Do not treat ML-discovered moderators as confirmatory.
Do not hide prompt/model/version details if they affect review decisions.
Do not let automation erase exclusion reasons or reviewer accountability.
Do not accept a screening log as auditable unless human decisions and exclusion reasons are recorded.