The Agent Skills Directory

[SAFE]: The skill is designed for deep, price-blind research into prediction markets like Polymarket and Kalshi, prioritizing primary sources, official records, and legal dockets.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes data from external websites and social media platforms, which introduces an attack surface for indirect prompt injection.
Ingestion points: Web content retrieved via search tools (wideband, exa-cli), and social media feeds via xpool.
Boundary markers: The instructions do not define explicit markdown delimiters (e.g., XML tags) to encapsulate retrieved content, though they do provide strict content-type filtering instructions.
Capability inventory: The agent can browse the web, execute specific search utilities, and read/write research artifacts (sources, claims, reports) to a PMKNB database. It lacks arbitrary shell execution or file system access.
Sanitization: The skill mandates 'price-blindness' by requiring the redaction of market odds, prices, and trader comments before the data enters the agent's reasoning loop, which serves as a significant privacy and methodology safeguard.
[COMMAND_EXECUTION]: The skill provides procedures for utilizing external search and discovery utilities such as wideband, exa-cli, and xpool. These tools are employed within their intended functional scope for information retrieval.
[DATA_EXFILTRATION]: The skill interacts with the PMKNB database to store research outputs. It includes explicit prohibitions against reading or recording sensitive market-implied data (such as PnL or positions), mitigating the risk of inadvertent data exposure.

pm-deep-analysis