Tool output format

LLxprt Code sends tool outputs to the model as plain multi-line text (not JSON).

This avoids double/triple JSON-encoding that produces heavily escaped strings, which are harder for the model to read (especially for code).

Format

The tool output text is formatted as:

status:
<success|error>

toolName:
<tool-name>

error:
<error text, if any>

output:
<raw tool output text>

Notes:

• output is intended to be raw multi-line text (for example, code or logs).
• If there is no error, error is an empty string.
• When output is missing, it is replaced with [no tool result].
• Some tools return structured (JSON-like) results. For OpenAI providers we prefer a human-readable multi-line rendering (e.g. error text, stdout/stderr) to avoid an extra JSON wrapper like {"error":"..."} and preserve real newlines.