The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external research files to generate content and perform automated reviews via another model.
Ingestion points: Research outlines and reports are read from PAPER_PLAN.md and NARRATIVE_REPORT.md into the agent context using the Read tool.
Boundary markers: There are no explicit delimiters or instructions used to isolate the ingested content from the agent's operational instructions.
Capability inventory: The skill utilizes Bash, Write, and mcp__codex__codex (which sends the generated content to an external model for review).
Sanitization: No content validation or sanitization is performed on the text read from the project files before it is processed or passed to the review model.
[EXTERNAL_DOWNLOADS]: The workflow performs automated network requests to fetch bibliographic data from academic databases.
Evidence: Uses curl to retrieve JSON and BibTeX data from dblp.org and doi.org (CrossRef) for citation verification. These are established academic metadata services.
[COMMAND_EXECUTION]: The skill uses shell commands for project management and file manipulation.
Evidence: Employs Bash to create directories, manage backups, and write LaTeX files in chunks using cat << 'EOF' patterns to ensure reliability.

aris-paper-write