opencode-qa

Warn

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill generates and executes Node.js and Bun scripts at runtime using shell heredocs. This pattern is extensively utilized in scripts/lsp-e2e.sh and scripts/lib/common.sh for environment probes and automated contract verification tests.
  • [DATA_EXFILTRATION]: Several helper scripts, such as scripts/db-session-by-id.sh and scripts/db-session-by-text.sh, access and read sensitive session data directly from the local SQLite database at ~/.local/share/opencode/opencode.db.
  • [COMMAND_EXECUTION]: The skill makes extensive use of system commands and subprocesses via tools like opencode, sqlite3, curl, jq, tmux, bun, node, npm, and python3 to facilitate its QA operations and manage local server instances.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides tools to extract and search session history from databases which store untrusted user messages. This creates a vulnerability surface if the extracted content is processed by an agent without sufficient sanitization.
  • Ingestion points: Session search and export functionality in scripts/db-session-by-text.sh and scripts/export-roundtrip.sh.
  • Boundary markers: The extraction scripts do not implement delimiters or instructions to differentiate retrieved data from executable agent commands.
  • Capability inventory: The skill's environment allows for significant command execution and system interaction through multiple CLI utilities.
  • Sanitization: While SQL inputs are escaped to prevent database injection, the message content itself is not filtered or sanitized for malicious instructions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 15, 2026, 01:35 PM
Security Audit — agent-trust-hub — opencode-qa