The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill handles highly sensitive private key data for a cryptocurrency wallet.
Evidence: The skill instructs the creation of a wallet file at ~/.t2000/wallet.key which stores plain Bech32 private keys.
Risk: Rule 5 explicitly states the wallet uses "No PIN. No encryption," meaning the security relies entirely on file system permissions.
Risk: Step 2 suggests using t2 init --import suiprivkey1xxx..., which passes a private key as a command-line argument. This is a known risk as the secret may be stored in plain text in the user's shell history files.
[EXTERNAL_DOWNLOADS]: The skill requires downloading and installing external software packages.
Evidence: npm install -g @t2000/cli in Step 1.
Note: The package @t2000/cli is a resource provided by the skill's vendor (t2000-afi).
[COMMAND_EXECUTION]: The skill guides the user through several shell commands to configure the local environment.
Evidence: Commands such as t2 init, t2 limit set, and t2 balance are used to interact with the local system.
Note: Risk is mitigated by Rule 4, which requires the agent to show the command and ask for user confirmation before execution.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow where the agent processes external data from a marketplace, which could potentially contain malicious instructions.
Ingestion points: The skill interacts with external service listings via the t2000_services tool and processes local configuration files.
Boundary markers: No specific delimiters or prompt isolation techniques are mentioned for handling untrusted marketplace data.
Capability inventory: The skill enables file system access (read/write to ~/.t2000/), network communication via a hosted MCP connector (https://mcp.t2000.ai/mcp), and financial transaction capabilities.
Sanitization: The skill mentions Bech32 validation for imported secrets.

t2000-setup