screening-manager_skill

Overview

This skill manages title/abstract screening and records inclusion/exclusion decisions into papers/screening_log.csv according to an approved protocol. It ensures every decision is traceable to protocol clauses and produces an auditable CSV for review. Use only after output/PROTOCOL.md is present and signed off by a human.

How this skill works

The skill reads output/PROTOCOL.md to extract inclusion/exclusion criteria, time windows, and mandatory outcomes. It loads a candidate pool from an available papers/*.jsonl or papers/core_set.csv file (or prompts to import one). Each candidate is screened deterministically: include only if title/abstract clearly meets inclusion and does not trigger exclusion; otherwise exclude with a protocol-grounded reason. Finally it writes papers/screening_log.csv with required columns and performs quick QA to ensure decisions and reasons are present.

When to use it

• During the title/abstract screening stage of a systematic review (C2) after protocol approval
• When output/PROTOCOL.md exists and has human sign-off
• When you have a candidate pool (papers_raw.jsonl, papers_dedup.jsonl, or core_set.csv) or can import one
• When you need an auditable, reproducible screening log for PRISMA or internal audit

Best practices

• Lock and human-sign output/PROTOCOL.md before screening begins
• Prefer deterministic rules: map each reason to a specific protocol clause
• Decide exclude when key information is missing and cite the missing requirement in reason
• Keep reviewer field explicit (HUMAN or CODEX) and timestamp each decision
• Spot-check include rows against the protocol as a QA step

Example use cases

• Screening a deduplicated candidate list exported to papers/papers_dedup.jsonl
• Converting a core training set (papers/core_set.csv) into a full screening log for audit
• Re-running screening after tightening protocol clauses to produce a revised screening_log.csv
• Producing an auditable CSV for inclusion in a systematic review supplement