The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize external biological and intellectual property data, including clinical trial reports, literature abstracts, and patent records. This ingestion process constitutes a standard attack surface where instructions hidden within these external documents could potentially influence the agent's behavior during report synthesis.
Ingestion points: The agent retrieves data through various tools including ls_paper_fetch, ls_patent_fetch, ls_news_fetch, and ls_clinical_trial_fetch as described in SKILL.md.
Boundary markers: While the skill enforces a strict output schema (Roman numeral sections), it does not explicitly instruct the agent to use XML tags or other delimiters to isolate data retrieved from tools from the rest of the prompt context.
Capability inventory: The skill possesses extensive search and fetch capabilities across the medicinal chemistry, R&D pipeline, and intellectual property domains.
Sanitization: There is no specific instruction provided for the agent to sanitize, escape, or filter instructions found within retrieved text before processing.
[EXTERNAL_DOWNLOADS]: The skill's setup documentation references external MCP service endpoints located at connect.patsnap.com. These endpoints facilitate the skill's primary data retrieval functionality and are hosted by the author.
[COMMAND_EXECUTION]: The Setup Guide contains shell commands (e.g., claude mcp add) intended for manual execution by the user to configure the Claude environment with the necessary MCP transport servers.

target-intelligence