I noticed the clawdbot invocation getting slower over the last couple of weeks, especially in weak hardware. Can add up quickly if repeatedly running commands

A few example benchmarks running on a cheap VPS 2 core / 2GB:

Summary

• Introduced lightweight CLI routing and config guards
• Tightened logging imports and lazy logger initialization
• Expanded focused tests for argv, env, frontmatter, utils, and subclis
• Standardized truthy env parsing and boolean handling across CLI/runtime

Per-file Changes

Overall Improvements

• Consistent env flag parsing across CLI, gateway, memory, and live-test toggles via parseBooleanValue() and isTruthyEnvValue()
• Clear separation of config validation through a shared config-guard instead of implicit pre-action hooks
• Faster CLI paths for common read-only commands by routing early and lazily registering subcommands
• More robust logging setup through direct subsystem imports and lazy console logger initialization

Reasoning

• Centralizing boolean/env parsing removes duplication and inconsistent truthy handling, and makes supported values explicit and test-backed.
• The config guard provides predictable config validation/migration behavior per command, without global side effects from pre-action hooks.
• Lazy subcommand registration and route-first paths reduce CLI startup time, especially for read-only and status-style commands.
• Logging import changes eliminate runtime failures and circularities caused by barrel exports.

Tests

• pnpm vitest run boolean.test.ts env.test.ts frontmatter.test.ts frontmatter.test.ts utils.test.ts argv.test.ts register.subclis.test.ts embeddings.test.ts globals.test.ts gmail-setup-utils.test.ts memory-cli.test.ts
• pnpm vitest run argv.test.ts register.subclis.test.ts
• pnpm lint && pnpm build && pnpm test

> oxlint --type-aware src test

Found 0 warnings and 0 errors.
Finished in 1.2s on 2236 files with 104 rules using 12 threads.

 Test Files  774 passed | 1 skipped (775)
      Tests  4036 passed | 7 skipped (4043)
   Start at  16:05:35
   Duration  141.35s (transform 11.32s, setup 212.33s, import 158.99s, tests 119.97s, environment 45ms)

• I've been running this version and testing as many commands as possible and just talking to multiple agents normally and running tasks

AI Assisted: GPT-5.2 Codex

To hopefully give some intuition and to make sure I validated routes correctly:

CLI entrypoints:

• Binary entry: entry.ts
• Main CLI runner: run-main.ts runCli().

Routing diagrams (current):

1. Top-level invocation

clawdbot
  -> [src/entry.ts]
      - profile/env prep + respawn guard
      -> import [src/cli/run-main.ts]
          -> runCli()

2. Fast-path routing (NEW, faster for read-only commands)

runCli()
  -> tryRouteCli() in [src/cli/route.ts]
      - health/status/sessions/agents list/memory status
      - runs command directly, returns early
  -> (if routed) exit

Why faster: avoids building Commander program + importing all subcommands.

3. Full CLI path (unchanged behavior, but LAZY subcommands now)

runCli()
  -> buildProgram() in [src/cli/program/build-program.ts]
      -> registerSubCliCommands() in [src/cli/program/register.subclis.ts]
          - placeholders only
          - real subcommand imported on demand
  -> program.parseAsync(...)

Why faster: only loads the chosen subcommand instead of importing all.

What’s new/different/faster:

• NEW: Fast-path routing for read-only commands in route.ts.
  Faster because it bypasses Commander setup and lazy imports entirely.

• NEW: Lazy subcommand registration in register.subclis.ts.
  Faster because only the invoked subcommand’s module tree is loaded.

• NEW: Deferred plugin registry loading via channel-options.ts and plugin-registry.ts.
  Faster because plugin discovery is avoided unless explicitly needed.

• NEW: Config caching in io.ts.
  Faster because repeated config reads in short-lived runs reuse cached content.

Summary flow (combined):

clawdbot
  -> [src/entry.ts]
  -> [src/cli/run-main.ts]
      -> tryRouteCli() [NEW fast path]
      -> else buildProgram() -> lazy subcommands [NEW lazy load]