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 catcatcatstudio/cat-skills --skill adversarygit clone --depth 1 https://github.com/catcatcatstudio/cat-skillsWrote 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/catcatcatstudio/cat-skills/adversary)<a href="https://agentmods.dev/skills/catcatcatstudio/cat-skills/adversary"><img src="https://agentmods.dev/badge/skills/catcatcatstudio/cat-skills/adversary/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/catcatcatstudio/cat-skills/adversary"><img src="https://agentmods.dev/badge/skills/catcatcatstudio/cat-skills/adversary.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 | $0.00073 | $0.02043 |
| Opus 5 | $0.00036 | $0.01022 |
| Sonnet 5 | $0.00015 | $0.00409 |
| Haiku 4.5 | $0.00007 | $0.00204 |
Grade A, and why
adversary 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 — 186 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/adversary — Structured Dissent
You pressure-test decisions by arguing against them at full strength. Not generic critique — a concrete alternative, argued honestly, with a verdict at the end.
Step 1: Identify the Target
Determine what's being pressure-tested from the argument, conversation context, open files, and project state.
Valid targets: a decision, strategy, architecture choice, positioning, plan, approach, design direction, trade-off call — anything where a reasonable person could choose differently.
Invalid targets: bugs (just fix them), syntax questions, factual lookups. If the target isn't a judgment call, say so and stop.
If clear: State the target in 1-2 sentences. Proceed.
If ambiguous: Ask ONE question:
"I see [X] and [Y] in play. Which decision should I pressure-test?"
Step 2: Detect the Lens
Auto-detect from context. The lens shapes what the adversary attacks and what "better" means.
| Lens | When | Adversary focuses on |
|---|---|---|
| Architecture | Code structure, tech stack, build vs buy, system design | Simpler alternatives, scaling traps, maintenance burden, over-engineering, hidden coupling |
| Strategy | Business decisions, prioritization, resource allocation, market positioning | Market assumptions, opportunity cost, resource reality, what competitors would exploit, timing risk |
| Marketing | Copy, positioning, messaging, content strategy, audience targeting | Who it doesn't land with, what it actually communicates vs intends, stronger positioning, channel mismatch |
| Design | UI/UX, visual direction, interaction patterns, information architecture | Whether aesthetic serves function, usability under real conditions, edge cases that break the concept, simpler alternatives that work harder |
| General | Anything that doesn't fit the above | What this decision commits you to downstream — second-order lock-in, options it kills, the thing you'll wish you'd considered in 6 months |
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 · 186 lines · 73 tokens per session scan A 07057a28a3f5
adversary is a skill published in the GitHub repository catcatcatstudio/cat-skills (3 stars, last pushed 13d ago), licensed MIT. It adds 73 tokens to every session and 2,043 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
watch
Watch live IDX turnover and order books for unusual activity — value surges, auto-reject approach, one-sided depth, walls appearing or withdrawn, quiet stocks waking up, UMA flags — plus delayed print-by-print attribution and end-of-day broker context. Use when the user runs /watch, or asks what is trading right now…
stockbit-status
Check whether the Stockbit session is still valid — main session expiry, whether trading credentials are stored, and the current trading mode. Use when the user asks "am I still logged in to Stockbit", "has my session expired", "is trading on", or when a stockbit MCP tool starts failing with an auth error.
stockbit-auth
Log in to Stockbit and capture the session for this server — opens the browser login flow and verifies the token was stored. Use when the user asks to log in or re-authenticate to Stockbit, or after a status check reports the session expired, HTTP 401, or logged out.
recursive-decomposition
Handle tasks that exceed the context window by decomposing them: size and filter the input, chunk it, run recursive sub-agents on independent parts, verify on small windows, and synthesise programmatically, following the Recursive Language Models (RLM) research by Zhang, Kraska and Khattab (2025). Use when a task…
bandar-check
Answer "who is accumulating this stock?" for an IDX ticker using Stockbit broker-flow data. Use when the user asks about bandar, big money, accumulation, distribution, foreign flow, asing, or who is on the other side of the tape.
trade-with-guardrails
Place an order through the user's own Stockbit account safely — check the mode, size the position from a risk budget, preview, read the summary back to the human, and only then write. Use when the user asks to buy, sell, amend or cancel an order, or to size a trade.