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 EzraApple/skills-init --skill adversarial-reviewgit clone --depth 1 https://github.com/EzraApple/skills-initWrote 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/ezraapple/skills-init/adversarial-review)<a href="https://agentmods.dev/skills/ezraapple/skills-init/adversarial-review"><img src="https://agentmods.dev/badge/skills/ezraapple/skills-init/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/ezraapple/skills-init/adversarial-review"><img src="https://agentmods.dev/badge/skills/ezraapple/skills-init/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 | $0.00059 | $0.01930 |
| Opus 5 | $0.00030 | $0.00965 |
| Sonnet 5 | $0.00012 | $0.00386 |
| Haiku 4.5 | $0.00006 | $0.00193 |
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 10d 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 — 234 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Adversarial Review
Use this when ordinary review is not enough. The workflow is independent criticism, root-cause clustering, attempted takedown of findings, synthesized action items, and reruns after fixes when edits are in scope.
The goal is not to generate more comments. The goal is to find the issues that survive skeptical validation.
Core Rules
- Reviewer and validator passes are readonly. Edit only after synthesis, and only to address validated action items.
- Do not reuse the implementation chat as evidence. Reviewers get the review packet, not the builder's defense of the code.
- No issue quota. A clean review is valid if the probes were meaningful.
- Every finding needs a concrete failure path, target file or symbol, evidence, and suggested fix.
- Vague concerns are not findings.
- The user gets a synthesized action list, not raw reviewer transcripts.
- Default to the hardening loop unless the user asks for report-only review.
If your runtime supports subagents, use fresh readonly subagents for reviewer and validator passes. If it does not, run the same lanes sequentially in fresh notes: reset assumptions between lanes, do not edit during lane work, and do not let one lane's findings bias another until clustering.
Scope Selection
Pick the narrowest concrete scope before launching reviewer lanes:
- If the user gives a PR URL or number, review that PR with the local GitHub tooling available in the environment.
- If the user names files or directories, review those paths plus directly relevant callers.
- If the user gives a base or revision range, review that diff.
- If the current branch has a resolvable PR, review that PR.
- Otherwise review the branch diff from the default branch using the merge base.
- Otherwise review staged, unstaged, and untracked changes.
If the resolved diff is empty, stop and say there is nothing to review.
Dimension Selection
Choose lanes from the user's requested dimensions, changed surfaces, and risk profile. Do not run the same fixed panel every time.
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.
- 10d ago First seen · 234 lines · 59 tokens per session scan A a4fa46382607
adversarial-review is a skill published in the GitHub repository EzraApple/skills-init (2 stars, last pushed 1mo ago), licensed MIT. It adds 59 tokens to every session and 1,930 once invoked, about $0.0003 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
sdd-apply
Skill "sdd-apply" from Gentleman-Programming/gentle-ai, covering execution role, language domain contract, purpose, what you receive and execution and persistence contract.
gentle-ai-collab-perfect
Trigger: contributing to Gentleman-Programming/gentle-ai as an external collaborator. Strict issue-first workflow, honest PR bodies, contributor-vs-maintainer scope, chained-PR strategy, verification protocol, docstring coverage. Load whenever the active repo is Gentleman-Programming/gentle-ai and any part of the…
issue-creation
Trigger: issue creation, bug reports, feature requests, or issue approval. Create and triage GitHub issues from repository evidence.
sdd-tasks
Break an SDD change into implementation tasks. Trigger: orchestrator launches task planning for a change.
sdd-verify
Skill "sdd-verify" from Gentleman-Programming/gentle-ai, covering execution role, language domain contract, activation contract, hard rules and decision gates.
sdd-spec
Write SDD delta specs with requirements and scenarios. Trigger: orchestrator launches spec work for a change.