一键导入
neural-address-review
Execute all fixes from a REVIEW.md — address warnings, blocking issues, and gaps found during review
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。
菜单
Execute all fixes from a REVIEW.md — address warnings, blocking issues, and gaps found during review
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 | neural-address-review |
| description | Execute all fixes from a REVIEW.md — address warnings, blocking issues, and gaps found during review |
You are executing the fixes identified in a REVIEW.md produced by review.
.neural/wip/.$ARGUMENTS matching a feature name, use that one..neural/wip/<feature>/REVIEW.md. If it does not exist, stop and tell the user to run /neural:neural-review first..neural/wip/<feature>/CONTEXT.md, .neural/wip/<feature>/PLAN.md, and any ADRs under .neural/wip/<feature>/docs/adr/. Fixes must honor those decisions.Extract all issues from REVIEW.md:
Also extract:
⚠️ partial or ✗ missing from the Completion Score table.PARTIAL or FAIL.For each issue to address, create a concrete fix task:
Fix Plan:
1. [BLOCKING] <issue> → <what to do> in <file>
2. [WARNING] <issue> → <what to do> in <file>
3. [PARTIAL] Task N: <what's missing> → <what to do>
4. [FAIL] Truth "<truth>": <what failed at which level> → <what to do>
Show this plan to the user and wait for confirmation before executing.
Apply fixes one at a time, in the order listed:
fix(<feature>): address review — <short description>If the same fix fails 3 times with materially different approaches, stop and report it unresolved — do not loop. Surface it to the user as a remaining blocking item.
After all fixes are applied:
Print a summary:
Address Review Complete:
Fixed: N blocking issues, M warnings
Remaining: K items (skipped by user)
Then ask: "Delete REVIEW.md? (y/n)" — if yes, delete it.
Finally suggest: "Fixes applied. Run /neural:neural-review again to verify, or /neural:neural-archive if you're confident."
Implementation planning with optional adversarial cross-review (Claude Code ⇄ Codex). Tasks are sequential vertical slices, each carrying its own testable behaviors. Pass --visual to also render the plan as a self-contained HTML page (PLAN.html). Pass --skills <skills> to preload skills that shape the plan and that execute loads before coding
Test-driven execution loop — one task at a time, red→green→refactor, atomic commits
Plan vs implementation verification with goal-backward analysis and test-quality audit
Move completed features from wip to archive
Socratic interview that captures domain language, decisions, and ADRs inside .neural/wip/<feature>/
Harvest knowledge from archived features into .neural/knowledge/ — run after archiving a feature, or invoke manually to rebuild the project knowledge base