math-guide-solver

What to consider before installing or running this skill: - Missing credential declarations: SKILL.md and README mention Claude API, Mathpix and Wolfram Alpha (optional). Those services require API keys/tokens, but the skill manifest declares no required env vars. Treat any prompt to provide API keys as coming from the external integrations — only add credentials after reviewing where and how they're used. - Inspect code before running: The package includes Python scripts (process_math_problem.py, generate_solution.py) and a Bash script (GITHUB_UPDATE_SCRIPT.sh). Review them locally — don't run the GitHub update script unless you understand and edit its hardcoded paths (/Users/... and /mnt/user-data...) because it copies/overwrites files and pushes tags to a remote repo. - Prompt-injection signal: The SKILL.md was flagged for unicode-control-chars. Open the SKILL.md and other docs in a hex-aware editor or view raw bytes to ensure there are no hidden control characters that could manipulate LLM prompts or tool behavior. - Network & secrets: The scripts reference external skills and APIs. If you enable external integrations, provision API keys in a controlled way (least privilege, dedicated keys, and in a sandbox account if possible). Monitor outbound network activity if you run the skill in an environment you control. - Missing artifacts: README references requirements.txt and tests, but requirements.txt was not included in the inspected manifest. Before running 'pip install -r requirements.txt' confirm which dependencies the code actually needs (skill.config.json lists PaddleOCR, math-images, matplotlib, Pillow). - Run in a sandbox first: Execute the scripts in an isolated environment (container or VM), with no production credentials mounted, verify behavior on simple inputs, and search for unexpected network calls (e.g., telemetry or exfil endpoints). - What would raise/lower my concern: If the publisher provides a trusted repository URL, a clear install script that lists required env vars, and no hidden control characters, this would raise confidence. Conversely, if the package attempts to read arbitrary local files, collect system secrets, or automatically make network calls to unknown endpoints, that would confirm malicious intent. Recommended immediate actions: open SKILL.md and all scripts in a plain-text editor, grep for 'http[s]://', 'api.', 'TOKEN', 'KEY', '/mnt', and for non-printable unicode; remove or edit the GITHUB_UPDATE_SCRIPT if you don't intend to run it; only supply external API keys after reviewing usage code and limiting keys appropriately.