[BUG] [v0.0.7] cortex exec JSON result uses inconsistent field naming between --output-format json (snake_case) and --output-format stream-json

[vasylchow]

Project

cortex

Description

cortex exec emits JSON output with inconsistent field naming conventions depending on the --output-format value. The json format uses snake_case keys (duration_ms, num_turns, result) while the stream-json format uses camelCase keys (durationMs, numTurns, finalText) for the same semantic data. This breaks automation scripts that switch between output formats or parse both.

Error Message

Debug Logs

System Information

Ubuntu 24.04

Screenshots

Steps to Reproduce

1. Run cortex exec --output-format json "hello" — observe duration_ms, num_turns in JSON output
2. Run cortex exec --output-format stream-json "hello" — observe durationMs, numTurns in JSON output
3. Compare the field naming conventions between the two outputs

Expected Behavior

Both output formats from the same command should use a consistent field naming convention (either snake_case or camelCase) for the same semantic fields.

Actual Behavior

--output-format json uses snake_case (duration_ms, num_turns, result, is_error) while --output-format stream-json uses camelCase (durationMs, numTurns, finalText, toolCalls). Scripts that parse exec JSON output will break when switching between formats because the field names differ.

Additional Context

At src/cortex-cli/src/exec_cmd/runner.rs lines 528–548, the Json format emits:
{
"type": "result",
"subtype": "success",
"is_error": false,
"result": "...",
"duration_ms": 1234,
"num_turns": 1,
"session_id": "..."
}
At lines 552–562, the StreamJson format emits:
{
"type": "completion",
"finalText": "...",
"numTurns": 1,
"durationMs": 1234,
"session_id": "...",
"timestamp": 1712345678000,
"toolCalls": 0
}

[echobt]

Bug Validation: VALID

Reason:
The issue reports an inconsistency in the JSON output field naming conventions between --output-format json and --output-format stream-json in cortex exec. Specifically, json uses snake_case (e.g., duration_ms, num_turns) while stream-json uses camelCase (e.g., durationMs, numTurns).

I verified this in the codebase by examining src/cortex-cli/src/exec_cmd/runner.rs.

For ExecOutputFormat::Json, the code emits:

serde_json::json!({
    "type": "result",
    "subtype": "success",
    "is_error": false,
    "result": final_message,
    "duration_ms": duration_ms,
    "num_turns": num_turns,
    "session_id": handle.conversation_id.to_string(),
})

For ExecOutputFormat::StreamJson, the code emits:

let completion = serde_json::json!({
    "type": "completion",
    "finalText": final_message,
    "numTurns": num_turns,
    "durationMs": duration_ms,
    "session_id": handle.conversation_id.to_string(),
    "timestamp": chrono::Utc::now().timestamp_millis(),
    "toolCalls": tool_calls_count,
});

This confirms the bug report: the json format uses duration_ms and num_turns, while the stream-json format uses durationMs and numTurns. This inconsistency can break automation scripts that parse the output. The screenshots provided also show the camelCase output for stream-json. No exact duplicates were found. Therefore, this is a valid bug.

Bug Process:

graph TD
    A[User runs cortex exec --output-format json] --> B[Output uses snake_case keys]
    C[User runs cortex exec --output-format stream-json] --> D[Output uses camelCase keys]
    B --> E[Expected: Consistent naming convention across formats]
    D --> E
    B --> F[Actual Bug: Inconsistent naming breaks automation scripts]
    D --> F
    style F fill:#ff6b6b

Loading