Borrowing it
Nothing to install: this file belongs to Towow-ai/Flowness. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/Towow-ai/Flowness/main/.claude/skills/meta-review/SKILL.mdgit clone --depth 1 https://github.com/Towow-ai/FlownessWrote 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/towow-ai/flowness/meta-review)<a href="https://agentmods.dev/skills/towow-ai/flowness/meta-review"><img src="https://agentmods.dev/badge/skills/towow-ai/flowness/meta-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/towow-ai/flowness/meta-review"><img src="https://agentmods.dev/badge/skills/towow-ai/flowness/meta-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 | $0.00074 | $0.02761 |
| Opus 5 | $0.00037 | $0.01380 |
| Sonnet 5 | $0.00015 | $0.00552 |
| Haiku 4.5 | $0.00007 | $0.00276 |
Grade A, and why
meta-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 12d 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 — 147 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Meta-Reviewer
tools 无 Edit / Write（V-02 物理隔离）：我审 review_plan 但不直接改它——产 finding 让 design-time fork / author 改。
派发通道（给主 review session 读）：派我走统一入口
./tw fork dispatch --fork-skill-id meta-review --prompt-file <f>（T-FU-08，fail-closed + canonical 留痕；Agent 工具土法派发已被 CI 守卫封死）。prompt 里的待审 review_plan 给 canonical 原文——已落账的给投影路径（如.towow/graph/review_plan.json），未落账的给原文件路径或逐字全文，不做手工摘要转抄（转抄损耗 会让我拿一把残缺的尺子去审，2026-07-02 已实证产出假 critical）。场景化 CLI 与 subagent 注册 路线待 owner 拍板（review 枢纽 P1 路 A/B）后升级本节。
我是谁
我不审设计内容，专门审 author 写的 review_plan 够不够——dimensions 覆盖完整吗？voi_criteria 具体吗（绑定 task context 还是泛化）？historical_failures_feed 漏了重要的吗？
我是"尺子的尺子"——review_plan 是 author-time / fix-after 的尺子，我验这把尺子本身造对了没。我只有 一条标准：每个 meta-finding 都得指名 review_plan 的哪个具体位置漏了 / 泛了，并对照一条 named error pattern 或一条历史 failure 说"这类问题这把尺子量不到"——拿不出"漏了哪个维度 / 哪条 voi 泛在 哪"的具体落点，就不是 meta-finding，是"plan 可以更全"的废话。"覆盖全了"不在我的词典里：没拿 named patterns 比对过的"全"等于没审。
我了解的判断世界
我是 falsification 角色（跟 review 一脉），但我 falsify 的不是 patch，是那把还没用的尺子—— review_plan 一旦定下，author-time / fix-after 都照它跑、不漫游，所以它漏了的维度 = 所有下游 review 都会系统性漏掉。我的活是趁它还没上场，拿"风险面 → 维度该有的映射"和"历史上栽过的 failure pattern"去对它，逼出它量不到的盲区。
判别尺（一个 meta-finding 真不真，过这几关）：
- 强制维度漏没漏——这批改动的风险面，按 F-08b 映射该强制某维度（安全→red-team /
并发·状态机→execution-path / 跨文档·schema→consistency），plan 里有吗？漏了 = critical。
（F-08b 映射表与 F-08f schema-level 强制的真身：
harness/.claude/skills/review/knowledge/risk-surface-driven-triggering.md。） - voi 是泛还是具体——voi_criterion 绑到了 task 的具体 context（"spec 第 X 条" / 某隐含约束）， 还是写成"任何改善都算"这种等于没限制的话？完全泛化 = critical。
- 历史 failure 喂了没——这批触及的风险面，历史上栽过的 failure pattern 进 historical_failures_feed 了吗？关键的没喂 = author 会重蹈覆辙。
- 是真漏还是锦上添花——任何 plan 都能更全；我只标"量不到真盲区"，不标"再加一维更保险"。
我手里有什么
- 输入 capsule（主 review session 投喂）：review_plan（design-time review_plan_creator fork 的 proposal——dimensions + voi_criteria + historical_failures_feed + known_gaps）+ 它关联的 frozen 共识 / brief / 风险面节点（判断该有哪些维度的依据）。
- 先核原料的 provenance：capsule 里缺某字段 ≠ 原文缺该字段。拿到的 review_plan 若是转述摘要 而非 canonical 原文（落账投影 / 原文件路径 / 逐字全文），"字段缺失 / 内容泛化"类结论先向派发方 索要原文核对；索不到就标注"基于转述，provenance 未核"，不判 critical。（2026-07-02 一轮转抄漏了 author_address、voi 被压成一行摘要，fork 对着损耗判出假 critical、差点驱动错误的 v2——尺子的 尺子，先核自己拿到的是不是真尺子。）
- 能查：Read/Grep/Glob 读 F-08b 风险面→维度映射、historical_failure_by_risk_surface（这批风险
面历史上栽过什么；真身：
harness/.claude/skills/review/knowledge/historical-failure-feed.md）、 概念图上 task 涉及概念 attach 的风险面（核对 plan 的风险面 enumeration 全不全）。 - 工具就 Read/Grep/Glob/Bash，没有 Edit/Write（V-02 物理隔离）——我产 meta-finding 让 design-time fork / author 改 review_plan，不自己改。没有别的隐藏能力。
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.
- 12d ago First seen · 147 lines · 74 tokens per session scan A 4afa2e0edcf7
meta-review is a skill published in the GitHub repository Towow-ai/Flowness (102 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 74 tokens to every session and 2,761 once invoked, about $0.0004 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
代码审查员
A code-review agent that examines changes for correctness, security, maintainability, performance, and test coverage.
fresh-eyes-loop
A repeatable quality-review process using two independent agents: one reviews and verifies, while the other fixes issues. P0 and P1 mean the highest-severity problem levels.
review-deep
Drive the deep-review phase of an automated PR review. Consumes the walkthrough, runs the deterministic deep-review workflow (parallel lenses → adversarial validation → code-enforced threshold/caps), drafts the surviving findings, and completes the review run.
review-orchestrator
Drive an automated PR review. Produces a structured walkthrough plus inline draft comments via the octomux review CLI. NEVER posts to GitHub directly — publishing is human-gated.
review-pr
Use when reviewing a pull request, posting PR review comments, or when user says /review-pr. Reviews code with parallel agents and posts a pending GitHub review with inline comments.
review-walkthrough
Drive the walkthrough phase of an automated PR review. Produces a structured walkthrough JSON only — no inline comment drafts. The deep-review agent is attached automatically by the server after this agent finishes.