The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute pdflatex for document compilation and grep for error analysis during the Phase 2 fix implementation. It also uses Bash to check for LaTeX errors and warnings in a non-interactive mode.
[DATA_EXFILTRATION]: Mathematical content from local LaTeX files is read using the Read tool and transmitted to an external reasoning model via the mcp__codex__codex tool. While this involves transmitting research data to an external provider, it is the primary intended function of the skill for remote mathematical review.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection (Category 8) because it ingests untrusted LaTeX files and interpolates their content into the reviewer prompt for the Codex model.
Ingestion points: Phase 0 reads project LaTeX files and reference materials.
Boundary markers: None identified; the proof content is directly inserted into the prompt template using a placeholder.
Capability inventory: The skill has access to powerful tools including Bash(*), Write, and Edit, which could be abused if the reviewer model is successfully subverted.
Sanitization: There is no evidence of LaTeX command sanitization or filtering of the input content before it is processed by the external model.

proof-checker