Logging guide

This guide covers logging patterns, conventions, and integration for @pencroff-lab/kore logger.

Overview

Aspect	Decision
Injection	DI via handler factory deps
Correlation	Child logger with `runId` binding
Dry-run	Child logger with `DRY-RUN` module prefix
Error logging	DEBUG at origin, ERROR at boundary only

Log Levels

Levels are bound directly on every Logger instance as readonly properties. Access them via the log function itself — no separate import needed:

// Levels are properties on the log function
log(log.INFO, 'Starting generation');
log(log.ERROR, 'Command failed', err);

// Works on any Logger instance (child loggers too)
const svcLog = log.child('claude');
svcLog(svcLog.DEBUG, 'Calling CLI');

Level	Use Case	Example
`log.TRACE`	Function entry/exit, loop iterations	`Entering parseMarkdown`
`log.DEBUG`	Variable values, intermediate results, error origin	`Parsed 5 sections`, `CLI returned error`
`log.INFO`	Major workflow steps, user-visible events	`Generating note for section 3`
`log.WARN`	Recoverable issues, retry attempts	`CLI timeout, retrying (1/3)`
`log.ERROR`	Failures at handler boundary	`Command failed` (logged once)
`log.FATAL`	Unrecoverable, process exits	`Database connection failed`

Call Signatures

The Logger is a callable function with overloaded signatures:

log('Server started');                              // INFO level (default)
log('Retrying connection', { attempt: 2 });         // INFO + context object
log('Retrying connection', 'attempt 2 of 3');       // INFO + detail string
log(log.ERROR, 'Connection lost');                  // Explicit level
log(log.ERROR, 'Connection lost', err);             // Level + Err instance
log(log.DEBUG, 'Query result', { rows: 42 });       // Level + context object

When the first argument is a valid level string, it’s used as the level. Otherwise, defaults to INFO.

Transports

The logger uses a transport abstraction for output. The built-in prettyTransport writes human-readable output to process.stderr with optional ANSI colors.

import { createLogger, prettyTransport, lvl } from "@pencroff-lab/kore";
import type { LogEntry, LogTransport } from "@pencroff-lab/kore";

// Default: prettyTransport with auto-detected colors
const appLog = createLogger('app');

// Custom transport (useful for testing)
const entries: LogEntry[] = [];
const spy: LogTransport = { write(e) { entries.push(e); } };
const testLog = createLogger('test', { transports: [spy], level: lvl.TRACE });

// Pretty transport with explicit options
const verboseLog = createLogger('app', {
  transports: [prettyTransport({
    colors: false,           // Disable ANSI (default: 'auto' based on TTY)
    timestamp: 'iso',        // ISO 8601 (default: 'short' HH:MM:SS.mmm)
  })],
});

Output format (pretty transport, colors disabled):

12:34:56.789 INF [app] [handler] Server started {"port":3000}
12:34:56.800 ERR [app] [handler] Request failed
  err: Err: Connection timeout [NET:TIMEOUT]

Err instances in context are rendered on their own indented line below the main line.

Logger DI Pattern

Logger is passed through deps, matching the handler factory pattern. Only import createLogger at the entry point; everywhere else receive Logger type via deps:

import { ok, type Logger, type ResultTuple } from "@pencroff-lab/kore";

interface HandlerDeps {
  db: Database;
  log: Logger;  // Injected logger function
  config: Config;
}

function createHandler(deps: HandlerDeps): ResultTuple<Handler> {
  // Validate deps...

  const handler = async (argv: ArgumentsCamelCase<Args>): Promise<void> => {
    const { log, db, config } = deps;
    const runId = crypto.randomUUID();

    // Create run-scoped logger with correlation
    const runLog = log.child('handler', { runId });

    // Create dry-run logger if needed
    const dryRunLog = argv.dryRun
      ? runLog.child('DRY-RUN')
      : null;

    const ctx: ExecutionContext = {
      runId,
      dryRun: argv.dryRun ?? false,
      log: runLog,
      dryRunLog,
    };

    runLog(runLog.INFO, 'Starting handler');
    // ... application logic using ctx
  };

  return ok(handler);
}

Run Correlation

Use child loggers with runId binding so all log lines from the same run can be traced:

const runLog = log.child('handler', { runId });

// All subsequent log calls include runId automatically
runLog(runLog.INFO, 'Starting generation', { plan: argv.plan });
runLog(runLog.INFO, 'Processing section', { index: 1, title: "Gradient Descent" });

Error Logging Strategy

Principle: Log DEBUG at origin, add context via .wrap() in intermediate layers, log ERROR once at boundary.

Service Layer (Origin) - DEBUG Only

import { attemptAsync, Err, fail, ok, type ResultTuple } from "@pencroff-lab/kore";

async function callClaude(prompt: string, ctx: ExecutionContext): Promise<ResultTuple<string>> {
  return attemptAsync(async () => {
    const result = await $`claude -p ${prompt}`.nothrow().quiet();

    if (result.exitCode !== 0) {
      const err = Err.from("CLI execution failed", { code: "CLI_ERROR" })
        .withMetadata({ exitCode: result.exitCode });

      // DEBUG at origin - captures technical detail
      ctx.log(ctx.log.DEBUG, 'Claude CLI failed', err);
      return fail(err);
    }

    return ok(result.stdout.toString());
  });
}

Intermediate Layer - Wrap Error, No Logging

import { fail, ok, type ResultTuple } from "@pencroff-lab/kore";

async function generateNote(
  section: Section,
  ctx: ExecutionContext
): Promise<ResultTuple<GeneratedNote>> {
  const prompt = buildPrompt(section);

  const [output, cliErr] = await callClaude(prompt, ctx);
  if (cliErr !== null) {
    // Wrap with context, don't log (avoid duplication)
    return fail(cliErr.wrap(`Failed to generate note for "${section.title}"`));
  }

  const [note, parseErr] = parseNoteOutput(output);
  if (parseErr !== null) {
    ctx.log(ctx.log.DEBUG, 'Note parsing failed', parseErr);
    return fail(parseErr.wrap("Failed to parse generated note"));
  }

  return ok(note);
}

Handler Boundary - ERROR Once

const handler = async (argv: Args): Promise<void> => {
  const ctx = createContext(argv, deps);

  const { log } = ctx;

  log(log.INFO, 'Starting note generation', { plan: argv.plan });

  for (const section of sections) {
    log(log.INFO, 'Processing section', {
      index: section.index,
      title: section.title
    });

    const [note, err] = await generateNote(section, ctx);
    if (err !== null) {
      // ERROR at boundary - single log with full wrapped context
      log(log.ERROR, 'Section processing failed', err);
      continue;  // or return based on strategy
    }

    log(log.INFO, 'Note generated', { path: note.path });
  }

  log(log.INFO, 'Generation complete');
};

Dry-Run Logging Pattern

Use a child logger with DRY-RUN module prefix for simulated operations:

import { attemptAsync, ok, type ResultTuple } from "@pencroff-lab/kore";

async function saveNote(
  note: GeneratedNote,
  ctx: ExecutionContext
): Promise<ResultTuple<string>> {
  const path = resolvePath(note);

  if (ctx.dryRun && ctx.dryRunLog) {
    // Use dedicated dry-run logger - message clearly marked
    ctx.dryRunLog(ctx.dryRunLog.INFO, 'Would write note', { path, size: note.content.length });
    return ok(path);
  }

  return attemptAsync(async () => {
    await Bun.write(path, note.content);
    ctx.log(ctx.log.INFO, 'Wrote note', { path });
    return ok(path);
  });
}

// Console output (pretty transport, colors disabled):
// 12:34:56.789 INF [app] [DRY-RUN] [handler] Would write note {"path":"...","size":4521}

Service Logger Injection

Services receive logger through their factory/constructor and create service-specific child loggers:

import { attemptValueAsync, type Logger, type ResultTuple } from "@pencroff-lab/kore";

interface ShellExecutor {
  run(prompt: string, model: string): Promise<string>;
}

interface ClaudeServiceDeps {
  shell: ShellExecutor;
  log: Logger;
  config: LlmConfig;
}

function createClaudeService(deps: ClaudeServiceDeps) {
  const { shell, log, config } = deps;

  // Create service-specific child logger
  const svcLog = log.child('claude');

  return {
    async generate(prompt: string): Promise<ResultTuple<string>> {
      svcLog(svcLog.DEBUG, 'Calling Claude CLI', {
        model: config.model,
        promptLength: prompt.length
      });

      return attemptValueAsync(() => shell.run(prompt, config.model));
    }
  };
}

CLI Initialization with Logger

import { createLogger, fail, ok, type ResultTuple } from "@pencroff-lab/kore";

async function initializeCli(): Promise<ResultTuple<void>> {
  // 1. Create root logger
  const log = createLogger('app');

  // 2. Load config
  const [config, configErr] = await loadConfig();
  if (configErr !== null) {
    log(log.FATAL, 'Config load failed', configErr);
    return fail(configErr);
  }

  // 3. Initialize DB
  const [db, dbErr] = createDatabase(config.paths.workspace_dir);
  if (dbErr !== null) {
    log(log.FATAL, 'Database init failed', dbErr);
    return fail(dbErr);
  }

  // 4. Create handler with log in deps
  const [handler, handlerErr] = createHandler({
    db,
    log,
    config
  });

  if (handlerErr !== null) {
    log(log.FATAL, 'Handler creation failed', handlerErr);
    return fail(handlerErr);
  }

  // 5. Register and run
  await yargs(hideBin(process.argv))
    .command({ command: "study_note from_plan", handler })
    .parse();

  return ok();
}

Execution Context

import type { Logger } from "@pencroff-lab/kore";

interface ExecutionContext {
  runId: string;
  dryRun: boolean;
  log: Logger;               // Run-scoped logger with runId binding
  dryRunLog: Logger | null;  // Only set when dryRun=true
}

Environment Configuration

# Log level (default: info)
# Valid values: trace, debug, info, warn, error, fatal
LOG_LEVEL=debug

The logger reads LOG_LEVEL from the environment at creation time. When not set, defaults to info. The level can also be set explicitly via createLogger options:

import { createLogger, lvl } from "@pencroff-lab/kore";

// Explicit level (overrides LOG_LEVEL env var)
const log = createLogger('app', { level: lvl.DEBUG });

Consumer code never imports lvl — use log.INFO, log.DEBUG, etc. instead. The lvl export is intended for createLogger options only.

The package also exports a default logger instance (log) created with no module name and default options. Prefer createLogger with an explicit module name for production use:

import { log } from "@pencroff-lab/kore";           // Default instance (no module)
import { createLogger } from "@pencroff-lab/kore";   // Named instance (preferred)

const appLog = createLogger('myapp');

Testing with Logger

Logger is silent in tests for v1. Use a mock logger that captures calls for assertion:

import sinon from "sinon";
import type { Logger } from "@pencroff-lab/kore";
import { expectOk } from "@pencroff-lab/kore/test";

export function createMockLog(): Logger {
  const log = sinon.stub() as unknown as Logger;

  Object.assign(log, {
    TRACE: "trace",
    DEBUG: "debug",
    INFO: "info",
    WARN: "warn",
    ERROR: "error",
    FATAL: "fatal",
    child: sinon.stub().returns(log),
  });

  return log;
}

// Usage in tests
const mockLog = createMockLog();
const deps = { log: mockLog, db: mockDb, config: testConfig };

const handler = expectOk(createHandler(deps));
await handler({ plan: "./test.md" });

sinon.assert.calledWith(
  mockLog as unknown as sinon.SinonStub,
  mockLog.INFO,
  sinon.match("Starting")
);

Naming Convention

Logger is a function type — use short names: log, ctx.log, svcLog, runLog.

Context	Name	Example
Root / deps	`log`	`deps.log`, `const log = createLogger('app')`
Run-scoped child	`runLog`	`const runLog = log.child('handler', { runId })`
Service child	`svcLog`	`const svcLog = log.child('claude')`
Dry-run child	`dryRunLog`	`const dryRunLog = runLog.child('DRY-RUN')`
Execution context	`ctx.log`	`ctx.log(ctx.log.INFO, 'Processing')`
Test mock	`mockLog`	`const mockLog = createMockLog()`

Summary

Inject log via deps - never import a global singleton
Use child loggers for run correlation (runId) and module scoping
DEBUG at origin, ERROR at boundary - avoid duplicate error logs
Wrap errors in intermediate layers - add context without logging
Dedicated dry-run logger - prefix with DRY-RUN module for simulated ops
FATAL for process exits - config load, DB init, handler creation failures
Silent in tests - mock log captures calls for Sinon assertions
Short names - log, ctx.log, svcLog, runLog (not logger)
Levels on the function - use log.INFO, log.ERROR etc., never import lvl in consumer code

Migration from throwing Documentation guide