The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides command templates that involve passing sensitive information as plaintext shell arguments. Examples include sign-with-plaintext-private-key --signer-private-key ed25519:<PRIVATE_KEY> and sign-with-seed-phrase '<SEED_PHRASE>'. This practice can lead to credential exposure in system logs and process lists.\n- [DATA_EXFILTRATION]: The skill references sensitive file paths that contain user credentials, specifically the legacy NEAR credentials directory at ~/.near-credentials/.\n- [COMMAND_EXECUTION]: The skill enables the agent to perform high-stakes administrative and financial actions on the NEAR blockchain, including deleting accounts, transferring tokens, and deploying smart contracts.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect injection as it generates shell commands based on external data. It lacks technical boundary markers and sanitization for inputs like account IDs or method arguments, relying instead on manual user approval.

near-cli-rs