Ship First, Verify Later

Your coding agent has been lying to you.

That's the implicit thesis of forward-implementation-first, a new skill for Claude Code, Codex, and other agents that hit 143 stars on GitHub in under 48 hours. The pitch is brutal and specific:

"Stop your coding agent from stalling real work on self-invented bookkeeping — receipts, hashes, locks, certification rituals. Ship first, then verify."

I've watched this happen. You ask an agent to build a feature and it burns 40% of its context window justifying how it's going to do it, generating a "certificate of completion" it invented on the spot, or running unnecessary validation loops. The agent is performing productivity theater.

The fix is a single instruction layer that overrides that behavior. The skill injects a directive: implement now, audit later. It's not anti-quality — it's anti-stall. It strips away the meta-work agents invent for themselves and forces them to produce.

This matters because the biggest bottleneck in agent-assisted development isn't model capability anymore. It's agent behavior. And this is one of the first tools I've seen that directly addresses the meta-problem: agents that look busy instead of being productive.

The 143 GitHub stars in a weekend suggest I'm not the only one who's noticed.