The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through its literature review and analysis phases. This is a common design pattern for research agents but introduces risk if external data contains malicious instructions.
Ingestion points: Untrusted data from research papers enters the agent context in SKILL.md (Phase 1) via the web_search and web_extract tools.
Boundary markers: The instructions do not define strict system-level delimiters or specific instructions to disregard potential commands embedded in academic text.
Capability inventory: The agent has access to powerful tools such as terminal, execute_code, and delegate_task, which could be abused if an injection were successful.
Sanitization: There is no evidence of filtering or sanitization of external academic content before interpolation into prompts.
[EXTERNAL_DOWNLOADS]: The skill references and downloads resources from well-known and trusted organizations.
A Makefile in templates/neurips2025/ uses curl to fetch style files from the official NeurIPS domain (media.neurips.cc).
The skill recommends the installation of the well-known SciencePlots Python package for visualization.
Reference documents in the templates/ directory link to official ACL GitHub repositories and Overleaf projects.
[METADATA_POISONING]: A discrepancy exists between the author listed in the YAML frontmatter of SKILL.md ("Orchestra Research") and the provided author context ("hurtzdonut559"). This appears to be a documentation artifact from the skill's development history rather than a deceptive attempt.
[COMMAND_EXECUTION]: The skill frequently instructs the agent to execute shell commands using the terminal and process tools for tasks such as LaTeX compilation, Git version control, and running experiment scripts.
[DYNAMIC_EXECUTION]: The skill methodology (specifically the autoreason loop) utilizes execute_code and delegate_task to dynamically analyze experiment results and generate parallel paper sections.

research-paper-writing