The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill employs dynamic execution techniques to manage dependencies and share functionality across related tools. Specifically, scripts/check-stranded-headings.py attempts to locate a Python interpreter containing the pymupdf library by searching the user's PIXI environments and $HOME directory, then re-executes itself using os.execv. Additionally, scripts/check-quotes.py uses importlib to dynamically load an extraction module from the westlaw skill's directory.
[COMMAND_EXECUTION]: The skill's scripts frequently use subprocess.run to invoke system utilities for PDF and text processing, such as pdfinfo, pdftotext, and jq. The script scripts/fetch-opinion.sh uses curl to interact with the CourtListener API and storage buckets to retrieve opinion text and PDFs.
[DYNAMIC_CONTEXT_INJECTION]: SKILL.md contains dynamic context placeholders (!command) that execute shell commands when the skill is loaded. These commands are used for documentation and workspace management, such as generating a table of contents via skill-toc and loading Typst formatting rules via typst-rules.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted court opinions fetched from external sources (e.g., CourtListener). This ingestion surface is mitigated by a mandatory "Iron Law of Authentic Text" enforced through scripts/check-quotes.py. This script performs a verbatim comparison between the generated excerpt and the retrieved source text, failing if any mismatch or "stitching signature" (indicating out-of-order text fragments) is detected, thereby preventing the agent from including unauthorized or malicious instructions from the source material.

elide-case