Christoffer91/ai-coding-skills

Fair-but-rigorous critical-friend pass on an idea, plan, prompt, or architecture before you commit — steelman, challenge, calibrated verdict. Modes: rubber-duck / critique / hard-challenge / decision. Can fan out parallel subagent lenses and pull an independent second-model opinion.

Evidence-driven debugging: a cheap systematic single-pass loop that escalates hard bugs to a dual-brain council — Claude and Codex CLI research independently, hypotheses are falsified, and candidate fixes race in disposable worktrees. Token-conscious with a mandatory usage report.

Dual-brain plan → execute → review → ship loop: Claude plans and reviews the PR, OpenAI Codex CLI critiques the plan, writes the code, opens the PR, and applies review edits. Deploy is risk-gated. Skill only — run orchestrate/install.sh from a clone for the dashboard, driver, and Codex-side skill.

Standard delivery pipeline for non-trivial work: coverage matrix (security/risk/review/tests/docs), adaptive routing, verification gates, and PR-ready output. Routes real implementation into the orchestrate loop.