The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes multiple local Python scripts (e.g., scripts/run_ledger.py, scripts/ars_apply_revision_patch.py, scripts/verify_submission_package.py) for managing the state of the writing pipeline, applying patches, and verifying manuscript integrity. It also suggests the use of system tools such as pandoc for document conversion and tectonic for LaTeX compilation.\n- [EXTERNAL_DOWNLOADS]: The citation_compliance_agent is designed to perform network lookups via retractionwatch.com and DOI resolvers to verify the validity and retraction status of academic citations. This is a documented functional requirement for the skill's academic integrity features.\n- [INDIRECT_PROMPT_INJECTION]: The pipeline ingests untrusted third-party data, such as reviewer comments and literature abstracts. The skill incorporates a robust 'canonical:instruction-data-boundary' directive throughout its agents, instructing them to treat all retrieved content as data and never promote it to instructions, mitigating potential injection attacks.\n- [DYNAMIC_EXECUTION]: The visualization_agent generates executable Python (matplotlib/seaborn) or R (ggplot2) code for the creation of publication-quality figures based on data extracted from the manuscript or provided by the user.\n- [PROMPT_INJECTION]: The skill defines a [direct-mode] token in the routing logic of SKILL.md that allows for bypassing standard clarification steps to route directly to specific agents. While a behavior override mechanism, it is an intentional internal control feature and does not appear to bypass safety protocols.

academic-paper