The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external bibliographic data (RIS, NBIB, CSV, PDF) which represents a surface for indirect prompt injection.
Ingestion points: The scripts scripts/prepare_records.py and scripts/prepare_fulltext.py ingest data from bibliographic export files and PDF documents into the agent's context.
Boundary markers: The skill uses explicit delimiters (=== SCREENING PROTOCOL ... ===) in templates/prompts.md and contains system instructions in agents/screening_reviewer_agent.md and SKILL.md to treat retrieved content as data, not commands.
Capability inventory: The screening reviewer subagents are restricted to the Read and Grep tools. The Python scripts perform local file reading and writing. No remote command execution tools or network-enabled agents are involved in processing the untrusted data.
Sanitization: The scripts/srlib.py helper script implements the spreadsheet_text function, which escapes formula prefixes (=, +, -, @) in generated CSV and Excel logs to prevent spreadsheet injection attacks.
[DYNAMIC_EXECUTION]: The skill generates and executes JavaScript workflow scripts to orchestrate multi-agent screening tasks.
Evidence: The scripts/build_workflow.py script assembles JavaScript templates (ta_screening.template.js, ft_screening.template.js) with localized configuration to be run via the Workflow tool. This is the intended core functionality of the skill and is constructed using local templates.

sr-screener