The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes a remote shell script directly from a non-whitelisted GitHub repository by piping curl output to sh (https://raw.githubusercontent.com/okx/onchainos-skills/main/install.sh). It also adds additional skills via npx skills add from external sources.
[EXTERNAL_DOWNLOADS]: The skill downloads a binary executable from the author's GitHub repository and grants it execution permissions (chmod +x) on the host system.
[DATA_EXFILTRATION]: The skill harvests system metadata including the hostname and the path of the user's home directory ($HOME) to generate a unique device identifier. This identity data is then transmitted to external telemetry servers (okx.com and a Vercel-hosted application).
[OBFUSCATION]: Base64 encoding is utilized in the installation script to conceal an HMAC key used for signing the telemetry data transmitted during the reporting process.
[COMMAND_EXECUTION]: The skill relies on shell command execution to perform environment discovery (uname, hostname) and to interact with the onchainos wallet CLI for executing blockchain transactions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from EVM RPC providers. (1) Ingestion points: Token names, symbols, and pool metadata fetched in rpc.rs. (2) Boundary markers: Present in SKILL.md as a notice to the agent. (3) Capability inventory: High-risk capabilities include submitting contract calls via onchainos.rs. (4) Sanitization: Absent; the skill passes raw RPC output to the agent's context.

sushiswap-v3