Harnesses

A harness is the runtime around a model — it doesn't supply raw intelligence (that's the model's job), it amplifies one. So we score the five things a harness actually controls:

• Helmet · Intelligence — how well it orchestrates the model (context engineering, planning, multi-step reasoning scaffolds).
• Chest · Autonomy — how far it runs unattended: one-shot → multi-step → self-verifying loop.
• Shoulders · Arsenal — the tools it gives the model: files, shell, browser, MCP, extensions.
• Torso · Reliability — how it recovers from errors and stays on-rails (gates, sandboxing, retries).
• Legs · Reach — how much context / codebase it can work across (indexing, repo-maps, retrieval).

Those five are editorial: judged 0–10 against each harness's documented capabilities on a fixed rubric. Buzz is the one hard, measured metric — the live mention count for each harness across WOLFSKIND BOT's own Source Network (our 1,500+ account X scan), computed by our own pipeline, so it reflects real practitioner attention rather than opinion.