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 BackToCimaCoppi/Praxis --skill adversarial-reviewgit clone --depth 1 https://github.com/BackToCimaCoppi/PraxisWrote 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/backtocimacoppi/praxis/adversarial-review)<a href="https://agentmods.dev/skills/backtocimacoppi/praxis/adversarial-review"><img src="https://agentmods.dev/badge/skills/backtocimacoppi/praxis/adversarial-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/backtocimacoppi/praxis/adversarial-review"><img src="https://agentmods.dev/badge/skills/backtocimacoppi/praxis/adversarial-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 | $0.04318 |
| Opus 5.5 | $0.00044 | $0.01727 |
| Sonnet 5.5 | $0.00022 | $0.00864 |
| Haiku 4.5 | $0.00011 | $0.00432 |
Grade A, and why
adversarial-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 1mo 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 — 226 lines — stays where its author put it; the contents beside it link to each section on GitHub.
对抗评审
0. 硬边界
- 同一
review_id + object_hash + truth_baseline_hash只开放一次,不自动开 R2/R3。对象或正式真值根本变化才可登记新身份。 - 每个实际选择的评审者都在授权项目根内独立只读调查,使用同一冻结对象与基线,不读其他人的结果。
- 当前主线程合并、核证和逐条裁决;评审者不代替用户选择新业务语义,不另设裁判模型。
- 评审报告是过程证据,正式 L1–L7 和授权裁决才是规格真值。
- 所有模型与运行参数禁用 Fast;不使用 Auto 或不明模型作为实际身份。
1. 触发时的桥接与模型选择
每次触发都从本 skill 目录运行 python3 scripts/discover_bridges.py。脚本仅报告三个 CLI 是否安装、是否登录、当前账户可见的模型 ID/名称和失败原因;不得把本机上一次的模型列表当作下一个用户的可用列表。读取脚本 JSON 后:
- 对不可用桥接明确说明未安装、未登录或查询失败;其他可用桥接照常提供。
- 按桥接展示模型标识和名称。可先筛出名称或描述中显示高推理能力的几个候选,提供完整清单的查询路径;候选筛选不能伪装成账户限制,也不能写死具体型号。
- 请用户选择至少一个“桥接 + 模型标识”;Codex 另选该模型实际支持的推理档。允许同一桥接选多个模型。若只有一个可用或用户只选一个,照样执行独立评审并给单模型发现报告;不得假造“对抗双方”、共识或排名。
- 用户选择后再次核对所选 ID 仍在本轮实时清单里。缺失、认证过期或模型已下架时告知用户并重选,禁止自行替换。
- 未收到选择时不启动评审。仅在用户已经提供本轮精确桥接/模型且核验通过时,直接继续。
发现脚本不输出账号、令牌或认证缓存。模型清单只证明可见,实际调用失败时按 §8 处理。
2. 总体流程
实时探测三个桥接 → 用户选择一个或多个桥接/模型 → 冻结评审身份与证据包
→ 所有已选模型同时独立只读评审 → 当前主线程合并与逐条裁决
→ 需要业务选择时请授权角色逐项裁决 → 多模型才点评排名
→ 按采纳项整改 → closed-remediation-review 固定范围验收 → PASS
3. 证据包
复杂场景同时读取 references/证据包制作规范.md;本节是强制最小集。
3.1 冻结内容
- review_id / object_hash / truth_baseline_hash。
- 评审对象:短小、临时或评审通道不可直接访问时内嵌全文;同一工作区内的大对象给绝对路径、哈希和精确范围,禁止摘要替代。
- 正式真值基线:适用的 L1–L7、冻结规格、工程红线和任务边界。
- 已决策·不得重开区:每项必须引用正式规格或 DEC-x;没有证据不得放入。
- 本次口径卡:设计、测试用例、章程或代码评审口径。
- 授权项目根
REVIEW_ROOT:默认解析为被评审对象所在 Git 工作树根;非 Git 项目才用用户明确授权的项目根。不得为了少给上下文而降成对象目录。 - 起始导航,以及明确的禁止读取项、禁止动作、输出格式和严重度预算。
钉对象,不钉视野;读得宽,评得窄:全部已选评审者必须评同一对象、同一真值基线和同一口径,但必须各自在 REVIEW_ROOT 内做独立勘察。证据包是冻结核心与调查种子,不是穷举白名单;各方实际发现路径可以不同。补读结论要给精确位置,调查得到的背景只能用于判断冻结对象,不能重开已决策项或改评整个项目。
证据包本身也要受审:评审者必须核验起始材料是否遗漏冲突真值、关键生产者/消费者或已经失效;发现缺口时用事实报告,不能默认制包方天然完整。
3.2 路径与临时目录
- 报告路径必须先解析为绝对路径,且位于用户授权的任务目录。
- 临时目录每轮用独立 mktemp -d,不得使用全局固定文件名。
- 记录评审前工作树快照;全部已选评审者完成后核验目标文件和工作树没有被它们修改。
- 原始 各实际评审者输出保留到报告验收完成,随后可清理临时目录。
4. 全部评审者共用指令
证据包末尾必须加入以下语义,各评审者仅替换角色名:
你是独立评审者,不是设计者、裁判或施工者。
开工顺序(强制):
1. 读完冻结对象、真值基线和起始导航;
2. 在授权项目根内亲自做一次独立只读勘察,不得只凭证据包清单作答;
3. 按任务相关性检索同一概念、业务标识、状态与错误码,核对正式规格、生产者、消费者、测试、配置/迁移/运行手册;
4. 把证据包完整性也当作待核验声明;继续检索已不再改变本次口径内结论后再收口。
What ships with it
7 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 1mo ago Changed · -115 lines · -104 tokens per session 0581599a9dbb
- 1mo ago First seen · 341 lines · 215 tokens per session scan A a1327fc6463d
adversarial-review is a skill published in the GitHub repository BackToCimaCoppi/Praxis (6 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 111 tokens to every session and 4,318 once invoked, about $0.0004 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 2026-08-31.
Other skills, from other repositories
ork-review-pr
Review a pull request or branch for correctness, regressions, security, operational risk, and missing evidence. Use when a request asks to review a PR, review a diff, find real bugs, assess merge risk, or provide evidence-backed review findings. Do not use for implementation or style-only cleanup.
ork-assess
Assess a code change, design, architecture, workflow, or competing options against explicit criteria and evidence. Use when a request asks to assess, rate, compare, identify trade-offs, evaluate readiness, or decide whether an approach is good enough. Do not use for a full pull-request review or to implement a chosen…
ork-verify
Verify that existing work is ready to merge, release, or hand off using an explicit evidence contract. Use when a request asks to verify, validate, prove, check readiness, run the relevant tests, or distinguish a claimed result from an observed one. Do not use to write missing tests or fix failures.
chief-of-staff
The CEO's right hand — maintains validated project context, prepares task contracts, logs cross-department decisions, and runs the weekly review. Use when the user says "what's in flight", "resume the project", "log this decision", "weekly review", "turn this idea into a brief", or when multi-department work needs…
appsec-review
Produces a bounded defensive application security review: threat model, security-sensitive diff, dependencies, and secrets hygiene with source-grounded findings. Use when the user says "security-review this release", "check this integration before shipping", "review the auth change", or "audit our agent's data access".
close-management
Runs the month-end close: generates a day-by-day runbook from D-2 to D+5 as CLOSE- .md with an owner matrix, dependency-ordered tasks (subledgers → accruals → reconciliations → flux → lock), numeric flux thresholds, a preparer/reviewer sign-off log, and close metrics such as days-to-close and reopened entries. Use…