The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The setup.sh script uses sudo install to copy the gate-cli binary to /usr/local/bin if the local installation fails or is bypassed. This requests elevated system privileges to perform file system operations in protected directories.
[EXTERNAL_DOWNLOADS]: The skill downloads binary executables from https://github.com/gate/gate-cli/releases/download/ as part of its dependency setup process defined in SKILL.md and setup.sh. While the repository is owned by the skill author, downloading and executing binary files remains a significant security vector.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a local shell script setup.sh to handle dependencies. This script performs network operations using curl, file system modifications via mkdir and tar, and system detection via uname. The agent also executes the gate-cli binary for all core trading and market operations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external market sources (tickers, currency lists, and contract details) that could be manipulated by third parties to include malicious instructions directed at the agent.
Ingestion points: Market data fetched via gate-cli in references/market-viewing.md, references/token-discovery.md, and references/trading-buy.md.
Boundary markers: None identified in the workflow to separate external token data from internal instructions.
Capability inventory: The skill has significant capabilities, including placing financial orders and executing shell commands via the CLI.
Sanitization: There is no documentation of input validation or escaping for market data returned by the CLI before it is processed by the agent.

gate-exchange-alpha