Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add baixinghao/intent-gate --skill red-blue-reviewgit clone --depth 1 https://github.com/baixinghao/intent-gateWrote 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.
[](https://agentmods.dev/skills/baixinghao/intent-gate/red-blue-review)<a href="https://agentmods.dev/skills/baixinghao/intent-gate/red-blue-review"><img src="https://agentmods.dev/badge/skills/baixinghao/intent-gate/red-blue-review/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.
<a href="https://agentmods.dev/skills/baixinghao/intent-gate/red-blue-review"><img src="https://agentmods.dev/badge/skills/baixinghao/intent-gate/red-blue-review.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00111 | $0.03637 |
| Opus 5 | $0.00056 | $0.01818 |
| Sonnet 5 | $0.00022 | $0.00727 |
| Haiku 4.5 | $0.00011 | $0.00364 |
Grade A, and why
red-blue-review 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.
How it starts
The opening of the file, as written. The whole thing — 197 lines — stays where its author put it; the contents beside it link to each section on GitHub.
红蓝对抗评审(蓝军 playbook)
对需求分析产物做对抗性审查:红军(需求分析方)产出 summary.md/DDL/对齐流水, 蓝军(你)以对立视角找茬,产出结构化 findings,驱动红军整改闭环。
可选项:本评审是插件自带的可选质量门,不是必经流程—— 人类点名启动,或 complex 需求交付时由你建议、人类同意后启动。 没启动意图时不要把评审内容塞进正常分析流程。
适用门槛:仅 complexity=complex 的需求值得评审;simple/medium 直接 PASS 放行并说明。
1. 独立性与信息节食(🔴 对抗有效性的根基)
- 🔴 蓝军必须在独立 session 执行(新开对话,或派遣与红军无共享上下文的子代理), 禁止与红军同 session——同 session 看到红军推理过程即继承其假设,对抗退化成自查。
- 🔴 信息不对称:蓝军只读"产物 + 过程文件 + rubric 文件", 禁止向人类索要红军的思维链、对话记录或"当时是怎么想的"。
- 🔴 蓝军不修改红军的任何产物,唯一输出是
_review/review-findings.md。 - 🟡 同厂商异 session 即可满足独立性;异厂商模型为加分项,非准入门槛。
2. 输入(全部来自 .harness/requests/{需求名}/)
| 文件 | 用途 |
|---|---|
summary.md |
被审主产物(报告) |
_review/alignment-log.md |
意图对齐流水(R1 注入保真复核的唯一依据) |
_review/review-request.md |
红军开单(产物清单 + 自评 + 建议评审重点) |
_review/review-findings.md(前轮,如存在) |
🔴 并集复核:逐条判定"仍有效→并入本轮 / 已失效→注明原因",禁止覆盖丢弃前轮成果 |
sql/*.sql(项目根目录) |
被审 DDL |
原始 PRD(doc/request/ 或用户提供) |
蓝军独立扫雷用(R3 歧义漏判检查) |
rubric 参照:项目红线文件(有则读)、术语 wiki(有则读)、
MCP prompt doc_analysis_playbook 的产出规范(§3.5-3.8 图表规范)。
🟡 进场前置检查:发现红军产物目录缺
_review/alignment-log.md→ 不必开审,直接结论 FAIL-可整改(过程证据缺失:没有对齐流水, R1 注入保真无从复核),要求红军补齐流水后重新开单。
2.5 执行流程(lint 先行 + 两遍制)
第 0 步:lint 先行(🔴 不可跳过)
调 MCP 工具 lint_summary(summary_path),机械检查(状态机终态/死状态/锚点引用/
规则引用/表读写矩阵)由工具完成。蓝军必须将其作为 findings 基线:
复核真实性后并入 findings(标注来源 [lint]),禁止忽略或重复劳动。
lint 覆盖了 R4/R5/R9 的机械半边,蓝军把精力留给语义判断。
第 1 遍:单产物检查(R1-R8)
逐表逐节过 R1-R8,产出单产物 findings。
第 2 遍:跨图 + 独立扫雷(R9 + R3)——lint 出矩阵,蓝军填判断
🔴 禁止手工建矩阵。lint 已在 _review/lint-report.md 生成三张矩阵骨架
(① 转移清单含事件/动作、② 表读写矩阵、③ 引用核对清单),蓝军的工作是
填判断列:矩阵中的空格、"待核"、❌ 格就是 findings 候选。
🔴 多出边状态的语义跟进(禁止踢回人类):lint L3 或矩阵①标出的"同一状态多条出边", 蓝军必须逐边对照 alignment-log 注入意图与决策表规则,自行判定触发条件是否互斥、 是否与注入意图矛盾(典型案件:注入意图说"超时自动刷新",状态机却画了"超时 → 踢出回首页")。 确实无法判定的,标 MAJOR 并写明缺什么信息, 禁止原样写"需人工确认"——语义分析是蓝军的活,不是人类的。
🔴 前轮 findings 并集(限 CRITICAL/MAJOR):进场发现前轮 review-findings.md 时,
仅对前轮 CRITICAL/MAJOR 逐条复核——仍有效的并入本轮(标注"前轮遗留"),
已失效的注明失效原因;MINOR 免检(不复核、不并入,防止低价值项滚动膨胀)。禁止静默丢弃。
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.
- 9d ago First seen · 197 lines · 111 tokens per session scan A 85d0561d71ec
red-blue-review is a skill published in the GitHub repository baixinghao/intent-gate (2 stars, last pushed 25d ago), licensed MIT. It adds 111 tokens to every session and 3,637 once invoked, about $0.0006 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 2026-08-31.
Other skills, from other repositories
refine-tests
Reviews existing test code for conformance to the test-designing-guide and test-writing-guide, then applies the refinements. Use this skill when the user wants to review or refine existing test code so it follows the project's test design and writing conventions. Typically invoked as /refine-tests .
work-review
Perform exhaustive code reviews using multi-agent analysis. Reviews PRs, branches, or current changes. Writes review.findings.json to the active session. Triggers on "review", "code review", "check PR".
review
Reviews code changes for concrete bugs with P0-P2 prioritized findings. Uses parallel architectural and implementation reviewers, then validates findings to remove false positives. Use when reviewing a caller-supplied diff, branch, staged changes, file set, or pull request, and use fix mode with caller-supplied…
plan-eng-review
Engineering plan review. Reviews a plan document for architecture, code quality, test coverage, and performance before implementation begins. Produces a coverage diagram, failure modes map, and parallelization strategy. Use when asked to "review this plan", "eng review", or "architecture review".
summary
Creates a decision-focused Moss change summary for reviewers who care about the problem, outcomes, trade-offs, and platform implications more than code structure. Use for pull-request summaries, release notes, branch or diff overviews, implementation handoffs, and explanations of what changed and why it matters.…
review
Pre-merge code review — the single canonical review of a change before it lands. Covers BOTH diff safety/structure (SQL safety, LLM trust-boundary violations, conditional side effects) AND engineering quality (architecture fit, edge cases, test coverage, performance). Use when asked to "review this PR", "code review"…