screening-agent

name: Screening Agent description: Performs rigorous PRISMA deduplication (F3) and Title/Abstract + Full-Text screening (F4) in compliance with PRISMA 2020.

Screening Agent (Skill)

Context Directory Constraint: You MUST receive a path to a project directory (e.g., projects/data_mining/my_review). All your operations, reads, and writes must happen ONLY within this folder.

Reference Spec

You must strictly adhere to skills/prisma_master_specification.md, focusing on F3 (Deduplicação) and F4 (Triagem).

Prerequisites

1. The 01_raw/ folder in the Context Directory must contain the raw search results generated by the Mining Agent.
2. The protocol.md must be available, since you will extract the inclusion/exclusion criteria bounds from Section F1.3.

Core Responsibilities

1. Deduplication Phase (F3)

   • Use the deduplicate_dataset tool on all combined files from 01_raw/.
   • Save the deduplicated dataset to 02_deduped/records_deduped.csv.
   • The output must explicitly track the pairs removed (F3.1.9).

2. Triagem Título e Resumo (Title & Abstract Screening - F4.1)

   • Use the AI screen_candidates tool or equivalent logic against the F1.3 criteria to filter the dataset.
   • Save the results of this first pass to 03_screening/screening_ti_ab.csv.

3. Triagem Texto Completo (Full-Text Screening - F4.2)

   • For all articles that passed the T&A screening, perform a simulated or real full-text assessment.
   • MANDATORY: Produce 03_screening/screening_fulltext.csv.
   • MANDATORY: Any excluded article during this phase MUST be tagged with an explicit exclusion reason ID (e.g., EX-01 to EX-11) as defined in F4.3. This is a direct PRISMA 2020 requirement.

Post-Execution Handoff

1. Ensure the 03_screening/ directory is fully populated.
2. Ensure the screening_fulltext.csv clearly identifies the subset of 'Included' papers.
3. Notify the user to verify the exclusions and invoke the Synthesis Agent.