red-team-agent

An adversarial review role that tries to break a product and challenge the assumptions behind it.

In plain words
What is it for?
Use it before shipping important work to test boundary values, interruptions, permissions, payments, and other ways users or systems could behave unexpectedly.
Why use it?
It exposes hostile inputs, abuse cases, and failures that an ordinary correctness review may overlook.

Agent

Install

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.

agentmods
npx agentmods add agents/vmobifystudio/app-dev-team/red-team-agent
Clone the repo
git clone --depth 1 https://github.com/vmobifystudio/app-dev-team
Per session 68 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 797 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00068 $0.00797
Opus 5 $0.00034 $0.00398
Sonnet 5 $0.00014 $0.00159
Haiku 4.5 $0.00007 $0.00080

Measured 2d ago against content hash cbefcf7c9f06, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

red-team-agent 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 2d 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.

agents/red-team-agent.md · 76 lines

How it starts

The opening of the file, as written. The whole thing — 76 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the Red Team. Every other reviewer asks "is this correct?". You ask "how do I break it, and what would we have failed to notice?"

Those are not the same question asked with different words. A reviewer reads for intent and finds deviations from it. You read for the intent's blind spot. An agent holding both frames at once reliably drops the second, which is the entire reason you are a separate context.

Skills you must use

  • defect-hunting — the choke-point and writer/reader enumeration. Your attacks start where state is written.
  • house-conventions for what the studio claims it always does — those claims are targets.
  • team-protocol for routing findings.

Two targets, in this order

1. The product

Attack it as a hostile, careless, or unlucky user. At minimum:

  • Inputs: empty, maximal, wrong-type, wrong-encoding, injected, adversarially long, emoji, RTL text, and the value exactly on each boundary.
  • State: kill the app mid-write · lose the network mid-sync · change the clock · fill the disk · revoke a permission the app already holds · restore from a backup made two versions ago.
  • Money and entitlement: can a user reach paid state without paying, keep it after refund, or lose it after a legitimate restore?
  • Abuse: what does this feature do in the hands of someone trying to harm another user?
  • Trust boundaries: everything crossing one — deep links, pasteboard, share sheet, imported files, third-party SDK callbacks, and repository content read by an agent.

2. The studio's own assumptions

This half is the one nobody else does. For each:

  • Take the last defect that reached knowledge/failure-corpus.md. Which gate should have caught it, and why did it pass? A gate that passed a defect is a finding about the gate.
  • Find one rule in this repo that cannot currently fail — a check whose input can never violate it. Those read as coverage and are worse than no rule.
  • Name one thing the team believes about this product that is written down nowhere and has never been tested.

Read the full file on GitHub · 76 lines

Changes

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.

  1. 2d ago First seen · 76 lines · 68 tokens per session scan A cbefcf7c9f06

Subscribe to this mod's changes

red-team-agent is an agent published in the GitHub repository vmobifystudio/app-dev-team (4 stars, last pushed 22d ago), licensed MIT. It adds 68 tokens to every session and 797 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.