The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone code from an unverified GitHub repository (https://github.com/daijinma/geo_marketing.git). This source is not recognized as a trusted organization or well-known service.
[COMMAND_EXECUTION]: The installation and setup process involves numerous shell commands with broad system impact, including:
git clone for remote code retrieval.
make setup and make db for environment initialization.
uv pip install and pnpm install for dependency management across multiple languages.
playwright install chromium for browser engine deployment.
docker-compose up for service orchestration.
wails build and go mod tidy for binary compilation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data originating from generative AI search engines, creating a vulnerability surface for indirect injection.
Ingestion points: Data is ingested through DeepSeekProvider and DoubaoProvider by scraping response_text and citation-link elements from live AI chat sessions.
Boundary markers: None identified. The skill directly extracts and processes the text content of the AI's response.
Capability inventory: The skill has the ability to write to PostgreSQL and SQLite databases, perform network requests via Playwright and requests, and execute background tasks.
Sanitization: Content is passed through tldextract for domain parsing and json.dumps for storage, but the raw response_text is stored without safety filtering.
[DYNAMIC_EXECUTION]: The monitoring service utilizes Playwright to automate browser interactions. This involves executing browser-level logic and navigating to external sites (e.g., chat.deepseek.com) to extract data, which inherently executes remote scripts in a controlled environment.

llm-sentry-geo-marketing