The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads binary executables from a remote repository during the installation process.
Evidence: SKILL.md and setup.sh reference download URLs for gate-cli from https://github.com/gate/gate-cli/releases/download/v0.6.2/.
[REMOTE_CODE_EXECUTION]: The skill downloads and installs a binary file (gate-cli) that is subsequently executed by the agent to perform account and investment operations.
Evidence: The setup.sh script downloads, extracts, and installs the binary, which is then mapped to tools in SKILL.md.
[PRIVILEGE_ESCALATION]: The installation script requests elevated privileges to modify system directories.
Evidence: setup.sh uses sudo install -m 755 to install the binary to /usr/local/bin if the local user-space installation fails.
[COMMAND_EXECUTION]: The skill core functionality relies on executing shell commands with user-influenced arguments.
Evidence: Frequent subprocess calls to gate-cli with parameters like amount, plan_id, and currency derived from user queries.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data from API responses and has write capabilities.
Ingestion points: Data retrieved from gate-cli cex earn dual plans, orders, and balance is processed by the agent in references/product-query.md and references/settlement-assets.md.
Boundary markers: Absent. There are no instructions providing delimiters or warnings to ignore malicious instructions embedded in the CLI output.
Capability inventory: The skill has write access through the gate-cli cex earn dual place command documented in references/subscription.md.
Sanitization: Absent. The skill instructions focus on mathematical transformations (APY conversion) but do not explicitly validate or sanitize the string content of the API responses before interpolation into final prompts.

gate-exchange-dual