What problem does it solve? AI coding agents drift when they work from model memory instead of grounded rules, producing fabricated APIs, unverified claims, and regressions. This Skill forces every session to load the invariant map and a structured repair loop before any design or edit happens. ## Core Features & Use Cases - Harness-first boot sequence: Loads the invariant map (.hydraAgent/policies/backs-invariants.yaml), the dev-mode playbook, and companion skills (the_path, elite_build_understanding) before any work begins. - Seven-step repair loop: Reproduce on live truth, write a failing RED contract test, fix the flaw class at the seam, verify on the real path, get an independent grade from a different model family, check concurrent sessions, then land. - Anti-hallucination discipline: Requires resolving every symbol against the installed artifact, falsifying your own change, and never claiming green off a mock or proxy. - Use Case: Before fixing a bug in a shared repo, invoke this Skill so the agent reproduces the failure live, writes a regression test that fails for the real reason, fixes the entire flaw class, and reports PROVEN or STILL-BUILDING. ## Quick Start Invoke the optimus skill to boot the harness-first dev-mode loop before starting any coding task in this repository.
