The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it ingests and processes untrusted data in the form of user-provided research papers and reviewer comments.
Ingestion points: Research paper drafts and external reviewer feedback are processed by the agents/pipeline_orchestrator_agent.md and agents/integrity_verification_agent.md.
Boundary markers: The instructions do not define explicit delimiters or instructions for the AI to ignore potentially embedded commands within the ingested academic content.
Capability inventory: The skill has the capability to perform file writes and execute system commands for document compilation.
Sanitization: There is no evidence of specific sanitization or validation of the input text before it is processed by the underlying language models.
[COMMAND_EXECUTION]: The skill involves dynamic execution through the compilation of LaTeX source code into PDF documents.
Evidence: The SKILL.md and agents/pipeline_orchestrator_agent.md specify the use of tectonic and pandoc during Stages 5 (FINALIZE) and 6 (PROCESS SUMMARY) to generate final academic papers and process records.
Context: This behavior is restricted to the generation of academic documents, which is the primary intended purpose of the skill, and uses well-known technology tools.

alterlab-research-pipeline