skills/edwinhu/workflows/pincite/Gen Agent Trust Hub

pincite

Warn

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: MEDIUMDYNAMIC_CONTEXT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTION
Full Analysis
  • [DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains instructions using the !command`` syntax which executes shell commands when the skill is loaded. The command attempts to execute a skill-toc binary from the user's home directory ($HOME/.claude/skills/plugin-utils/bin/skill-toc) to generate a table of contents.
  • [COMMAND_EXECUTION]: The script scripts/pincite.py utilizes subprocess.run to execute external command-line tools pdftotext and pdftoppm. These tools are used to extract text and images from local PDF files as part of the citation verification process.
  • [DATA_EXFILTRATION]: The skill transmits document snippets and metadata to external Google APIs (generativelanguage.googleapis.com and books.google.com) to retrieve and verify page numbers. Furthermore, scripts/review_build.py attempts to read API credentials from specific system paths such as /run/user/1000/agenix/gemini-api-key and paths defined by the XDG_RUNTIME_DIR environment variable.
  • [DYNAMIC_EXECUTION]: The scripts/gbooks.py script automates a local web browser using the Chrome DevTools Protocol (CDP) via the websocket-client library to retrieve data from Google Books. Additionally, assets/review/serve.py starts a local HTTP server that binds to the loopback address and accepts POST requests to write citation decisions to a local file, utilizing origin and header checks for security.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 5, 2026, 10:36 AM