The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data in the form of reviewer decision letters (text or PDF) and manuscript files to generate response documents.
Ingestion points: Reviewer decision letters and manuscript files (e.g., paper/main.tex, paper/main.qmd) are read to extract comments and verify changes.
Boundary markers: Reviewer comments are quoted verbatim using markdown blockquote syntax (>) in the generated response documents to distinguish them from author responses.
Capability inventory: The skill utilizes Read, Write, Edit, Bash, Grep, and Glob tools, and executes bundled Python validation scripts (check_response_claims.py, check_density_complaint.py) to verify edits.
Sanitization: While the skill enforces verbatim quoting of reviewer comments, there is no explicit sanitization or filtering of the input text, making it a surface for indirect prompt injection if an attacker-controlled review letter contains malicious instructions.

revise