I will set up and harden your claude code or codex coding agent workflow

I set up and harden coding-agent workflows for teams running Claude Code or Codex.

What you get, fixed scope and in writing:

- Harness audit: a read of your agent configuration and recent sessions, a findings list ordered by what each failure costs you, and the fix for each.

- CI hardening and provenance: tests across the platforms you support, lint, types and security scanning as separate checks, a daily dependency audit, workflow linting, releases with checksums and build attestations.

- Harness build: instruction files, a skill catalogue for the procedures your team repeats, hooks that refuse the tool calls which break a rule, commit and PR conventions, an attribution trail for agent-written commits.

Delivery is asynchronous: commits on a branch in your repository, one logical change each, with a running note and a written runbook at handover.

Background: 442 pull requests merged into 60 external repositories, directed through coding agents, in Rust, Go, Python, Java and C; author of self-hosted usage dashboards for Claude Code, Codex and Kimi, and a browser bridge for coding agents.

Send the repository, the team size, the agents you run, and what is going wrong.