The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill repeatedly references and directs users to okx.ai (e.g., in SKILL.md, references/ai-support.md, and references/intro.md). This domain has been flagged by automated scanners (URLite) as a 'CryptScam' and is not the official okx.com domain.
[EXTERNAL_DOWNLOADS]: The file _shared/preflight.md instructs the agent to download install.sh and installer-checksums.txt and execute them if the onchainos CLI is not found. This 'curl | bash' pattern is dangerous, especially when the instructions originate from a skill associated with flagged malicious URLs.
[COMMAND_EXECUTION]: The skill relies on the execution of numerous shell commands via the onchainos CLI tool, including sensitive operations like onchainos upgrade --force, onchainos wallet status, and onchainos wallet login. If the underlying CLI tool is compromised or pointed at the flagged okx.ai domain, these commands could facilitate data theft or system compromise.
[CREDENTIALS_UNSAFE]: In references/how-to-play.md, the skill instructs users to obtain and set sensitive API credentials (OKX_API_KEY, OKX_SECRET_KEY, OKX_PASSPHRASE). While it includes a warning not to paste these into the chat, the use of these credentials in an environment associated with a flagged scam domain (okx.ai) poses a severe risk of credential harvesting.
[INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface for indirect prompt injection (Category 8). It ingests untrusted data from user queries and external CLI command outputs (e.g., onchainos agent get-my-agents, onchainos agent search) and uses this data to drive further logic and routing without explicit sanitization or boundary markers.

okx-guide