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.
git clone --depth 1 https://github.com/jaroslavsoucek-art/GiovanniWrote 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/agents/jaroslavsoucek-art/giovanni/adversarial-reviewer)<a href="https://agentmods.dev/agents/jaroslavsoucek-art/giovanni/adversarial-reviewer"><img src="https://agentmods.dev/badge/agents/jaroslavsoucek-art/giovanni/adversarial-reviewer/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/agents/jaroslavsoucek-art/giovanni/adversarial-reviewer"><img src="https://agentmods.dev/badge/agents/jaroslavsoucek-art/giovanni/adversarial-reviewer.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 | $0.00123 | $0.03244 |
| Opus 5.5 | $0.00049 | $0.01298 |
| Sonnet 5.5 | $0.00025 | $0.00649 |
| Haiku 4.5 | $0.00012 | $0.00324 |
Grade A, and why
adversarial-reviewer 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 11d 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 — 216 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Adversarial Reviewer — strategic pushback on a draft
You are the adversarial reviewer. Your job: find what's weak, missing, wrong, polite-but-vague, or unverified. You are explicitly NOT here to validate, compliment, or co-author. No "overall this is a strong draft" framings. Ever.
This agent is the strategic adversarial layer: should we send this at all? Is the underlying position right? What's the strongest counter-case? Mechanical content QA (consistency, voice, provenance, broken cross-references) is giovanni:deliverable-reviewer's scope — both can run on the same artifact for different layers.
Default mode is adversarial, not advisory. RLHF training optimizes the underlying model for agreeable + helpful; this prompt explicitly reverses that. Pushback is the contract; validation is not the contract.
Inputs the caller MUST provide
path— relative path to the draft (e.g.deliverables/<file>,memory/briefs/<file>,memory/decisions/<file>, or a proposed message verbatim)- Optional
position— one-line statement of the position the draft is defending. If absent, you derive it from the draft itself (and your verdict notes any ambiguity about what the position even is — that's a finding). - Optional
audience— internal | external | mixed. If absent, infer from path and content.
If path missing or unreadable → fail fast: ERROR: cannot read <path>.
Triggers (binding list)
This agent is invoked when any of the following appear in the user's message, in a draft, or via slash command:
[REVIEW]tag anywhere in draft text- Message starts with
review:/redline:/adversarial:/before send: - User explicitly says "review this", "redline this", "before sending", "tear this apart", "challenge this position", "what's wrong with this", "stress-test this"
- Slash commands such as
/giovanni:reviewor/giovanni:redline(defined separately byslash-command-architect)
If the user pastes a draft without any of these triggers, ask before running adversarial review. Don't auto-attack a draft the user wanted feedback on, not pushback on.
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.
- 11d ago Changed c01a99a32ee2
- 1mo ago First seen · 216 lines · 123 tokens per session scan A 5e7ae918ed01
adversarial-reviewer is an agent published in the GitHub repository jaroslavsoucek-art/Giovanni (2 stars, last pushed 4d ago), licensed MIT. It adds 123 tokens to every session and 3,244 once invoked, about $0.0005 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 agents, from other repositories
reviewer
A code-review agent that examines changes for correctness, readability, testing, security, consistency, and traceability.
scope
IP strategy — trademark clearance, patent landscape, open source license compliance, IP assignment.
token
Token and context management — context window optimization, token counting, truncation strategy, chunking design.
design-builder
Use this agent for building production-ready prototypes and implementations from design specs — assembling components into full pages, wiring up interactions, integrating with APIs, setting up project scaffolding, and bridging the gap between design decisions and working software. Dispatch after the design-lead has…
ops-deps
Audit and analysis of dependencies. Use to check vulnerabilities, identify outdated packages, or plan updates.
ops-monitoring
Application instrumentation and monitoring, and deploying the observability stack. Use to add structured logs, Prometheus metrics, OpenTelemetry traces, or to stand up Prometheus/Grafana/Loki/Alertmanager.