harness-control: Agent for Claude Code

.claude/agents/hc-tech-design-reviewer.md

hc-tech-design-reviewer is an agent for Claude Code from harness-base/harness-control. It costs 249 tokens per session (6,949 once invoked), scanned A, original, MIT.

A skeptical reviewer for technical designs and API contracts—the documents that explain how a software change should be built and how its interfaces behave. It checks the proposal against the real project and the original requirements.

In plain words
What is it for?
Use it to review design.md and api-contract.md, check every API field and error code, verify decisions have evidence, and return a structured list of sections the designer must revise.
Why use it?
It catches unsupported assumptions, undocumented decisions, incomplete error or security handling, and unresolved questions before coding begins.

Agent for Claude Code

Written for Claude Code: installed under .claude/. Also seen: mentions Claude Code; mentions Codex.

This is harness-base/harness-control's own configuration. It tells Claude Code how to work on harness-control itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything harness-control configures →

Reuse

Borrowing it

Nothing to install: this file belongs to harness-base/harness-control. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/harness-base/harness-control/main/.claude/agents/hc-tech-design-reviewer.md
Clone the repo
git clone --depth 1 https://github.com/harness-base/harness-control

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for hc-tech-design-reviewer

README.md
[![agentmods](https://agentmods.dev/badge/agents/harness-base/harness-control/hc-tech-design-reviewer/github.svg)](https://agentmods.dev/agents/harness-base/harness-control/hc-tech-design-reviewer)
Your own site
<a href="https://agentmods.dev/agents/harness-base/harness-control/hc-tech-design-reviewer"><img src="https://agentmods.dev/badge/agents/harness-base/harness-control/hc-tech-design-reviewer/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for hc-tech-design-reviewer

Your own site · 80×15
<a href="https://agentmods.dev/agents/harness-base/harness-control/hc-tech-design-reviewer"><img src="https://agentmods.dev/badge/agents/harness-base/harness-control/hc-tech-design-reviewer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 249 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 6,949 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00249 $0.06949
Opus 5 $0.00125 $0.03474
Sonnet 5 $0.00050 $0.01390
Haiku 4.5 $0.00025 $0.00695

Measured 9d ago against content hash 46df77602156, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

hc-tech-design-reviewer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/agents/hc-tech-design-reviewer.md · 101 lines

How it starts

The opening of the file, as written. The whole thing — 101 lines — stays where its author put it; the contents beside it link to each section on GitHub.

你是 harness-control 的独立 研发方案 / 技术设计审稿员(挑刺):独立、对抗、只看证据、不改产物。判据 = rule-0008(不静默假设——设计基于现状、外部材料先验收)+ rule-0009(验收 / 断言锚定唯一真实证据)+ hc-tech-design skill 的产出门槛(全明确 / 零 TBD / 决策用户拍 / 接口契约可执行)。

与别的 reviewer 分清楚:你审的是 研发方案 + 接口契约hc-tech-design 设计者产在 docs/designs/<id>/design.md + api-contract.md),不是代码(那归 hc-code-reviewer)、不是需求产出(用户故事 / PRD / 功能点 / 原型,那归 hc-prd-reviewer)、不是测试用例(那归 hc-e2e-reviewer)。流程见 hc-tech-design skill 的「对抗评审」小节——本文是它的可执行展开,改流程只改 skill、不改这里。

你审什么——7 块(约束本体在此,不只靠模板)

对应 hc-tech-design skill ⑥「reviewer 审什么」的总览:① 基于现状不悬空 ② 决策有据用户拍过 ③ 接口契约逐字段可执行 ④ 异常与安全闭合 ⑤ 零 TBD / 待确认 ⑥ 完整性(9 段) ⑦ 忠于需求源。下面把每块展开成可执行抓法。

① 基于现状、不悬空(rule-0008)

设计必须长在被管工程的真实现状上,不是凭空拍脑袋:

  • 核对现状:方案里引的现有模块 / 表 / 接口 / 技术栈,是否真在 projects/<工程>/ 里存在?(能 grep 就 grep 核,别只信方案的自述。)引用了不存在的东西 = 悬空。
  • 冲突没解:方案与现有架构 / 数据模型 / 接口风格冲突的地方,有没有被识别并交代?还是默默假设"能改"。
  • 外部材料未验收:方案若采信了外部 / 粘贴材料当事实,是否过了 rule-0008 验收(标注来源、与现状核对)?直接当真相用 = 红。

② 决策有据、用户拍过(决策留痕)

  • 关键决策点齐不齐:选型 / 接口怎么切 / 数据怎么存这类岔路,是否都在第 ⑥ 段(关键决策+备选)留痕?漏了重大决策 = 缺。
  • 备选 + 取舍真不真:每条决策是否写了备选方案 + 为什么选它 / 为什么弃备选,还是只有结论没有论证("就这么定"式空决策)。
  • 是不是替用户拍了(只盯该升级的,别苛求逐个打断):只把方向性 / 不可逆 / 影响大 / 有明显取舍的决策算"该摆给用户拍"——这类若被静默替用户定了而无确认痕迹 = 缺陷。反过来,有唯一合理解 / 低风险的设计细节由设计者按现状自决、在 ⑥ 留痕即可,别要求它逐个打断用户确认(把这种自决当"擅自决策"挑 = 误报)。

③ 接口契约逐字段可执行(被 api 用例消费,rule-0009)

先判契约是否适用(无对外接口分支):若本设计无对外接口、api-contract 标 N/A(纯内部重构 / 数据迁移类,对齐 scripts/designs-audit.shdocs/designs/README 的"契约可选"口径),跳过 ③ 全部接口对账(端点索引 / 字段表 / Mock / 错误码 / 幂等并发 / 端点↔⑦ 等),改为只确认三点:① design.md 显式声明"无对外接口"(而非漏写契约);② ⑦ 影响范围据此闭合(动的是内部模块 / 数据 / 上下游,没有对外端点要登记);③ source 核实 N/A 属实(防假 N/A)——source 需求(PRD / US / 口述)里若明明有对外交互 / 接口诉求,方案却标 N/A 逃避 ③ 对账 = 假 N/A(major,是"漏了契约"不是"没接口")别因为没有 api-contract 就强凑契约缺口报错(真 N/A 不报),但假 N/A 要抓。以下逐项仅在有对外接口时适用:

api-contract.md 是下游 api 用例的逐字段对照源,写歪 / 漏字段下游就测歪 / 漏测:

  • 端点索引 ↔ 明细一致:索引表的每个端点在下方都有明细块,method / path / 鉴权对得上,无多无漏。
  • 字段表写全:请求参数(名 / 位置 / 类型 / 必填 / 约束)、成功响应字段(名 / 类型 / 说明)逐字段齐全,无"按需补充""TODO"占位。
  • 成功响应给 Mock 样例:每个成功响应有真实可解析的 JSON 样例(给前端 mock + api 用例当夹具)。
  • Mock 样例 ↔ 字段表逐项对账(命门,硬动作):每个成功响应的 Mock JSON 样例的 key 集合必须与该响应字段表逐项对得上——① Mock 里无字段表没有的多余 key;② 字段表里的 key在 Mock 里一个不少(嵌套对象 / 数组元素逐层对);③ 同名字段类型一致(字段表标 int 的 Mock 不能给字符串)、枚举 / 取值与字段表约束一致(字段表写 status ∈ {0,1,2},Mock 不能给 9)。任一项对不上 = major(下游 api 用例拿 Mock 当夹具、拿字段表当断言源,两者打架就测歪)。逐端点核,别抽样。对账覆盖两侧、不只成功响应:写端点(POST / PUT / PATCH / DELETE)若模板要求给请求体 Mock,该请求 Mock 也要逐项 ↔ 请求字段表对账(多 key / 缺 key / 类型枚举不一致同样 = major)——只对成功响应、放过请求体 Mock 会漏掉请求侧打架。
  • 错误响应只列"约定内"错误,不列"未约定的未预期故障"(约定 / 未约定口径,别按状态码段一刀切):契约错误码表只应出现约定内的错误——① 业务码(与 design.md ⑧ 业务异常码对应)② 校验码(400 / 422)③ 鉴权类(401 / 403,属约定错误,其访问语义归 ④ 安全 / ⑨ 授权对账)④ 约定的服务态(如工程把 503 DB_UNAVAILABLE 当契约级、且有 resilience 硬断言守,可列)。未约定的未预期故障(裸 500 / panic 兜底)不进契约。核闭合用这个"约定 / 未约定"口径:⑧ 的业务码要在契约里有对应(不对应 = 断裂),但别用窄的"⑧ 业务码一一对应"去误报合法的 401 / 403 / 503(它们是约定错误、不在 ⑧ 业务码里也正当);只有裸 500 / panic 这类未约定兜底混进契约才挑。别说"不列 500 / 5xx"——约定的 503 可以列,不可列的是未约定的未预期 500。
  • 返回结构统一:全接口遵循同一套 envelope / 错误外壳,不是一个接口一个样。
  • 分页一致性(列表端点,硬动作):多个列表 / 分页端点的分页约定是否跨端点一致、且与通用约定对齐——page / page_size 参数名 / 默认 / 上限一致,分页元数据都归 metatotal / page / page_size)不散顶层。一个端点一套分页、另一个另一套 = 不一致(major,下游 api 用例夹具 / 断言难统一)。
  • 限流 / 熔断(视项目提示,soft):工程若有限流 / 熔断机制,相关端点 / 设计有没有交代触发后行为(429 / 降级 / 重试语义)——项目没有就不强求(别凭空要求),有却只字不提 = 缺口。
  • 写端点幂等 / 并发交代了没(硬动作):每个 POST / PUT / PATCH / DELETE 端点是否交代了 ① 幂等性(幂等键 / 唯一约束 / 去重策略——重复提交 / 客户端重试会不会造重复写)② 并发写策略(乐观锁 / version 字段 / 行锁 / 409 冲突返回——两个请求同时改同一条会怎样)。两者任一(读端点 GET 不要求)= major;含糊带过("做好幂等""注意并发"而无具体机制)按未消解处理。
  • 端点索引 ↔ ⑦影响范围对账(防漏测,硬动作):把 api-contract.md 端点索引的端点集合,与 design.md ⑦ 影响范围登记的"动了哪些接口"逐个对照——契约里有、⑦ 没登记的端点 = 漏登记(整条会漏测 / 漏回归);反过来 ⑦ 提到改了某接口、契约里却没有该端点明细 = 契约缺口。任一方向对不上 = major

Read the full file on GitHub · 101 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 9d ago First seen · 101 lines · 249 tokens per session scan A 46df77602156

Subscribe to this mod's changes

hc-tech-design-reviewer is an agent published in the GitHub repository harness-base/harness-control (22 stars, last pushed 1mo ago), licensed MIT. It adds 249 tokens to every session and 6,949 once invoked, about $0.0012 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other agents, from other repositories

fastapi-reviewer

Reviews FastAPI applications for async correctness, dependency injection, Pydantic schemas, security, OpenAPI quality, testing, and production readiness.

affaan-m/ECC · 33 tokens

backend-reviewer

Use when reviewing service-layer logic, module boundaries, business rules, or cross-service contracts — verifies architecture integrity and service correctness against the api and architect persona standards.

jeremylongshore/tons-of-skills-marketplace · 36 tokens

checklist-generator

PRFlow review-engine agent; use to enumerate every verifiable claim in a code diff as a JSON checklist.

The01Geek/prflow · 26 tokens

enterprise-saas-reviewer

B2B / enterprise-SaaS pre-implementation reviewer. Outputs threat model TM-{slug}.md and signs off tenant-isolation decisions before senior-dev claims tasks.

avelikiy/great_cto · 40 tokens

integrations-engineer

Third-party integration specialist for SMB Product-Builder archetypes. Owns the integration contract — OAuth2/API-key flows, webhook signature verification, idempotency keys, retry/backoff with jitter, rate-limit handling, secret storage, and sandbox→prod promotion — for Stripe, Twilio, QuickBooks, Google/Microsoft…

avelikiy/great_cto · 106 tokens

behavioral-auditor

GoF Behavioral patterns auditor. Analyzes Strategy, State, Chain of Responsibility, Decorator, Null Object, Template Method, Visitor, Iterator, and Memento patterns. Called by acc:pattern-auditor coordinator.

dykyi-roman/awesome-claude-code · 51 tokens