The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash(uv:*) and Bash(ls*) to perform routine file system checks. Specifically, it uses ls to verify the presence of required deliverables like PDF files, YAML specifications, and CI workflows. These operations are scoped to checking file existence as defined in the compliance protocol.
[REMOTE_CODE_EXECUTION]: The skill contains a Python script for word counting that uses regular expressions to strip LaTeX environments (like lstlisting or tikzpicture) and commands. This script is intended to be executed locally via uv or a similar runner to process the user's main.tex file. There is no evidence of remote code fetching or dynamic execution of untrusted external content.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of PDF briefs and LaTeX submissions. Ingestion points: The agent reads PDF content (via the Read tool) and LaTeX source files (SKILL.md). Boundary markers: The instructions explicitly tell the agent to "report exactly what is met, missing, or over-limit" and to "never modify the submission." While no explicit XML-style delimiters are mentioned, the task is strictly scoped to extraction and reporting. Capability inventory: The skill has access to Read, Glob, and Bash. Sanitization: The Python word-count script performs basic sanitization by stripping LaTeX commands and non-prose environments. The severity is low because the agent is instructed to act as a report-only auditor.

brief-compliance-check