systematic-debugging – Coding skill for AI agents

Systematic Debugging

Overview

Random fixes waste time and create new bugs. Quick patches mask underlying issues.

Core principle: ALWAYS find root cause before attempting fixes. Symptom fixes are failure.

Violating the letter of this process is violating the spirit of debugging.

The Iron Law

NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST

If you haven't completed Phase 1, you cannot propose fixes.

The Feedback Loop Rule

The feedback loop is the debugging work. Before reading code to build a theory, create or identify a tight command that can go red on the user's exact symptom and green when the bug is fixed. A tight loop is fast, deterministic, agent-runnable, and specific enough to catch this bug — not merely "doesn't crash".

When a clean repro is hard, spend disproportionate effort building the loop. Guessing without a red-capable loop is the failure mode this skill exists to prevent.

When to Use

Use for ANY technical issue:

• Test failures
• Bugs in production
• Unexpected behavior
• Performance problems
• Build failures
• Integration issues

Use this ESPECIALLY when:

• Under time pressure (emergencies make guessing tempting)
• "Just one quick fix" seems obvious
• You've already tried multiple fixes
• Previous fix didn't work
• You don't fully understand the issue

Don't skip when:

• Issue seems simple (simple bugs have root causes too)
• You're in a hurry (rushing guarantees rework)
• Someone wants it fixed NOW (systematic is faster than thrashing)

The Four Phases

You MUST complete each phase before proceeding to the next.

---

Phase 1: Root Cause Investigation

BEFORE attempting ANY fix:

1. Read Error Messages Carefully

• Don't skip past errors or warnings
• They often contain the exact solution
• Read stack traces completely
• Note line numbers, file paths, error codes

Action: Use read_file on the relevant source files. Use search_files to find the error string in the codebase.

2. Build a Tight Feedback Loop

• Can you trigger the user's exact symptom with one command?
• Does the command fail for this bug and only pass once the bug is fixed?
• Is it fast enough to run repeatedly?
• Is it deterministic? For flaky bugs, can you raise the reproduction rate high enough to debug?
• If not reproducible → gather more data, don't guess.

Ways to construct a loop — try in roughly this order:

1. Failing test at the seam that reaches the bug: unit, integration, or end-to-end.

2. HTTP script / curl against a running dev server.

3. CLI invocation with fixture input, diffing stdout/stderr against expected output.

4. Headless browser script (Playwright/Puppeteer) asserting on DOM, console, or network.

5. Replay a captured trace: HAR, request payload, event log, queue message, or webhook body.

6. Throwaway harness that boots the smallest useful slice of the system and calls the failing path.

7. Property / fuzz loop when the bug is intermittent wrong output over a broad input space.

8. Bisection harness suitable for git bisect run when the bug appeared between two known states.

9. Differential loop comparing old vs new version, two configs, two providers, or two datasets.

10. Human-in-the-loop script only as a last resort: script the human steps and capture their result so the loop stays structured.

Tighten the loop once it exists:

• Make it faster: cache setup, narrow scope, skip unrelated initialization.
• Make the signal sharper: assert the exact symptom, not generic success.
• Make it more deterministic: pin time, seed randomness, isolate filesystem, freeze network.

For non-deterministic bugs, the immediate goal is a higher reproduction rate, not perfection. Run the trigger 100x, parallelize, add stress, narrow timing windows, or inject sleeps. A 50% flake is debuggable; a 1% flake usually is not.

Action: Use the terminal tool to run the tight loop:

# Run a specific failing test
pytest tests/test_module.py::test_name -v

# Or run a scripted repro
python scripts/repro_bug.py

# Or run a high-repetition flaky repro
for i in {1..100}; do pytest tests/test_flake.py::test_name -q || break; done

3. Check Recent Changes

• What changed that could cause this?
• Git diff, recent commits
• New dependencies, config changes

Action:

# Recent commits
git log --oneline -10

# Uncommitted changes
git diff

# Changes in specific file
git log -p --follow src/problematic_file.py | head -100

4. Gather Evidence in Multi-Component Systems

WHEN system has multiple components (API → service → database, CI → build → deploy):

BEFORE proposing fixes, add diagnostic instrumentation:

For EACH component boundary:

• Log what data enters the component
• Log what data exits the component
• Verify environment/config propagation
• Check state at each layer

Run once to gather evidence showing WHERE it breaks.

THEN analyze evidence to identify the failing component.

THEN investigate that specific component.

5. Trace Data Flow

WHEN error is deep in the call stack:

• Where does the bad value originate?
• What called this function with the bad value?
• Keep tracing upstream until you find the source
• Fix at the source, not at the symptom

Action: Use search_files to trace references:

# Find where the function is called
search_files("function_name(", path="src/", file_glob="*.py")

# Find where the variable is set
search_files("variable_name\\s*=", path="src/", file_glob="*.py")

Phase 1 Completion Checklist

• [ ] Error messages fully read and understood
• [ ] A tight loop command exists and has been run at least once
• [ ] Loop is red-capable: it asserts the user's exact symptom, not a nearby failure
• [ ] Loop is deterministic, or a flaky bug has a high enough reproduction rate to debug
• [ ] Recent changes identified and reviewed
• [ ] Evidence gathered (logs, state, data flow)
• [ ] Problem isolated to specific component/code
• [ ] Root cause hypotheses can be stated and tested

STOP: Do not proceed to Phase 2 until you understand WHY it's happening.

---

Phase 2: Pattern Analysis

Find the pattern before fixing:

0. Minimize the Reproduction

Once the loop is red, shrink the repro to the smallest scenario that still goes red. Cut inputs, callers, config, data, and steps one at a time, re-running the loop after each cut. Keep only what is load-bearing for the failure.

Done when removing any remaining element makes the loop go green. A minimal repro narrows the hypothesis space and often becomes the cleanest regression test.

1. Find Working Examples

• Locate similar working code in the same codebase
• What works that's similar to what's broken?

Action: Use search_files to find comparable patterns:

search_files("similar_pattern", path="src/", file_glob="*.py")

2. Compare Against References

• If implementing a pattern, read the reference implementation COMPLETELY
• Don't skim — read every line
• Understand the pattern fully before applying

3. Identify Differences

• What's different between working and broken?
• List every difference, however small
• Don't assume "that can't matter"

4. Understand Dependencies

• What other components does this need?
• What settings, config, environment?
• What assumptions does it make?

---

Phase 3: Hypothesis and Testing

Scientific method:

1. Form Ranked Falsifiable Hypotheses

• Generate 3–5 plausible hypotheses before testing any single one.
• Rank them by likelihood and cheapness to falsify.
• State the prediction each hypothesis makes: "If X is the cause, then changing or observing Y should make Z happen."
• Discard or sharpen any hypothesis that does not make a testable prediction.

If the user is present, show the ranked list before testing. They may have domain knowledge that instantly re-ranks it. If the user is AFK, proceed with your ranking.

2. Test Minimally

• Test the highest-ranked hypothesis with the smallest possible probe.
• Change one variable at a time.
• Don't fix multiple things at once.
• Prefer debugger/REPL inspection when available; one breakpoint beats ten logs.
• If you add logs, tag every temporary line with a unique prefix such as [DEBUG-a4f2] so cleanup is a single search.

3. Verify Before Continuing

• Did it work? → Phase 4
• Didn't work? → Form NEW hypothesis
• DON'T add more fixes on top

4. When You Don't Know

• Say "I don't understand X"
• Don't pretend to know
• Ask the user for help
• Research more

---

Phase 4: Implementation

Fix the root cause, not the symptom:

1. Create Failing Test Case

• Simplest possible reproduction
• Automated test if possible
• MUST have before fixing
• Use the test-driven-development skill

2. Implement Single Fix

• Address the root cause identified
• ONE change at a time
• No "while I'm here" improvements
• No bundled refactor

… (full content is in the download package)