Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/yunsii/wezdecknpx agentmods add skills/yunsii/wezdeck/adversarial-reviewWrote 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/yunsii/wezdeck/adversarial-review)<a href="https://agentmods.dev/skills/yunsii/wezdeck/adversarial-review"><img src="https://agentmods.dev/badge/skills/yunsii/wezdeck/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/yunsii/wezdeck/adversarial-review"><img src="https://agentmods.dev/badge/skills/yunsii/wezdeck/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 | $0.00085 | $0.02599 |
| Opus 5 | $0.00043 | $0.01300 |
| Sonnet 5 | $0.00017 | $0.00520 |
| Haiku 4.5 | $0.00009 | $0.00260 |
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 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 — 225 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Adversarial review (platform skill — single source)
Who runs: the coding agent (OpenClaw Main / Claude-TUI / Codex-TUI), not the human.
Never tell the user to copy-paste run.sh as the primary path.
Never replace this with a solo monologue and call it 对抗审查 (that is 设计批判).
This directory is the only skill + runner unit. Other paths are symlinks (user-level or in-repo discovery). Do not maintain a second SKILL.md body.
Who runs what
| Actor | Duty |
|---|---|
| Human | Intent only:「对抗审查」「审一下这批改动」「按推荐审查」 |
| Any agent | Load this skill → resolve TOOL + TARGET → run → disclose |
| Host headless backends | claude / codex / grok as find/refute (not ACP by default) |
When to load / run
- User asks for 对抗审查 / adversarial review / multi-role review
- Write-task checklist: acceptance suggests review
- Runtime code/scripts changed (any git TARGET the agent may review)
- After editing this toolkit itself →
dogfood
Skip / redirect:
- Pure docs/tests-only (runner auto-skips); user explicitly skips with reason
- Design docs / RFC / ADR / 方案提案 (no runtime diff): do not force this
skill. Use 设计评审 (structured checklist in host
validation.md→ Design proposal review) or brainstorm only when the user still wants alternative approaches. Never label a doc critique 对抗审查.
Multi-role minimum
| Role | Stance | Runner stage |
|---|---|---|
| find / reviewer | guilty-until-proven | stage 1 |
| refute / refuter | kill weak findings | stage 2 |
| repro | empirical in sandbox | stage 3 |
- Prefer different families than the writer (strategy B via
--writer). - Same capability twice OK → label SINGLE-MODEL / degraded; do not skip refute.
- Solo essay without opposite roles → 设计批判, not 对抗审查.
Paths: TOOL vs TARGET (critical)
Do not assume ./scripts/dev/adversarial-review exists in the repo under review.
| Name | Meaning |
|---|---|
| TOOL_HOME | This skill/runner directory (where run.sh lives) |
| TARGET | Git repo being reviewed (--repo or current git toplevel) |
What ships with it
20 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.
- lib/context-pack.sh 21 KB runs code
- lib/findings-schema.json 1.4 KB
- lib/impact/extract-symbols.sh 2.0 KB runs code
- lib/impact/impact.sh 5.1 KB runs code
- lib/impact/resolvers/grep.sh 2.6 KB runs code
- lib/impact/test-impact.sh 2.4 KB runs code
- lib/impact/test-pack-slice.sh 4.2 KB runs code
- lib/provider.sh 7.4 KB runs code
- lib/providers/claude.sh 740 B runs code
- lib/providers/codex.sh 2.3 KB runs code
- lib/providers/grok.sh 902 B runs code
- lib/roles-lib.sh 672 B runs code
- lib/roles.conf 1.2 KB
- lib/rubric-lib.sh 849 B runs code
- lib/rubric.conf 1.3 KB
- lib/select-backends.sh 5.2 KB runs code
- prompts/critic.md 2.6 KB
- prompts/refute.md 1020 B
- prompts/repro.md 1.1 KB
- run.sh 27 KB runs code
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 · 225 lines · 85 tokens per session scan A 332810f06317
adversarial-review is a skill published in the GitHub repository yunsii/wezdeck (5 stars, last pushed yesterday), licensed MIT. It adds 85 tokens to every session and 2,599 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 2026-08-31.
Other skills, from other repositories
printing-press-output-review
Internal sub-skill: agentic review of a printed CLI's sampled command output for plausibility issues that rule-based checks can't encode (substring-match relevance, format bugs, silent source drops, ranking failures). Invoked via the Skill tool by the main printing-press skill at Phase 4.85 and printing-press-polish…
systemic-issue-triage
Trigger: new issue, bug report, triage, backlog, issue flood, community report, root cause, dead-end, blocked user. Attack issues by root class, never one-by-one; fixes must shrink the system, not grow it.
issue-root-resolution
Trigger: root audit, atacar la raíz, issue roots, backlog roots, mechanism map, deletion-driven fix, resolver issues de raíz, close outdated issues. Audit and resolve issue clusters by verified root cause.
review-loop
Run the adversarial verification loop — implement, then hand the change to a fresh checker that did not write it, fix what it finds, and re-dispatch until APPROVE. Use before claiming any behavioural change is done, and on requests like "review loop", "adversarial review", "independent review", "get this verified"…
semgrep-rule-creator
Creates custom Semgrep rules for detecting security vulnerabilities, bug patterns, and code patterns. Use when writing Semgrep rules or building custom static analysis detections.
fix
Dispatch fix subagent for FIX-FIRST gaps from review, re-review, then diagnose unresolved failures after 2 loops.