I built agentcli-go to make CLI projects deterministic and AI-agent friendly.

  Instead of starting from scripts, it scaffolds a Go CLI and includes harness engineering from day 1:

  - doctor for contract checks
  - smoke for basic end-to-end sanity
  - verify for full validation loop

  Quickstart:

  go install github.com/gh-xj/agentcli-go/cmd/agentcli@latest
  mkdir -p /tmp/agentcli-demo && cd /tmp/agentcli-demo
  agentcli new --module example.com/demo demo
  agentcli add command --dir ./demo --preset file-sync sync-data
  agentcli doctor --dir ./demo --json
  cd demo && task verify

  Goal: reduce script drift and make automation reliable for both humans and coding agents.

  I’d love blunt feedback on:

  - where this is too opinionated
  - what extension points are missing
  - whether harness engineering should be mandatory or opt-in

  Repo: https://github.com/gh-xj/agentcli-go

  First comment
  If you’ve tried AI-assisted automation, what failed first in practice: scaffolding, validation contracts, or iterative bug-fix loops? I’m trying to make those failure modes explicit in the framework.