The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted draft responses, creating a surface for indirect prompt injection where malicious content could attempt to influence the validator's output. However, the skill's specific focus on requiring evidence (OHLCV data, citations) acts as a constraint.
Ingestion points: The agent processes a 'draft response' provided in the input context.
Boundary markers: Absent; the instructions do not require the use of delimiters for untrusted data.
Capability inventory: The skill identifies and recommends the use of data_get_ohlcv, web_fetch, and local scripts like read_news.ts for evidence gathering.
Sanitization: The skill performs pattern matching and quote extraction but does not explicitly sanitize the input text.
[EXTERNAL_DOWNLOADS]: The skill recommends fetching data from well-known services like DeFiLlama and Alternative.me for protocol and market metrics. These are standard industry sources for the skill's stated purpose.

skeptic