pybreakz 0.1.0

A zero-dependency Python debugger CLI built for Claude Code

pybreakz

A zero-dependency Python debugger CLI built for Claude Code.

Run your script with breakpoints, capture locals and stack state, get clean structured output — all in one shot. No interactive session required.

---

Install

pip install pybreakz

Or install from source:

git clone https://github.com/allannapier/py_debug_cli.git
cd py_debug_cli
pip install -e .

Requires Python 3.9+. No pip dependencies.

---

Usage

Basic breakpoints

pybreakz run script.py --breakpoints 42
pybreakz run script.py --breakpoints 10,25,42

Captures all locals at each breakpoint hit.

Watch specific expressions

pybreakz run script.py --breakpoints 42 --watch "user_id,response.status,len(items)"

Evaluates each expression in the breakpoint's local scope.

Evaluate arbitrary expressions

pybreakz run script.py --breakpoints 42 --eval "df.shape,type(result),items[:3]"

Like --watch but separate section in output — useful for computed values.

Capture on exception

pybreakz run script.py --on-exception
pybreakz run script.py --on-exception --watch "self,request,data"

Captures locals, stack, and full traceback at the crash site.

Conditional breakpoints

pybreakz run script.py --breakpoints "42:x>100,67:status=='error'"

Only fires when the condition evaluates to True.

Pass arguments to your script

pybreakz run script.py --breakpoints 10 -- --input data.csv --verbose

Anything after -- is forwarded to the script as sys.argv.

JSON output (for programmatic use)

pybreakz run script.py --breakpoints 42 --format json

Timeout

pybreakz run script.py --breakpoints 10 --timeout 60   # 60s limit
pybreakz run script.py --breakpoints 10 --timeout 0    # no limit

Default timeout is 30 seconds.

---

Example output

════════════════════════════════════════════════════════════
  pybreakz report  →  script.py
════════════════════════════════════════════════════════════

┌─ BREAKPOINT HIT #1  script.py:42
│  Source:      result = process(item)
│
│  Call Stack (innermost last):
│  <module>()  [script.py:80]
│  main()  [script.py:60]
│  run_batch()  [script.py:42]
│
│  Locals:
│    items                = list[150 items]: [{'id': 1, ...}, ...]
│    item                 = {'id': 1, 'name': 'foo', 'active': True}
│    result               = None
│  Watched:
│    item['id']           = 1
│    len(items)           = 150
└────────────────────────────────────────────────────────────

---

How Claude Code uses it

Claude Code treats pybreakz like any other shell command. A typical debugging loop:

1. Claude reads your code and identifies suspicious lines
2. Calls pybreakz run script.py --breakpoints 34,67 --watch "x,response"
3. Reads the report output
4. Adjusts breakpoints or patches the bug
5. Repeats if needed

The --on-exception flag is especially useful as a first pass — just run it and let the crash tell Claude where to look.

---

Options reference

Flag	Short	Description
--breakpoints LINES	-b	Comma-separated line numbers, optionally with conditions (42:x>0)
--watch EXPRS	-w	Comma-separated expressions to evaluate at each hit
--eval EXPRS	-e	Additional expressions to evaluate (shown separately)
--on-exception	-x	Capture state on unhandled exception
--timeout SECS	-t	Script timeout in seconds (default: 30)
--format FORMAT	-f	Output format: text (default) or json
--max-hits N		Max breakpoint hits to record (default: 20)

---

Limitations

• Works on .py scripts run directly. Not designed for long-running servers or async event loops.
• Breakpoints must be on executable lines (not blank lines, comments, or def/class headers — use the first line inside the function body).
• sys.settrace has a small performance overhead — not suitable for tight performance benchmarks.
• Multi-threaded scripts: only the main thread is traced.