Quick harness comparison: clawd-code vs openai/codex

clawd-code

• Small top level
• Mainly src/ + tests/
• Python-first
• README frames it as a clean-room Claude Code porting workspace
• Current emphasis is manifesting, parity audit, command/tool inventories, and routing

openai/codex

• Big monorepo
• codex-cli/ packaging layer
• codex-rs/ real Rust implementation
• docs/ is extensive
• Clearly organized around a maintained, installable product

clawd-code

The README explicitly says the repo is now a Python-first porting workspace rather than a tracked snapshot. The CLI entrypoint in src/main.py is mostly for:

summary
manifest
parity-audit
subsystems
commands
tools
route
show-command
show-tool

That strongly suggests a harness skeleton focused on representation, inspection, and porting progress rather than a fully polished runtime.

openai/codex

The docs and package manifests say “shipping CLI product” everywhere. The Rust README makes the architecture explicit:

core/   business logic
exec/   headless automation CLI
tui/    fullscreen TUI
cli/    command multitool wrapper

That is a much more mature and productionized split.

clawd-code

• Interpretable
• Compact
• Easy to scan quickly
• Likely easier to hack on fast
• Looks unfinished as a complete end-user harness

openai/codex

• Industrial
• Heavier operational surface
• Much richer docs around config/auth/sandbox/MCP
• Clearly designed for release engineering and real users
• Harder to absorb quickly because of repo size/complexity

Use clawd-code if you want to...

• understand the harness ideas quickly
• inspect a smaller porting workspace
• hack on Python metadata/parity/runtime scaffolding

Use openai/codex if you want to...

• study a mature shipping agent CLI
• see how product-grade sandbox/TUI/exec/docs are organized
• learn from a real release-oriented harness stack