菜单
Fork-style adversarial review workflow for the current diff or a proposed code change. Use to find correctness, regression, security, performance, maintainability, test, and documentation issues before final delivery.
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
Metric, SQL, dbt, dashboard, semantic-layer, and BI validation discipline for high-stakes analytics work. Use for grain, joins, nulls, filters, freshness, lineage, reconciliation, dashboard semantics, and decision relevance checks.
Compact task-state and handoff ledger for long Opus 4.8 High sessions, subagent orchestration, compaction, and resumable workflows. Use to preserve decisions, files, commands, evidence, risks, and next actions while dropping noisy logs.
Classify high-rigor work into S0-S4 scope levels and choose the minimal sufficient subagent set. Use for ultracode-high orchestration, complex coding, analytics, BI, strategy, writing, audits, migrations, and multi-domain tasks.
Evidence, claim, assumption, and uncertainty ledger for high-rigor agent work. Use when synthesizing research, analytics, code review, strategy, or writing so important claims are labeled as verified, calculated, inferred, assumed, or unknown with source evidence.
Final response contract for ultracode-high and high-rigor subagent workflows. Use to synthesize work into a relayable result with changes, verification, final-gate improvements, risks, and next actions.
Gate vocabulary and runner support for ultracode-high. Use to define precondition and completion gates, prefer deterministic command/file/grep gates where possible, flag attestation gates honestly, and run gates with the bundled gate_check.py helper.
| name | high-code-review |
| description | Fork-style adversarial review workflow for the current diff or a proposed code change. Use to find correctness, regression, security, performance, maintainability, test, and documentation issues before final delivery. |
| model | claude-opus-4-8 |
| effort | high |
| context | fork |
Review as a code reviewer. Findings first, summaries second.
This skill is part of ultracode-high. For the full orchestration map, tool
catalog, agent roster, and card schemas, read
../ultracode-high/references/runtime-context.md when available. Your findings
are not prose polish; they are candidate blockers with mechanisms and checks.
severity: critical | high | medium | low | suggestion
file:
line:
mechanism:
impact:
recommendation:
verification:
Prefer fewer, verified findings over broad speculation. If no issues are found, say so and name residual test gaps.