The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data including news headlines, regulatory filings, and web search results from financial data providers like DefiLlama and Glassnode. This represents a surface for indirect prompt injection if malicious content is embedded in those sources.
Ingestion points: News content via longbridge news, regulatory filings via longbridge filing, and web search results for DeFi APY and on-chain metrics.
Boundary markers: The instructions repeatedly emphasize citing actual sources, timestamps, and reporting periods. It requires distinguishing estimates from observations and includes a mandatory independence statement for IPO analysis.
Capability inventory: The skill executes shell commands via the longbridge CLI and utilizes web search capabilities to supplement data.
Sanitization: The skill uses structured JSON output for CLI data and requires formatting results into specific research templates, though it does not explicitly describe sanitization of free-text news content.
[COMMAND_EXECUTION]: The skill documents and utilizes a variety of subcommands for the longbridge CLI tool to fetch market data, financials, and account information. These operations are scoped to research and planning tasks as described in the documentation.

longbridge-research