The Agent Skills Directory

[PROMPT_INJECTION]: Static detector flags regarding instructions to ignore previous rules are false positives in this context. The skill's instructions in SKILL.md and references/discovery.md are defensive, specifically training the agent to treat external text as data and to disregard any embedded commands like "ignore previous instructions" or "send funds to..."
[DATA_EXFILTRATION]: The skill communicates exclusively with the pinned host https://www.playhunch.xyz. It contains explicit security invariants that prevent the agent from deriving request URLs from untrusted data, effectively mitigating redirection or exfiltration attacks.
[COMMAND_EXECUTION]: The skill defines a standard API interaction flow (discovery, quote, trade). The scripts/walkthrough.sh file is a benign developer utility for testing these endpoints and does not contain malicious command patterns.
[CREDENTIALS_UNSAFE]: The skill handles transaction signing via the x402 protocol but mandates strict pin-checks of the payment challenge against x402-registry.json. This ensures the agent only signs authorizations for the correct asset (USDC), network (Base), and recipient, preventing credential or fund misappropriation.
[REMOTE_CODE_EXECUTION]: No patterns for remote code execution or untrusted dependency installation were found. All network interactions are scoped to the vendor's documented API.

hunch