The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands (pdftotext, pdfinfo, unzip, grep) to perform a 'mechanical leak sweep' on the submission PDF and supplementary archives. These commands are standard utilities used locally for text extraction and metadata inspection.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to audit the contents of a research paper (main.pdf) and supplementary files to verify compliance with conference checklists. This creates an ingestion surface where instructions hidden within the paper's text could potentially influence the agent's audit report.
Ingestion points: The agent processes content from main.pdf via pdftotext and lists files in supp.zip via unzip -l as described in SKILL.md.
Boundary markers: Absent; there are no specific instructions to ignore embedded prompts within the audited documents.
Capability inventory: The skill possesses shell execution capabilities for file analysis (pdftotext, pdfinfo, unzip, grep).
Sanitization: Absent; the skill does not specify any filtering or escaping for the text extracted from the submission files.

emnlp-submission