protocol-writer_skill

Overview

This skill writes an executable, auditable systematic review protocol into output/PROTOCOL.md focused on clearly reproducible searches, screening, and extraction rules. It enforces operational criteria, concrete queries per source, a screening log, and an explicit HUMAN approval gate before any screening begins. The protocol is authored from the perspective of a methodologist and includes an auditor checklist to prevent silent drift.

How this skill works

The skill translates scope notes into 1–3 testable review questions and enumerates in-scope vs out-of-scope items. It lists databases and manual routes, then produces copy/paste executable search strings with time windows and search dates. It then emits operational inclusion/exclusion rules, a staged screening plan with labels and tie-break rules, a fully specified extraction schema (definitions, allowed values, unknowns), and a risk-of-bias plan.

When to use it

• At the start of a systematic review pipeline to lock the protocol before screening/extraction
• When you need reproducible, auditable search and screening rules
• If stakeholders require explicit HUMAN approval before screening begins
• When multiple reviewers need a single operational contract to follow

Best practices

• Translate goals into 1–3 explicit research questions using observable terms
• Provide executable queries per database (fields, operators, dates) and record search date
• Use concrete MUST-HAVE and MUST-NOT criteria; avoid subjective terms like “relevant”
• Specify screening labels and a single tie-break policy; record all decisions in a screening_log.csv
• Define extraction columns with definitions, allowed values/units, and how to mark unknowns
• Include a short decision log and an explicit HUMAN approval statement before screening

Example use cases

• Start a review on agent-based evaluations: produce queries for arXiv, ACL Anthology, PubMed and a screening schema
• Lock protocol for a multi-reviewer team so screening can proceed reproducibly with tie-break rules
• Create an auditable extraction table (papers/extraction_table.csv) defining metrics, units, and unknown handling
• Generate a protocol that an auditor can read and re-run searches, then verify screening decisions