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 telagod/code-abyss --skill orchestrating-adversarial-reviewsgit clone --depth 1 https://github.com/telagod/code-abyssWrote 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/telagod/code-abyss/orchestrating-adversarial-reviews)<a href="https://agentmods.dev/skills/telagod/code-abyss/orchestrating-adversarial-reviews"><img src="https://agentmods.dev/badge/skills/telagod/code-abyss/orchestrating-adversarial-reviews.svg" alt="Measured on agentmods" 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 | $0.00143 | $0.01529 |
| Opus 5 | $0.00072 | $0.00764 |
| Sonnet 5 | $0.00029 | $0.00306 |
| Haiku 4.5 | $0.00014 | $0.00153 |
Grade C, and why
orchestrating-adversarial-reviews scanned grade C with 1 finding 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 7d 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.
Hidden instructionshighPrompt injection
Directives inside HTML comments, invisible characters or bidirectional overrides are read by the model and not by the person reviewing the file.
<!-- safety-scan: ignore RM_RF_ROOT,CURL_PIPE_SH,PROMPT_INJECTION 本 skill 把危险命令(| tail 吞退出码、docker rm 误删、agent 谎报)列为反模式教学，自身不执行 --> How it starts
The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.
对抗验证编排 · orchestrating-adversarial-reviews
单个 agent 会谎报"已修复 / 全覆盖 / 没问题"。结论的可信度不来自"谁说的"，来自"扛过几次推翻"。 本 skill 是编排骨架：fan-out 发现 → 三棱镜对抗验证 → 证明性 guard → 守卫式上线。 信级：运行时行为 / 证明测试 > 多 agent 多数裁决 > 单 agent 自报（永远
[unverified]）。
核心信条
- 不信单 agent 自报。 finder 会噪音误报，implementer 会谎称"已修复/全覆盖"。每个高价值结论必须被独立 agent 用不同视角尝试推翻。
- 可证伪 > 可声称。 修复必须配一个"退回漏洞代码就 FAIL、修好才 PASS"的 load-bearing 证明测试。没有证明测试的"已修复"等于没修。
- 失败方向要对。 守卫链里任何一步的退出码都不能被管道遮住；破坏性动作前置可逆检查。
何时使用
| 场景 | 用 | 理由 |
|---|---|---|
| 授权安全审计 / 加固闭环 | ✅ | 首个范例，见 workflow |
| 大面积代码审查（多维度、需高可信） | ✅ | dimensions → find → 对抗验证 |
| 研究综合 / 事实核查（结论要扛得住） | ✅ | 多源 fan-out + 证伪棱镜 |
| 大规模迁移 / 重构（site 发现 + 逐项验证） | ✅ | pipeline 逐项独立 + 证明测试 |
何时不使用
- ❌ 单文件、低风险、机械改动——直接做完跑测试，别套编排（参见
shipping-changes的"何时不使用"）。 - ❌ 用户没有 opt-in 多-agent 编排 / 没开 ultracode——Workflow 会 fan-out 几十个 agent 烧大量 token，必须显式授权。
- ❌ 只需要"找什么洞"的知识——那是
securing-systems/analyzing-security，本 skill 不重写知识，只编排。
编排骨架（三相）
Recon (fan-out) 每维一个 finder, 并行深读, schema 出结构化 findings
| pipeline 而非 barrier: 维度A的发现可在维度B还在找时就进验证
v
Verify (三棱镜) 每条 finding 派 N 个 verifier, 各执一镜, 默认怀疑
| 可利用性 / 正确性 / 证伪猎杀 —— 票数 >= 多数 才保留
v
Synthesize / Ship 合成定级报告; 若是修复任务 -> 证明测试 guard -> build-first 上线
对应 Workflow 工具的 pipeline(items, findStage, verifyStage)（默认无栅栏，墙钟最短）。仅当"下一阶段需全部上一阶段结果"（去重 / 早退 / 跨条比较）才用 parallel 栅栏。
四大护栏（实战血泪，按重要度）
- 三棱镜对抗验证（防 finder 噪音）——每条发现派视角各异的 verifier（可利用性 / 正确性 / 证伪猎杀），默认怀疑，多数票
confirmed才保留。 - 证明性测试 guard / load-bearing（防 implementer 谎报）——修复配真行为测试，且测试本身能"反向证伪"（退回漏洞版必 FAIL）。本 skill 的命门。
- build-first + 退出码 guard（防误删 / 半成品上线）——先构建验证新件再动旧件；
cmd | tail会吞掉cmd的退出码，判码用cmd > log 2>&1; rc=$?。 - 语境校准降噪——fan-out 前钉死威胁模型 / 评判语境，否则收一堆无效发现。
What ships with it
1 file 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.
- 7d ago First seen · 75 lines · 143 tokens per session scan C 83fe7dc394a9
orchestrating-adversarial-reviews is a skill published in the GitHub repository telagod/code-abyss (239 stars, last pushed 1mo ago), licensed MIT. It adds 143 tokens to every session and 1,529 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it C with 1 finding (hidden instructions). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
pr-test-analyzer
Reviews pull-request test coverage for critical gaps, edge cases, and brittle implementation-coupled tests. Prioritises behavioral coverage over line coverage and rates each finding 1-10 by criticality. Use when reviewing PRs that add or modify functionality and you want to assess whether the accompanying tests are…
review-work
Post-implementation review orchestrator. Launches 5 parallel background sub-agents: Oracle (goal/constraint verification), Oracle (code quality), Oracle (security), unspecified-high (hands-on QA execution), unspecified-high (context mining from GitHub/git/Slack/Notion). All must pass for review to pass. MUST USE…
verify-implementation
A workflow that runs a project’s verification skills to produce a report on coding patterns, architecture rules, and project conventions. It is intended for work after implementation, before a pull request, or during code review.
red-team-review
Unified adversarial review: v4.3 Strategic Matrix (MTA-004). 7-phase framework: Priors → Rubric → Adversarial Lenses → SWOT/TOWS → MCDA Decision Engine → Blind Spot/Kill Switch → Executive Summary. Absorbs: bias-detector.
include-test-files-that-assert-on-behavior-being-changed-in-decl
When delegating a task affected by this skill, include.
vhs-e2e-gif
Record a test run, a TUI session, or any terminal command as a GIF with VHS and attach it to a GitHub PR as a release-hosted asset, never a repo commit. Use when asked to record an e2e run, demo a fix on a PR, attach a GIF or screen recording to a pull request, show a test passing visually, or produce a terminal…