The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches official patent classification data from the World Intellectual Property Organization (wipo.int) and the China National Intellectual Property Administration (cnipa.gov.cn). Additionally, it downloads pre-trained AI embedding models from well-known hubs including HuggingFace (huggingface.co) and ModelScope (modelscope.cn). These downloads are performed for legitimate skill operations and target trusted or official domains.
[COMMAND_EXECUTION]: The skill executes the obsidian CLI tool using subprocess.run to verify vault integrity and perform local data queries. This execution is limited to a specific known tool and is integrated into the user's existing Obsidian workflow.
[INDIRECT_PROMPT_INJECTION]: The skill ingests markdown files from the user's Obsidian vault, which introduces a potential surface for indirect prompt injection if note content contains malicious instructions.
Ingestion points: Markdown interpretation notes located within the Research/Patents directory of the user's vault (e.g., vault_index.py).
Boundary markers: The skill does not implement explicit delimiters or instructions to ignore embedded commands within the processed markdown content.
Capability inventory: The skill manages a local web server (binding only to 127.0.0.1), executes the obsidian CLI tool, and performs local file system operations related to an SQLite cache and model storage.
Sanitization: The frontend uses escapeHtml to sanitize data before rendering it in the browser, providing protection against basic cross-site scripting (XSS) from malicious note content.

patent-map