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 AbyssCN/oh-my-dag --skill omd-reviewgit clone --depth 1 https://github.com/AbyssCN/oh-my-dagWrote 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/abysscn/oh-my-dag/omd-review)<a href="https://agentmods.dev/skills/abysscn/oh-my-dag/omd-review"><img src="https://agentmods.dev/badge/skills/abysscn/oh-my-dag/omd-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/abysscn/oh-my-dag/omd-review"><img src="https://agentmods.dev/badge/skills/abysscn/oh-my-dag/omd-review.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00101 | $0.01517 |
| Opus 5 | $0.00051 | $0.00758 |
| Sonnet 5 | $0.00020 | $0.00303 |
| Haiku 4.5 | $0.00010 | $0.00152 |
Grade A, and why
omd-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 11d 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 — 58 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/omd-review — 对抗式改动审查
经 omd MCP dag_review(可能带 mcp__omd__ 前缀;未加载先 ToolSearch "dag_review")对一批改动做对抗审查:默认多维并行 find(召回)+ 确定性跨模型 verify(证伪);deep 档单 agent 读全仓 + 实测(实测最优,见档位选择)。
用法(异步三段式)
-
gate:按 blast radius 选,不按行数(真源harness/docs/GATES.md,与引擎scripts/dag-review.ts逐档一致)。不给 = G2。gate 什么改动 实际跑什么 G0文档 / 机械(注释、配置、纯改名、格式) ⚠ 免审短路——不取 diff、不发模型、直接 exit 0。它不是"便宜的浅扫",是"不审";要便宜的召回选 G1 G1骨架(新模块脚手架、签名、接线) contract+boundary两维G2常规逻辑(普通业务码、带行为测试的重构) correctness+security+boundary三维G3敏感:schema · 认证 · 安全边界 · 不可逆 G2 三维 + contract+spec轴强制(对照 SDD,--no-spec在 G3 直接报错)+ owner 终审轮数硬上限:G1/G2 各 1 轮,G3 = 1 轮 + 至多 1 个修复 cycle。
-
scope:收窄审查范围(路径/模块);不给 = 当前 branch diff 全量。 -
deep:深审档(true)。一个 pi agent 读全仓 + 实测(库/runtime API 主张必跑bun -e复现,治外部 API 幻觉)→ 确定性跨模型 verify。比默认贵,但精度更高、自然去重。
dag_review(gate?, scope?, deep?)→runId;dag_status(runId)轮询(跑几分钟,别重复发起);dag_result(runId)→ 报告落盘路径,自己Read再转述(按严重度先讲 top 项)。
档位选择:默认 vs 深审(deep)
- 默认(多维并行 find + 确定性 verify):廉价召回档,diff-可见的 bug 够抓,适合日常/大 diff。
- 深审
deep:true(单 agent):实测最干净(3 方召回测:满召回 + 0 假阳性 + 推理最好 + 自然去重)。用于敏感接缝(会计/authz/迁移/状态机)、需读 diff 外真身、或默认档出噪音时。贵在一个 agent 全仓读+跑,别对大批量机械 diff 滥用。 - 模型(review 自成体系,不碰引擎
verifier角色):find =review角色(OMD_REVIEW_MODEL);verify =OMD_REVIEW_VERIFY_MODEL覆盖,否则回落 find 模型(跨模型 verify 靠显式配OMD_REVIEW_VERIFY_MODEL)。无需任何配置也能跑,fallback 优雅降级、不假设 key。
误报裁决程序(承终裁手册:finding ≠ ground truth)
车队产出是候选,你终裁。每条 finding:
- 定位代码事实——必须看 diff 视野外(审查器只看 diff,这是误报主源)。
- oracle 证伪:typecheck / 测试 / 活库实测 / 既有测试钉。
- 证伪成立 → 驳,记驳的依据;证伪不了 → 修;拿不准 →
?升级 owner。
已知误报模式(直接警惕):
- 「X 未导出/未定义」而 X 是 diff 外既有代码(ugrep 确认真身,build 绿即证伪)。
- 「缺权限守卫」而守卫以 JOIN/EXISTS 已在查询里(空集测试即证伪)。
- 用「可能/如果内部没校验」推测行为——去读那个函数,不接受推测性 P0。
gate 分档:插在 ROI 高处,别铺满(承终裁手册)
审核 ROI = P(缺陷) × 逃逸代价 ÷ 成本。
- P(缺陷)高 = 机械铺量 / 敏感接缝(会计写·状态机·迁移)/ 大 diff。
- 逃逸代价高 = 账面污染 / 绕审计——错了是灾难,再贵也审(上 G3)。
- ROI 决定的是审不审、审多深;分档轴仍是 blast radius——骨架接线 + typecheck/test 全绿 → G1 够,别 G3。 但别拿 G0 当"小改的便宜档":G0 = 不审,只对文档/机械改动成立。
- 0 P0/P1 ≠ 白审,可能是"插错位置"信号:过程审(写码前判定)> 结果审(post-code gate)。
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.
- 11d ago First seen · 58 lines · 101 tokens per session scan A f51ba98022f9
omd-review is a skill published in the GitHub repository AbyssCN/oh-my-dag (39 stars, last pushed 2d ago), licensed MIT. It adds 101 tokens to every session and 1,517 once invoked, about $0.0005 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.
Other skills, from other repositories
control-center-review
Review one captured Control Center build before runtime activation.
review-pr
Review GitHub pull requests. Use when user says "review PR", "review pull request", or "/review-pr".
codex
Delegate coding work to the OpenAI Codex CLI for repository changes, reviews, and focused fixes.
spawn-reviewers
Spawn and collect the reviewer fleet at stage20spawnreviewers. Consumes spawn.json.spec (the authoritative spawn spec from derive-spawn-spec / derive-static-spec), resolves CODEINTELALLOWED, builds per-agent prompts from the per-agent template + role suffixes (Bug Hunter A/B, Unified Auditor, Domain Critics, Impact…
singleton-dispatch
PLN-725 single-agent dispatch for stage11extractsignals and stage15coveragecritic. After each prepare stage writes its manifest, this skill reads the status field (cachehit / skipped / needsagent) and, only on needsagent, spawns one synchronous singleton Task that writes the by-convention pln725.json output the…
verify-findings
Dispatch and collect the finding-verifier fleet at stage23verifyfindings (PLN-722). Reads verifymanifest.json (written by stage22bverifyprepare), spawns one falsify-oriented verifier Task per toverify[] entry with mode-specific Task scheduling (GitHub mode dispatches verifiers synchronously; local mode uses parallel…