The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted third-party content, such as manuscripts, academic search results, and reviewer feedback. This creates an attack surface where malicious instructions hidden in these materials could influence the agent's behavior.
Ingestion points: SKILL.md defines boundaries for retrieved external content, and agents/revision_coach_agent.md parses unstructured text from reviewer emails or PDFs.
Boundary markers: The skill implements an explicit instruction-data-boundary protocol in SKILL.md and uses structured HTML comments () to isolate data from instructions.
Capability inventory: The skill executes shell commands for document formatting (pandoc, tectonic) and runs various Python scripts for integrity verification and workflow management.
Sanitization: It utilizes a ground_truth_isolation_pattern.md (referenced in SKILL.md) to maintain the integrity of system instructions over data.
[COMMAND_EXECUTION]: The skill utilizes shell commands and local Python scripts to perform document transformations and validations.
agents/formatter_agent.md executes pandoc, tectonic, xelatex, bibtex, and biber to generate academic outputs in formats like PDF and DOCX.
Numerous Python scripts (e.g., scripts/check_acronyms.py, scripts/ars_apply_revision_patch.py, scripts/verify_submission_package.py) are invoked to manage the writing pipeline and verify manuscript integrity.
[EXTERNAL_DOWNLOADS]: The skill references and retrieves information from well-known academic, institutional, and technical services.
agents/citation_compliance_agent.md cross-references the Retraction Watch Database (retractionwatch.com).
references/venue_disclosure_policies.md contains many URLs for journal policy pages on domains such as nature.com, science.org, icmje.org, ieee.org, and aclweb.org. These are legitimate, established domains used for academic data retrieval.
[DYNAMIC_EXECUTION]: The visualization_agent.md generates executable Python (matplotlib/seaborn) or R (ggplot2) code to create publication-quality figures based on data provided by the user or extracted from the manuscript.
This script generation is handled through standardized templates and is a core functional component of the skill's visualization capabilities.

academic-paper