SolonCode in CI: Headless Runs, JSONL Events, and Remote Workspace Execution

Description and use

SolonCode enables headless AI agent runs in CI, with explicit tool permissions, turn and budget limits, and machine‑readable JSON output for automation.

Problem

Integrating interactive coding agents into CI pipelines requires non‑interactive, machine‑readable execution, explicit tool permissions, and bounded resource usage.

Outcome

SolonCode enables CI pipelines to run AI agents headlessly, enforce tool permissions, bound execution, and produce machine‑parseable JSON results, improving automation reliability.

Workflow

• 1. Invoke soloncode run with a prompt and options (output‑format, allowedTools, disallowedTools, permission‑mode, max‑turns, max‑budget‑usd).
• 2. Agent executes non‑interactively, producing a JSON result object (or text/stream‑json).
• 3. CI wrapper checks exit status and parses JSON to determine success, errors, or budget overruns.
• 4. Use --json-schema to constrain structured output when needed.
• 5. For long tasks, use stream-json with --verbose to capture event stream and optionally tee to a file.
• 6. Optionally resume sessions by reusing the session ID from the first run for a second phase (analysis then modification).

Evidence

SolonCode has a separate run entry point for that boundary. It is a one-shot, non-interactive execution mode

Source

The distinction between the exit code and the JSON body matters. A result can be valid JSON and still represent an incomplete or over-budget execution.

Source