菜单
Periodic review of claudemax's own scaffolding against the current Opus capability. Flags load-bearing vs vestigial pieces so the harness stays sharp as models improve. Run after every major model release or quarterly, whichever comes first.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | harness-audit |
| description | Periodic review of claudemax's own scaffolding against the current Opus capability. Flags load-bearing vs vestigial pieces so the harness stays sharp as models improve. Run after every major model release or quarterly, whichever comes first. |
Models improve. Yesterday's helpful scaffolding becomes today's friction. This skill audits claudemax against the current model and surfaces what should be deleted, simplified, or kept.
docs/SKILL_CATALOG.md. Flags any skill whose purpose now overlaps another (e.g., a primitive the umbrella has subsumed).packages/runtime/src/prompts.ts. Flag any prompt > 1500 chars whose every clause is not load-bearing for the current Opus baseline.packages/core/src/router.ts heuristics against the current Opus default capabilities. Flag rules that no longer change behavior on the current model..claude/hooks/ invocation against current-model failure modes. Flag hooks that haven't fired in N runs and modes that have started slipping through.memory.runs.description: should still reflect what the skill actually does. Flag drift.A HARNESS_AUDIT.md artifact in the repo root with sections:
## load-bearing (keep)
- <item> — <one-line reason it's still earning its place>
## redundant (delete or merge)
- <item> — <what overlaps it, why deletion is safe>
## simplify
- <item> — <what to cut from it without losing the load-bearing piece>
## measurement gaps
- <thing we cannot decide without data> — <what to instrument>
The output is human-reviewed; this skill does NOT auto-delete anything.
/harness-audit --against opus-X.Y).cmax doctor flags credit consumption climbing without proportional effectiveness gain.docs/SOTA_2026.md).The Anthropic harness-design guide explicitly recommends: "strip away pieces that are no longer load-bearing to performance" as models improve. A harness that only grows is a harness that gets in its own way. /harness-audit is the codification of that subtraction discipline.
Ask. Achieve. The canonical user-facing entry to claudemax — describe your goal, the SOTA-2026 pipeline (deepresearch → multispec → parallel /goal → verify) handles the rest. Same engine as /cmax with a friendlier verb. CLI equivalent is `cmax ask "<goal>"`.
The default FAT umbrella. Auto-runs deepresearch + multispec decomposition + parallel /goal + verify. Opus for planning/judgment/verification; Sonnet for sub-Spec execution (escalates per router). Multispec engine selects Mode A (SDK subagents) or Mode B (Claude Code Agent Teams) automatically based on work size. Your default daily-driver.
Max-effectiveness umbrella. Opus for EVERYTHING — planning, decomposition, sub-Spec /goal execution, verification. Use for novel domains, security/auth/payments, or when you want the highest ceiling regardless of cost.
Show the model-routing decision for a task (Fable/Opus/Sonnet/Haiku) with reasoning. Use before launching a packet to sanity-check the tier and to override when you have stronger judgment than the heuristic.
Reviews a proposed diff for correctness, security, style, taste alignment. Distinct from /audit (which reviews existing code, not diffs) and /verify (which checks SPEC completion).
Spec quality gate. Every completion condition must have a mechanically-checkable verifyHint (file path, command, test name, behavior). Blocks /goal handoff if the spec wouldn't be verifiable.