Single Line Error Summaries

Single-Line Error Summarization Protocol

Overview

When reporting a test failure or build error back to the user or across multi-agent boundaries, default agents dump the complete 80-to-120 line raw terminal traceback into the chat transcript.

Uncondensed stack traces:

1. Pollute Context Memory: 80 lines of V8/CPython internal runtime frames consume 1,200+ tokens per failure turn.
2. Obscure Root Cause: The human developer must scroll through pages of boilerplate middleware frames to find the single line of user code that broke.
3. Induce Token Waste in Follow-Up Turns: Subsequent turns re-send the entire verbose error log.

The Single-Line Error Summarization Protocol parses tracebacks down to a standardized 1-Line Root Cause Token: 🚨 [FAIL: path/to/file:line] ErrorType: Root Cause Message.

100-Line Raw Traceback vs. Single-Line Error Summary

┌─────────────────────────────────────────────────────────────┐
│                 Error Reporting Density                     │
│                                                             │
│  Raw Unparsed Traceback (85 Lines / 950 Tokens):            │
│  Traceback (most recent call last):                         │
│    File "/usr/lib/python3.11/asyncio/runners.py", line 118  │
│    File "/usr/lib/python3.11/asyncio/tasks.py", line 277    │
│    File "/app/src/services/billing.py", line 42, in charge  │
│      user_record = db.get_user(user_id)                     │
│    File "/app/src/db/client.py", line 18, in get_user       │
│      raise KeyError(f"User {user_id} not found in database")│
│    ... [75 lines of internal ASGI framework frames]         │
│                                                             │
│  Single-Line Error Summary (18 Tokens - 98.1% Reduction):   │
│  🚨 [FAIL: src/db/client.py:18] KeyError: User 404 not found│
│  ↳ 18 clean tokens, instant pinpoint identification         │
└─────────────────────────────────────────────────────────────┘

The Standardized Single-Line Error Schema

Format all runtime and compiler failures using this deterministic signature:

🚨 [FAIL: <relative_file_path>:<line_number>] <ExceptionName>: <Direct Failure Message>

Multi-Language Single-Line Examples:

Production Python Traceback Condenser

import re

def condense_traceback(raw_trace: str) -> str:
    """Condenses multi-line Python/Node stack traces into a single root cause line."""
    # Check Python Traceback (Extract last frame in user code)
    py_frames = re.findall(r'File "([^"]+)", line (\d+), in (\w+)\n\s*(.+)\n(\w+Error: .+)', raw_trace)
    if py_frames:
        file_path, line_no, func, code_line, err_msg = py_frames[-1]
        return f"🚨 [FAIL: {file_path}:{line_no}] {err_msg.strip()}"

    # Check JS / TS TypeError / ReferenceError
    js_match = re.search(r'(\w+Error: .+)\n\s+at (?:async )?(.+) \((.+):(\d+):(\d+)\)', raw_trace)
    if js_match:
        err_msg, func, file_path, line_no, col = js_match.groups()
        return f"🚨 [FAIL: {file_path}:{line_no}] {err_msg.strip()}"

    # Fallback: Return first error line
    for line in raw_trace.splitlines():
        if "Error:" in line or "Exception:" in line or "FAIL" in line:
            return f"🚨 {line.strip()}"
            
    return raw_trace.splitlines()[-1].strip()

Benchmark Comparison

Reporting 40 test failures and runtime exceptions during a debugging session:

Agent Operational Directive

MANDATORY: When communicating test or build failures in chat or subagent returns, agents must condense multi-line stack traces into single-line root cause tokens (🚨 [FAIL: file:line] ErrorType: Message).