Debugging Methodology

Debugging Methodology

Overview

Stop guessing. Add instrumentation. Follow the evidence. When something fails, the answer is never "maybe it's X" followed by random changes. Add instrumentation that produces the specific data needed to explain the failure. Measure, don't speculate.

Why this matters: Random fixes waste time and create new bugs. Quick patches mask underlying issues. Systematic debugging is faster than guess-and-check thrashing, because every probe you add narrows the search space instead of widening it.

The Iron Law

NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST

If you haven't gathered evidence, you cannot propose fixes.

When to Use

• A test fails after a code change
• A build breaks or types stop compiling
• Runtime behavior doesn't match expectations
• A bug report arrives
• A failure is intermittent and needs to be pinned down
• "It works locally but fails in CI"
• A Result returns err() when you expected ok()

When NOT to use: Trivial, self-evident issues with a one-line obvious cause (a typo in a string, a missing import the compiler already names). For those, fix and move on. Everything else gets the protocol.

Related: Run [code-flow-analysis](../code-flow-analysis/SKILL.md) BEFORE instrumenting to map the execution path. Use [investigation-modes](../investigation-modes/SKILL.md) to size the effort. Debug err() returns with [result-types](../result-types/SKILL.md), instrument fn(args, deps) functions per [fn-args-deps](../fn-args-deps/SKILL.md), close out with [verification-before-completion](../verification-before-completion/SKILL.md), and state your certainty with [confidence-levels](../confidence-levels/SKILL.md).

The Anti-Pattern

Problem occurs
  → "Maybe it's a race condition"
  → Change something random
  → Still broken
  → "Could be caching"
  → Change something else
  → User frustrated, hours wasted

Why this happens: Insufficient data. You're debugging blind.

The Protocol

1. Document the Symptom Exactly

// WRONG: "The API returns an error"
// RIGHT: Exact symptom with context

/*
Symptom: getUser returns err('DB_ERROR')
Expected: ok(user) with { id: '123', name: 'Alice' }
Actual: err('DB_ERROR')
Context: userId='123', environment=test, after migration
Error message: "Connection refused to localhost:5432"
*/

Capture:

• Exact error message (copy-paste, don't paraphrase)
• Expected vs actual behavior
• Input values that triggered the failure
• Environment context

2. Add Instrumentation FIRST

Before forming ANY hypothesis, make the invisible visible:

// BEFORE: Silent failure
async function getUser(
  args: { userId: string },
  deps: GetUserDeps
): Promise> {
  const user = await deps.db.findUser(args.userId);
  if (!user) return err('NOT_FOUND');
  return ok(user);
}

// AFTER: Observable failure
async function getUser(
  args: { userId: string },
  deps: GetUserDeps
): Promise> {
  deps.logger.debug('getUser called', { userId: args.userId });

  try {
    deps.logger.debug('Calling db.findUser', { userId: args.userId });
    const user = await deps.db.findUser(args.userId);
    deps.logger.debug('db.findUser returned', {
      userId: args.userId,
      found: !!user,
      userKeys: user ? Object.keys(user) : null
    });

    if (!user) {
      deps.logger.info('User not found', { userId: args.userId });
      return err('NOT_FOUND');
    }

    deps.logger.debug('Returning user', { userId: user.id });
    return ok(user);
  } catch (error) {
    deps.logger.error('Database error in getUser', {
      userId: args.userId,
      error: error instanceof Error ? error.message : String(error),
      stack: error instanceof Error ? error.stack : undefined
    });
    return err('DB_ERROR');
  }
}

3. Instrument at Decision Points

Add logging at every branch, every external call, every state change:

| Point | What to Log | |-------|-------------| | Function entry | All args, relevant deps state | | External calls | Request params, before and after | | Conditionals | Which branch taken, why | | Transformations | Input shape, output shape | | Error paths | Full error object, context | | Function exit | Return value, duration |

4. Use OpenTelemetry Spans

For complex flows, use tracing:

import { trace } from 'autotel';

async function processOrder(
  args: { orderId: string },
  deps: ProcessOrderDeps
): Promise> {
  return trace('processOrder', { orderId: args.orderId }, async (span) => {
    span.setAttribute('order.id', args.orderId);

    const validationResult = await trace('validateOrder', async () => {
      return validateOrder(args, deps);
    });

    if (!validationResult.ok) {
      span.setAttribute('error.type', validationResult.error);
      return validationResult;
    }

    const paymentResult = await trace('processPayment', async () => {
      return processPayment({ order: validationResult.value }, deps);
    });

    span.setAttribute('payment.success', paymentResult.ok);
    return paymentResult;
  });
}

5. Form Evidence-Based Hypothesis

Only AFTER you have data:

Evidence from logs:
- getUser called with userId='123' ✓
- db.findUser called ✓
- db.findUser threw: "Connection refused to localhost:5432"

Hypothesis: Database connection failed
Evidence: Error message explicitly says "Connection refused"
Test: Check if database is running, check connection string

Hypothesis must:

• Be based on observed data (not speculation)
• Explain ALL symptoms
• Be testable with a single change

6. One Change at a Time

WRONG:
  → Change connection string AND add retry AND increase timeout
  → "It works now!" (but which fix was it?)

RIGHT:
  → Check database is running (it wasn't)
  → Start database
  → Retest → Works
  → Root cause: Database container wasn't started

Result Type Debugging

When debugging Result-returning functions:

Check the Error Path

// Add explicit logging for err() returns
if (!user) {
  const errorContext = {
    userId: args.userId,
    dbQueryExecuted: true,
    queryDuration: Date.now() - startTime,
  };
  deps.logger.info('Returning NOT_FOUND', errorContext);
  return err('NOT_FOUND');
}

Trace Through Workflows

// When createWorkflow fails, add step-level logging
const workflow = createWorkflow()
  .pipe(step('validate', async (input) => {
    deps.logger.debug('Step: validate', { input });
    const result = await validate(input, deps);
    deps.logger.debug('Step: validate result', { ok: result.ok });
    return result;
  }))
  .pipe(step('process', async (input) => {
    deps.logger.debug('Step: process', { input });
    const result = await process(input, deps);
    deps.logger.debug('Step: process result', { ok: result.ok });
    return result;
  }));

Verify Deps Are Correct

// When mocking in tests, verify deps behavior
it('returns NOT_FOUND when user missing', async () => {
  const deps = mock();
  deps.db.findUser.mockResolvedValue(null);

  // Debug: verify mock is configured correctly
  console.log('Mock configured:', {
    findUser: deps.db.findUser.getMockName(),
    mockReturnValue: await deps.db.findUser('test'),
  });

  const result = await getUser({ userId: '123' }, deps);

  // Debug: verify what was called
  console.log('findUser called with:', deps.db.findUser.mock.calls);
  console.log('Result:', result);

  expect(result.ok).toBe(false);
});

Debugging Checklist

Before guessing, answer these:

| Question | How to Answer | |----------|---------------| | What are the input values? | Log function entry | | Which code path executed? | Log at conditionals | | What did external calls return? | Log before/after deps calls | | What error was thrown? | Log in catch block with stack | | What was the return value? | Log function exit |

Anti-Patterns

Never Guess

WRONG: "It's probably a race condition"
RIGHT: "Let me add timestamps to see the execution order"

WRONG: "Maybe the mock isn't working"
RIGHT: "Let me log what the mock returns when called"

WRONG: "Could be a caching issue"
RIGHT: "Let me log cache hits/misses with keys"

Never Change Multiple Things

WRONG: Fix A, B, and C → "It works!" (which one fixed it?)
RIGHT: Fix A → Test → Still broken → Fix B → Test → Works (B was the issue)

Never Remove Instrumentation Prematurely

WRONG: Problem seems fixed → Delete all logging
RIGHT: Problem confirmed fixed → Keep structured logging, remove debug-level only

Never Assume Code Does What It Says

// Function is CALLED getUser but does it RETURN user?
async function getUser(args, deps) {
  // Don't assume. Log the actual return value.
  const result = /* ... */;
  deps.logger.debug('getUser returning', { result });
  return result;
}

Integration with Testing

Debug Failing Tests Systematically

it('processes order correctly', async () => {
  // 1. Log test setup
  console.log('=== TEST: processes order correctly ===');

  const deps = mock();
  deps.db.findOrder.mockResolvedValue(mockOrder);
  console.log('Deps configured:', {
    findOrderReturns: mockOrder
  });

  // 2. Log function call
  console.log('Calling processOrder with:', { orderId: '123' });
  const result = await processOrder({ orderId: '123' }, deps);

  // 3. Log result before assertion
  console.log('Result:', JSON.stringify(result, null, 2));

  // 4. Now assert
  expect(result.ok).toBe(true);
});

Verify Mock Interactions

// After test fails, check what was actually called
console.log('Mock calls:', {
  findOrder: deps.db.findOrder.mock.calls,
  saveOrder: deps.db.saveOrder.mock.calls,
  sendEmail: deps.mailer.send.mock.calls,
});

Decision Tree

Problem occurs
    │
    ▼
Can you see exact failure point?
    NO → Add function entry/exit logging
    YES ↓
Do you know input values at failure?
    NO → Log all args and deps state
    YES ↓
Do you know which code path executed?
    NO → Log at every conditional
    YES ↓
Do you know what external calls returned?
    NO → Log before/after deps calls
    YES ↓
Do you know the error details?
    NO → Log full error with stack
    YES ↓
Form hypothesis from evidence → Test with ONE change

Quick Reference

| Symptom | Instrumentation to Add | |---------|----------------------| | Wrong return value | Log at every return statement | | Function never called | Log at caller and callee entry | | External call fails | Log request params and response | | Intermittent failure | Add timestamps, log state | | Test passes locally, fails in CI | Log environment, deps versions | | Result is err() unexpectedly | Log at every err() return |

Debugging Workflows

When a createWorkflow() pipeline fails, debug systematically:

1. Identify Which Step Failed

const workflow = createWorkflow({ validateOrder, processPayment, sendConfirmation });

const result = await workflow(async (step) => {
  deps.logger.debug('Starting workflow');

  const validated = await step(() => validateOrder(args, deps));
  deps.logger.debug('validateOrder completed', { ok: validated !== undefined });

  const payment = await step(() => processPayment({ order: validated }, deps));
  deps.logger.debug('processPayment completed', { ok: payment !== undefined });

  const confirmation = await step(() => sendConfirmation({ order: validated, payment }, deps));
  deps.logger.debug('sendConfirmation completed', { ok: confirmation !== undefined });

  return { order: validated, confirmation };
});

if (!result.ok) {
  deps.logger.error('Workflow failed', { error: result.error });
  // Error type tells you which step failed
}

2. Use Existing Traces

If you have OpenTelemetry set up (see observability skill), use existing traces:

In Jaeger/Honeycomb:
1. Find the trace by traceId or time range
2. Look for the RED span (error status)
3. Expand span attributes for error details
4. Check parent spans for context

Query example:
  service.name = "my-service"
  status = ERROR
  span.kind = INTERNAL

3. Debug step.try() Failures

When third-party code throws:

const config = await step.try(
  () => {
    deps.logger.debug('Parsing config JSON', { raw: user.configJson });
    return JSON.parse(user.configJson);
  },
  {
    error: 'INVALID_CONFIG' as const,
    onError: (e) => {
      // Log the actual exception before it's converted to Result
      deps.logger.error('JSON.parse failed', {
        error: e instanceof Error ? e.message : String(e),
        raw: user.configJson?.substring(0, 100),  // First 100 chars
      });
    },
  }
);

Integration with Observability

When debugging in production, leverage your existing traces:

1. Find the trace - Use traceId from logs or error reports
2. Walk the spans - Each step() creates a child span
3. Check attributes - Look for error.type, error.message
4. Correlate with logs - Use traceId to find related log entries

// In your error handler, include trace context
app.use((err, req, res, next) => {
  const span = trace.getSpan(context.active());
  logger.error('Request failed', {
    error: err.message,
    traceId: span?.spanContext().traceId,
    spanId: span?.spanContext().spanId,
    path: req.path,
  });
  res.status(500).json({ error: 'Internal error', traceId: span?.spanContext().traceId });
});

Red Flags - STOP and Follow Process

If you catch yourself thinking:

• "Quick fix for now, investigate later"
• "Just try changing X and see if it works"
• "Add multiple changes, run tests"
• "It's probably X, let me fix that"
• "I don't fully understand but this might work"
• Proposing solutions before tracing data flow
• "One more fix attempt" (when already tried 2+)

ALL of these mean: STOP. Return to gathering evidence.

The 3-Fix Rule

If 3+ fixes failed:

1. STOP attempting more fixes
2. Question the architecture
3. Consider: Is the pattern sound?
4. Discuss with team before attempting more fixes

Pattern indicating architectural problem:

• Each fix reveals new problems in different places
• Fixes require massive refactoring
• Each fix creates new symptoms elsewhere

Common Rationalizations

| Excuse | Reality | |--------|---------| | "Issue is simple, don't need process" | Simple issues have root causes too | | "Emergency, no time for process" | Systematic debugging is faster than thrashing | | "Just try this first, then investigate" | First fix sets the pattern. Do it right. | | "Multiple fixes at once saves time" | Can't isolate what worked. Causes new bugs. | | "I see the problem, let me fix it" | Seeing symptoms is not understanding root cause | | "One more fix attempt" (after 2+ failures) | 3+ failures means an architectural problem |

Verification

A bug is not fixed until:

• [ ] The exact symptom was captured (error message, expected vs actual, inputs)
• [ ] Evidence from instrumentation explains the failure (not a guess)
• [ ] The hypothesis was confirmed with a single isolating change
• [ ] Root cause is identified, not just where it manifests
• [ ] A regression test exists that fails without the fix and passes with it
• [ ] The original failure no longer reproduces, end-to-end
• [ ] All tests pass and the build succeeds
• [ ] Debug-level instrumentation is removed; structured logging/traces are kept

Remember

• Evidence before hypothesis - Don't guess, measure
• One change at a time - Know what fixed it
• Structured logging pays off - Pino plus OpenTelemetry
• Keep observability - Debug logs become production traces
• Use existing traces - Don't re-instrument if traces already exist

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jagreehal
• Source: jagreehal/jagreehal-claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.