opencode-sdk

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • Indirect Prompt Injection (HIGH): The skill enables the agent to read local files and subscribe to external, real-time event streams. Malicious instructions embedded in these data sources could hijack the agent's logic to execute unauthorized operations via the SDK's high-privilege functions.
  • Capability Inventory (HIGH): The skill provides tools for session manipulation, TUI control, and file modification. Combined with data ingestion, this creates a 'Read-to-Write' loop where untrusted data can influence system-altering commands.
  • Unverifiable Dependencies (MEDIUM): The skill relies on @opencode-ai/sdk, a package from a non-standard/untrusted source, which is installed via npm during setup.
  • Credential Exposure (MEDIUM): The toolkit includes authentication setup methods (client.auth.set). If the agent is successfully injected, it could be coerced into leaking API keys or credentials stored within these configurations.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 12:51 PM