opencode-qa

Pass

Audited by Gen Agent Trust Hub on Jun 6, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local system commands (sqlite3, curl, jq, tmux) and the opencode binary itself. These executions are confined to local operations for the purpose of testing and verifying the coding agent's functionality.
  • [DATA_EXFILTRATION]: The skill performs read-only queries against the local opencode SQLite database (~/.local/share/opencode/opencode.db). This behavior is the primary intended function of the 'Case D' investigation tools and is clearly documented. No network exfiltration of this data was observed; network operations are restricted to local health checks and SSE probes.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes user-generated session titles and message parts from the local database. It mitigates common injection risks by employing SQL escaping (oqa_sql_escape) in its helper scripts and using structured JSON output for data handling.
  • [EXTERNAL_DOWNLOADS]: The skill uses curl to interact with the local HTTP server it spawns during testing. It does not download or execute scripts from remote or untrusted internet sources.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 6, 2026, 10:31 AM
Security Audit — agent-trust-hub — opencode-qa