The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes internal Python scripts to perform document linting and profile verification.
Evidence: SKILL.md and scripts/run_preflight.py define the execution of bundled checking logic.
[EXTERNAL_DOWNLOADS]: The instructions involve fetching conference Call For Papers (CFP) URLs to ensure venue requirements are up-to-date.
Evidence: SKILL.md Step 2 instructs the agent to verify policies against the profile's cfp_url.
[PROMPT_INJECTION]: The skill processes untrusted LaTeX source code that may include directives to read local files, representing an indirect injection surface.
Ingestion points: scripts/texlib.py (_load_lines) recursively follows \input and \include directives.
Boundary markers: The linter limits recursion depth to 8, but it does not implement directory-based sandboxing for file path resolution.
Capability inventory: The tool reads local files and reports finding snippets back to the agent or user.
Sanitization: The parser performs basic sanitization through comment stripping and LaTeX command removal during analysis.
[SAFE]: The skill implements a hardened, custom-subset YAML parser in scripts/venue_profile.py to avoid the security risks associated with full-featured deserialization libraries.

preflight-check