Measure and tune the AI SDLC harness so the pipeline delivers on a cheap model — run scenario × model evals against the sandbox fixture, then fix the failing instruction rather than raising the model tier. Use when a task only works on the expensive model, when adding or changing a playbook or rule, when per-PR cost is climbing, or when writing a new eval scenario.
Install
npx skills add emgiezet/aisdlc --skill harness-evalnpx skills add emgiezet/aisdlc --skill harness-eval -a claude-code -gnpx skills add emgiezet/aisdlc --skill harness-eval -a codex -ggemini skills install https://github.com/emgiezet/aisdlc.git --path plugins/sdlc/skills/harness-evalgh skill install emgiezet/aisdlc harness-evalgit clone --depth 1 https://github.com/emgiezet/aisdlc /tmp/emgiezet__aisdlc
mkdir -p ~/.claude/skills ~/.agents/skills
cp -R /tmp/emgiezet__aisdlc/plugins/sdlc/skills/harness-eval ~/.claude/skills/harness-eval # Claude Code
cp -R /tmp/emgiezet__aisdlc/plugins/sdlc/skills/harness-eval ~/.agents/skills/harness-eval # Codex, Cursor, Gemini CLI, Copilot, OpenCode