nnabuuu/harness-engineering-toolkit
Skill Claude CodeCodex
Scan a codebase to detect the five most common harness self-deceptions, generate a narrative diagnostic report, and optionally auto-fix findings. Based on the 4-Layer Harness Model. Trigger: /harness-audit, "audit my harness", "scan my project", "harness 审计", "检查我的 harness", or when user wants an automated assessment…
nnabuuu/harness-engineering-toolkit
Skill Claude CodeCodex
End-to-end harness creation: interview → spec → runnable project. Two entry points: (A) start from scratch — interviews the user, produces HARNESSSPEC.md, then builds; (B) start from existing HARNESSSPEC.md — skips straight to build. Three build modes: document, code, investigation. Triggers: 'plan harness', 'build…
nnabuuu/harness-engineering-toolkit
Skill Claude CodeCodex
Retrospective on completed harness tasks. Analyzes run data (scores, eval reports, changelogs) to identify convergence issues, bottleneck dimensions, repeated work, and cost inefficiency. Generates concrete prompt/rubric improvements. Optionally archives completed tasks. Standalone archive mode available — skip…
nnabuuu/harness-engineering-toolkit
Skill Claude CodeCodex
Interactive harness maturity diagnostic using the Inversion pattern. Does NOT ask "do you have X" — assumes you do, then probes until the gap reveals itself. Surfaces the five most common self-deceptions in harness engineering. Works for ANY industry. Trigger: /harness-self-check, "evaluate my harness", "check my…
nnabuuu/harness-engineering-toolkit
Skill Claude CodeCodex
Audit the skills installed in a harness and decide, for each one, whether to retire, trim, or keep it. Use this whenever the user wants to clean up skills, asks whether a skill is still worth its tokens, complains that skills fire too often or burn usage limits, mentions that a model upgrade made their agent slower or…