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 SeaOf0/dsh-redteam-model --skill independent-reviewgit clone --depth 1 https://github.com/SeaOf0/dsh-redteam-modelWrote 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/seaof0/dsh-redteam-model/independent-review)<a href="https://agentmods.dev/skills/seaof0/dsh-redteam-model/independent-review"><img src="https://agentmods.dev/badge/skills/seaof0/dsh-redteam-model/independent-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/seaof0/dsh-redteam-model/independent-review"><img src="https://agentmods.dev/badge/skills/seaof0/dsh-redteam-model/independent-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 | $0.00138 | $0.01546 |
| Opus 5 | $0.00069 | $0.00773 |
| Sonnet 5 | $0.00028 | $0.00309 |
| Haiku 4.5 | $0.00014 | $0.00155 |
Grade A, and why
independent-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 — 77 lines — stays where its author put it; the contents beside it link to each section on GitHub.
独立复核员模板(independent-review)
位置:
plugins/dsh-redteam-model/shared/skills/,九预设共同加载(八专业模式 + redteam light 通用模式)。
复核员的独立性规则(构造 prompt 时必须遵守)
- 只给原始材料,不给主链结论:派工 prompt 附上原始证据(命令回显/请求响应包/ 代码片段/反汇编/样本哈希与 trace),不告诉它主链怎么判的、定了几级。
- 要求独立重算:复核员必须用自己的方法重走关键路径(重算调用链/重放关键请求/ 重读反汇编段),禁止只做「看起来合理」的确认。
- 输出二选一,禁止骑墙:最终输出只能是
确认(附独立证据)或挑战(附反证与替代解释),不允许「基本同意但保留意见」。 - 证据引用规范:结论必须引用原始材料中的具体位置(file:line / 时间戳 / 请求包编号), 与各模式六字段报告的证据标准一致。
- 回传格式(紧凑回传协议):复核结论回传父会话只含——结论(确认/挑战)+ 关键证据 编号(evidence-index 编号/file:line)+ 产物哈希 + 与主链的差异清单(如有); 禁止回传证据全文/大段原文——父会话需要时按编号读盘核对。证据本体已在盘上, 回传全文只消耗上下文不增加信息。
light 语义(redteam light 通用模式派单时的复核裁剪)
light 通用模式(研究员总控)按「confirmed 级才复核」派单:partial/unknown 级浅做结论不派 复核员(带等级标注直接进总结);被路由任务的专业强门禁由对应专业模式的 gate 负责,本技能 不代行。八专业模式沿用全量复核语义不变。
复核员 prompt 模板(总控派工时套用)
你是独立复核员。以下是待复核 finding 的原始材料(未经加工):
<原始证据材料,不含任何主链结论>
任务:
1. 独立重走关键路径:<按模式给具体指令——pentest:重放关键请求对照响应;
code-audit:从入口重算调用链到 sink;binary-analysis:重读反汇编关键段;
attack-defense:重验利用链关键步骤;av-evasion:核对构建命令/哈希/判定日志>
2. 排除误报来源:环境差异 / 工具自身误报 / 我方测试残留。
3. 输出(二选一,禁止骑墙):
确认 —— 附你的独立证据;
挑战 —— 附反证与替代解释。
4. 若被指派 gate 校验(见下节),同时输出 gate-pass / gate-fail 与缺失字段清单。
阶段门禁校验(兼任门禁官,当前 attack-defense 使用,其他模式可选)
复核员在交叉复核之外,可被总控指派对阶段产物做 gate 校验:
- 输入:该阶段落盘产物(
artifacts/+evidence-index.md)与该阶段必需字段清单 (各 playbook「阶段契约」表)。 - 校验:逐字段核对存在性与证据挂钩(每个字段是否有证据引用/哈希/时间戳支撑, 不接受「口头声称已完成」);证据血缘完整性——派生证据的 parent 必须指向存在且可信的 父条目(root 除外),血缘断链或父条目不可信 → 该证据降级「疑似」并列入 gate-fail 缺项。
- 输出:
gate-pass(可进下一阶段)或gate-fail+ 缺失字段清单(打回该阶段)。 - 总控规则:报告只收带复核员 gate-pass 签名的 finding;阶段推进以 gate-pass 为准。
跨 harness 双签(独立性同上注)流程(关键 finding / 计分项)——用户触发制
前置(铁律):双签不主动执行。 DSH 独立复核通过的结论即为最终输出;跨 harness 双签由用户决定——各模式在报告结尾把它列为建议项(注明适用场景与触发方式),用户说复核才走本流程。
- 第一签:DSH 独立子代理按上述模板复核 → 确认(主动必做,与双签无关)。
- 第二签(用户批准后):
subagent_claude_code以同一套原始材料(同样不给结论)独立复核 → 确认。 - 两签一致 → 结论升级为「跨 harness 双签一致」;不一致 → 双方结论并列上报用户裁决,或降级为「疑似」。
- claude 不可用 → 按兜底链降级(codex → DSH 双子代理分别独立复核), 并在报告中注明复核方式。
- claude 的确认同样需要证据,不接受无证据背书(persona 通用规则)。
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 · 77 lines · 138 tokens per session scan A cbfcb9cb782c
independent-review is a skill published in the GitHub repository SeaOf0/dsh-redteam-model (398 stars, last pushed yesterday), licensed MIT. It adds 138 tokens to every session and 1,546 once invoked, about $0.0007 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 2026-09-03.
Other skills, from other repositories
novel-reviewer
A review and editing workflow for existing novel chapters. It checks story logic, timelines, character states, unresolved clues, world rules, and writing style.
workflow-manager
A writing workflow tracker for chapter-based projects. It records whether each chapter is being assembled, drafted, reviewed, confirmed, polished, or archived.
review-chapter
A six-part review of a manuscript chapter. It compares the chapter with the outline and reports quality, coverage, review status, and problems supported by evidence from the text.
workflow
A command set for starting and managing step-by-step automated workflows.
local-search
A local search component that finds web results using installed command-line tools and page parsers. It supports general web search and extra sources for areas such as Chinese pages, news, code questions, academic work, and reference material.
manage-taskboard
Manage work in the native DeepSeek Harness Taskboard with exact task ids and optimistic versions. Use when an Agent must inspect project work, claim an eligible todo, record progress or blockers, verify an implementation, submit it for human review, or release its own claim; also use when a human asks how to accept…