gauntlet

gauntlet is a cursor rule for Cursor from robertsfeir/atelier-pipeline. It costs 8,143 tokens per session, scanned A, original, Apache-2.0.

A coordinated audit in which separate agents review an application’s architecture, code, tests, security, interface, product fit, and code quality.

In plain words
What is it for?
Use it for a broad, user-started review that runs review rounds in parallel, checks upstream specifications first, and produces a combined findings register.
Why use it?
It gathers independent findings across the codebase while keeping reviewers from influencing one another before results are combined.

Cursor rule for Cursor

Written for Cursor: a Cursor plugin manifest. Also seen: mentions subagents; positional $N argument.

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 rules/robertsfeir/atelier-pipeline/gauntlet
Clone the repo
git clone --depth 1 https://github.com/robertsfeir/atelier-pipeline

Made for: Cursor.

Wrote 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.

agentmods badge for gauntlet

README.md
[![agentmods](https://agentmods.dev/badge/rules/robertsfeir/atelier-pipeline/gauntlet.svg)](https://agentmods.dev/rules/robertsfeir/atelier-pipeline/gauntlet)
Your own site
<a href="https://agentmods.dev/rules/robertsfeir/atelier-pipeline/gauntlet"><img src="https://agentmods.dev/badge/rules/robertsfeir/atelier-pipeline/gauntlet.svg" alt="Measured on agentmods" height="20"></a>
Per session 8,143 This file is loaded in full into every session.
When invoked 8,143 The same file — it is already loaded in full.
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 5.1 $0.08143 $0.08143
Opus 5 $0.04071 $0.04071
Sonnet 5 $0.01629 $0.01629
Haiku 4.5 $0.00814 $0.00814

Measured 6d ago against content hash 02ae44d48ba4, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

gauntlet 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 6d 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.

.cursor-plugin/rules/gauntlet.mdc · 1,026 lines

How it starts

The opening of the file, as written. The whole thing — 1,026 lines — stays where its author put it; the contents beside it link to each section on GitHub.

The Gauntlet — Full-Spectrum Codebase Audit

Run a full-spectrum, multi-agent audit covering every layer of the application: architecture, implementation, testing, security, frontend/UX, product alignment, and blind code review. All rounds run in parallel — contamination is prevented by instruction, not by timing. Phase 2 begins only after all active rounds complete.

Trigger: User-initiated only. Eva does not auto-trigger the Gauntlet.


Execution Rules (read before starting Phase 0)

  1. Parallel execution. All active rounds in Phase 1 run concurrently. Eva launches every active round as a background agent in a single message. Phase 2 (Combined Register) begins only after all active rounds have written their findings files.
  2. No cross-contamination. Agents do not read each other's findings files during their own review phase. No agent sees what any other agent wrote.
  3. Brain-first. If brain is available (brain_available: true in pipeline-state.md), query it before invoking each agent. Inject results as <brain-context>.
  4. Upstream-first. Every agent reads specs, UX docs, and ADRs before reading code.
  5. No hedging. If something is wrong, state it with file and line. Avoid "consider whether" language for clear violations.
  6. Composability thread. Every agent evaluates composability from their lens. This is a first-class concern alongside correctness.
  7. Positive observations required. Every agent must include at least three things done well. A report that only surfaces problems is not balanced.
  8. All rounds run Opus. No model downgrade regardless of task scope or pipeline sizing. The Gauntlet is explicitly high-effort. Omitting the Opus model parameter is a Gauntlet configuration error.
  9. Output directory. All findings files go to docs/reviews/gauntlet-{YYYY-MM-DD}/. Eva announces the path at the start of Phase 0. The user decides whether to commit this directory to version history after reviewing the report.

Read the full file on GitHub · 1,026 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. 6d ago First seen · 1,026 lines · 8,143 tokens per session scan A 02ae44d48ba4

Subscribe to this mod's changes

gauntlet is a cursor rule published in the GitHub repository robertsfeir/atelier-pipeline (25 stars, last pushed 2mo ago), licensed Apache-2.0. It adds 8,143 tokens to every session, about $0.0407 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 2026-08-30.