Skip to content

ccch1mneyyy/dsh-TUI / .agents/skills/review

review

Verified

Review or de-slop concrete changes in ccch1mneyyy/dsh-TUI at maintainer level: PR numbers or URLs, branches, commit ranges, patch files, staged or unstaged worktrees, and scoped repository-hygiene requests. Use for evidence-first correctness, contract, repository-rule, process, and behavior-preserving cleanup review, including debug residue, dead or duplicate code, speculative abstractions, stale comments or tests, and unrelated diff churn. Do not use for abstract designs with no concrete repository artifact; use audit for repository-wide assessments and vuln-check for security-only checks.

FileKindSize
SKILL.mdskill md10.5 KB
references/contract-gates.mdmarkdown5.0 KB
references/deslop-gates.mdmarkdown5.9 KB
references/evidence-base.mdmarkdown7.2 KB
references/redlines.mdmarkdown4.8 KB
references/report-contract.mdmarkdown3.1 KB
references/verification-map.mdmarkdown5.1 KB
scripts/contract-snapshot.mjsexecutablescript15.6 KB
scripts/scan-diff-hygiene.mjsexecutablescript9.6 KB

references/verification-map.md

View on GitHub
# dsh-TUI 验证映射
先读当前 package.json 的 scripts、.github/workflows/ci.yml、脚本头部、docs/contributing.md;本表仅选择方法,漂移依仓库真源并记报告。

## 禁止虚构
根级 test/lint 禁令以 SKILL.md “全局不变量与模式”为唯一真源。scripts/ 混有回归/复现/探针/迁移/取证,不全目录当测试跑;跑前读头部辨 import src/ 或 lib/types/、普通 `node` 或 `node --import tsx/esm`;只记实跑命令/退出码/SHA/关键输出,不将应通过写成已通过。

## 最小通用关口
源码改动先从当前 scripts 核 `pnpm compile`、`pnpm verify:build`、`pnpm verify:package`、`pnpm verify:bun-package`(仅 scripts/CI 仍定义时)的存在及真实组成,不机械全跑;纯文档/workflow/YAML 依贡献指南分流,verify:build 按编译依赖链顺序。`pnpm build` 通常聚合 compile/门禁,须展开真实结果、不重复计独立证据;required checks/平台 job 从当前 CI 提取,不手抄旧表。

## 改动面 → 验证类别
| 改动面 | 至少核对 |
|---|---|
| package exports/bin/dependencies | manifest gate、clean compile、package tarball、入口 smoke、Bun 目标(若适用) |
| adapter import 或上游版本 | boundary、contract、manifest-deps、alpha/source/patch/web coexistence 等当前镜像门禁 |
| cordis.patch.yml | patch-surface 复算、真实上游应用语义 |
| session/channel submit/steer/pending | 对应 channel 聚焦脚本、恢复/rewind/fork 与事件白名单 |
| compaction/transcript folding | compact、session switch、resume 与行投影回归 |
| Chat/按键/模态优先级 | keymap、受影响 modal、输入草稿、help/overlay 让位与泄漏 |
| 共享渲染、消息列表、工具卡 | 当前 CI UI 回归、scroll/resticky、窄终端、inline/fullscreen |
| ask/question/approval | askpanel、layout、abort/consumed、真实 store 快照 |
| theme/i18n | theme runtime/persistence、i18n key/type、双语文档和环境固定 |
| Ink/Yoga/滚动/命中测试 | renderer 专用 verify/repro、resize、selection、pointer、真实终端/伪终端 |
| Windows/POSIX 路径、spawn、socket | 平台 job、含空格路径、UTF-8 byte 预算、权限位守卫、Windows 实机 |
| update/download/archive | checksum、redirect/DNS、检查与使用间竞态、symlink、原子替换、平台资产 |
| plugin/extension/admission | spec、grants、storage、messages、ledger、commands、negotiation、lifecycle 当前门禁 |
| workflow/CI gate | actionlint/zizmor(若仓库采用)、permissions、fork trust、fail-open、路径过滤、required-check 聚合 |
| 维护技能 .agents/skills/ | frontmatter、相对引用、脚本 self-test、不入 npm package files |

具体脚本名从上列当前包清单/贡献指南搜索,不按表猜。

## 选择算法
依次:diff 列行为/owner/公共契约/用户可见面 → 各找最近已有回归 → 核 CI/聚合调用或明本地补证 → 先最小聚焦,再依失败/风险扩大聚合 → 终端可见/平台敏感补真实演练(无头绿不足)→ 报告分静态证明/自动回归/人工演练/未执行。

## 不可信 fork 执行门
本节是 fork 执行门唯一真源:head 任意命令可能执行贡献者改过的 package lifecycle/scripts、verify/repro、编译插件/loader、workspace/link 依赖、CI helper。默认静态;须用户明确许可,隔离工作树/容器,无仓库 secrets/云凭据/SSH agent/浏览器 cookie,只用最小只读凭据或无凭据,限网络/CPU/内存/磁盘/时间,不挂敏感 HOME,先审 package lifecycle 与脚本 diff 再跑。

## 反假绿检查
### 坏基线
回归须在未修实现失败;优先隔离树反向 patch、最小故障注入破被保护不变量、对照最新目标分支/明确坏 commit;不改坏共享树、不破坏性 reset 回滚。

### 独立 oracle
oracle(判定依据)不以实现同 helper 算 expected、不复制实现或同源自比;不恒真/只验符号字符串存在、不用本地化文案/随机内容/宽松子串替结构断言;事件/exports/features 白名单直比官方/目标全集;渲染用稳定区/语言无关哨兵;异步轮询不变量/事件,不固定 sleep 碰运气。

### 真实路径
按脚本设计 import 上列真实源码或干净构建产物,走真实 registration/open/subscribe/dispose;mock 仅替不可控外部边界且形状符当前上游类型;读写隔离临时目录、恢复 env/listener。

### CI 挂载
脚本存在不证保护:须 package script/workflow 实引、路径过滤不误跳、上游 failed/cancelled/空输出时聚合 fail-closed、最新 head checks 非空且符当前 SHA。

## Deslop apply 验证
前后验证编排唯一真源为 SKILL.md 阶段 6,命令仍依本表;删除测试/门禁不算清理,须先证独立覆盖仍在。

## 报告记录模板
| 命令/核验 | 类型 | SHA | 结果 | 说明 |
|---|---|---|---|---|
| `...` | static / regression / manual | `...` | exit / pass / fail / blocked | 失败归因、环境、覆盖不变量 |

静态确定的阻断机制直接报告,不藏“建议补跑测试”;测试用来证明影响/修复。
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall