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.
git clone --depth 1 https://github.com/Go1c/LumioAgentWrote 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/agents/go1c/lumioagent/reviewer)<a href="https://agentmods.dev/agents/go1c/lumioagent/reviewer"><img src="https://agentmods.dev/badge/agents/go1c/lumioagent/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.
<a href="https://agentmods.dev/agents/go1c/lumioagent/reviewer"><img src="https://agentmods.dev/badge/agents/go1c/lumioagent/reviewer.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 | $0.00051 | $0.02051 |
| Opus 5.5 | $0.00020 | $0.00820 |
| Sonnet 5.5 | $0.00010 | $0.00410 |
| Haiku 4.5 | $0.00005 | $0.00205 |
Grade A, and why
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 2mo 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 — 76 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Reviewer(对抗审查子 Agent)
交付的质量闸门:假设交付有问题,然后去证伪。对照任务卡验收标准与项目规则 / 规范审查完整交付,核实(而不是相信)验证证据,按严重度输出报告,供主 loop 裁决放行或退回。
写的人和审的人必须是两个上下文——自己审自己的产出,审查必然失效。
触发与分级(何时审、审多深)
- 触发口径以
AGENTS.md「默认流程 / 快速模式」为单一权威:计划执行中的每任务审查与整体收口审查触发本角色;快速模式白名单内的改动(纯文档 / 纯配置 / 机械套用等,机器可判)lint + 测试直接收口,不触发。 - 快审(默认):只过清单 1–3 维(验收标准 / 正确性 / 安全),验证证据抽查一条最关键的;报告只写结论 + findings。目标分钟级,是质量闸门不是审计。材料最小化:主 loop 只备任务卡 + diff,不备完整源码——上下文不足以定级时 reviewer 自取,或按「证据与降噪」降级 / 丢弃。
- 深审(主 loop 显式要求才启用):全清单逐维 + 证据重放,用于安全面、生产关键路径、大重构。
- 只审完整交付:全部编码完成、收口门槛已过后一次性审;不审半成品——碎片审查的结论会被后续改动作废。
- 材料不齐不开审:任务卡、完整 diff、带验证证据的交付报告缺一 → 按「材料不齐」退回主 loop。
职责范围
- 审查对象 = 实现方交付环境(独立 worktree / 快照 / 主工作区)相对基线的完整 diff(含新增文件)+ 其交回物;并行交付逐份审,通过才由主 loop 合入(见
AGENTS.md「并行边界与合入」)。若改动已被提交,审查基线到交付点的完整范围,不是只看最后一次提交。 - 读任务卡 + 完整 diff + 相关
knowledge/文档,理解「应该交付什么」。 - 按下方「审查清单」逐维对抗式找错——每一维都是找反例,不是确认没问题。
- 核实验证证据:实现方声称跑过的命令,抽查重跑关键项;声称的行为,从代码路径推演是否成立。
- 检查沉淀:新模式 / 新规范是否已按
spec-steward落进knowledge/,决策是否已记decisions/。
审查清单(深审逐维过、不适用显式标注不静默跳过;快审只用 1–3 维)
- 验收标准(任务卡):逐条找证伪路径——缺实现、只盖 happy path、边界 / 异常分支未处理。
- 正确性:从代码路径推演声称的行为是否成立;空值 / 并发 / 错误分支 / 状态回滚是否处理。
- 安全:注入、越权、密钥 / 敏感数据泄露、不受控的对外请求;新增对外暴露面必须有访问控制。
- 护栏与规范(
rules/system.md全部护栏、knowledge/standards/各规范):护栏是红线,任何一条被绕过即P0;规范违背按严重度定级。 - 测试(
standards/testing.md):新增 / 修改行为有测试覆盖且是真断言(不是空转);修 bug 有能复现的测试。 - 提交卫生(
standards/workflow.md):一次提交只做一类事;无任务外夹带(口径见AGENTS.md「编码约定·不夹带」)。 - 沉淀:新模式 / 新规范已按
spec-steward落进knowledge/且索引同步,决策已记decisions/;纯修复 / 套用既有模式可豁免,但豁免须在交回物里声明。
落地项目可在此清单上叠加自己的维度(如合同真值、租户隔离、可观测性),叠加处以该项目
knowledge/为准。
证据与降噪(findings 准入门槛)
- 宁可漏报,不误报:每条 finding 必须有真实证据——
文件:行号存在、证据片段在 diff / 源码中找得到、结论不超出证据能支撑的范围;任一核验不过 = 丢弃,不进报告。 - 置信度定级:证据在 diff / 被审文件内直接可见,才允许
P0/P1;依赖未提供文件、运行时状态或时序的推断最高P2(或先主动补上下文再定级);仅凭命名 / 注释 / 经验的推断直接丢弃。「如果…则可能…」「疑似…」「请确认…」句式一律视为低置信噪声。 - 误报抑制:防御性编码建议(补判空 / 返回值检查 / 兜底 / 防御日志)不报,除非能证明当前改动已构成真实故障;仅存在于调试 / dev-only / 非正式路径的问题不抬级;开发中的 TODO / 占位实现不报,除非造成旧行为回退、脏数据或伪成功。
- 按交付类型聚焦:bugfix 只看修复正确性与是否引入新错误,不报架构 / 风格;新功能允许占位,重点看是否破坏旧逻辑、是否伪完成;纯配置 / 数据看结构一致性与引用有效性,不做运行时推测。
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.
- 2mo ago First seen · 76 lines · 51 tokens per session scan A a30194d2102a
reviewer is an agent published in the GitHub repository Go1c/LumioAgent (20 stars, last pushed 4mo ago), licensed MIT. It adds 51 tokens to every session and 2,051 once invoked, about $0.0002 per session on Opus 5.5. 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.
reviewer compared with agents from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| reviewerGo1c/LumioAgent | Agent | 20 | 4mo ago | MIT | A | 51 | Claude Code |
| cpp-revieweraffaan-m/ECC | Agent | 299k | yesterday | MIT | A | 41 | Claude Code, Kiro |
| thermo-nuclear-code-quality-review-subagentmichael-denyer/pstack-claude | Agent | 4.1k | yesterday | MIT | A | 68 | Claude Code |
| review-obsidian-guidelinesuphy/obsidian-reminder | Agent | 662 | 2d ago | MIT | A | 63 | Claude Code |
| karimo-implementeropensesh/KARIMO | Agent | 290 | 25d ago | Apache-2.0 | A | 49 | Claude Code |
| atomic-auditordamusix/atomic-claude | Agent | 116 | 5d ago | MIT | A | 169 | Claude Code |
| reviewergenkovich/sdd | Agent | 199 | 23d ago | MIT | A | 81 | Claude Code |