The Agent Skills Directory

[COMMAND_EXECUTION]: Uses which, latex --version, and kpsewhich in SKILL.md and references/environment.md to detect local developer tools and LaTeX packages.- [COMMAND_EXECUTION]: The scripts/prepare_logo.py file executes the rsvg-convert binary via subprocess.run to process user-provided SVG logos.- [DYNAMIC_EXECUTION]: Programmatically generates and compiles LaTeX documents via assets/Makefile, utilizing local engines such as pdflatex, latexmk, or tectonic.- [INDIRECT_PROMPT_INJECTION]: The skill ingests codebase content like comments and documentation (Phase 3 Research). Evidence Chain: (1) Ingestion points: Codebase files identified via find and grep. (2) Boundary markers: Absent, but explicit agent instructions are provided to treat input as untrusted. (3) Capability inventory: Subprocess calls, file system writes, and LaTeX compilation. (4) Sanitization: Specifically instructs the agent to scan collected text for AI instructions and strip or rephrase them before generating the manual.

talisman