gauntlet

A repository onboarding test that checks whether a new contributor can use the available documentation and scripts to build and submit a small task.

In plain words
What is it for?
Use it to test onboarding guides, helper scripts, and task-card templates, and to track whether contributors can complete a task without help.
Why use it?
It reveals missing or misleading onboarding material by testing the contributor experience in a clean environment. Each attempt records what failed and updates the contributor materials.

Command

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 commands/dwarvesf/dwarves-kit/gauntlet
Clone the repo
git clone --depth 1 https://github.com/dwarvesf/dwarves-kit
Per session 111 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,045 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. 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 $0.00111 $0.03045
Opus 5 $0.00056 $0.01522
Sonnet 5 $0.00022 $0.00609
Haiku 4.5 $0.00011 $0.00304

Measured 2d ago against content hash f8906507a856, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

gauntlet scanned grade B with 1 finding 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.

Strips warnings and disclaimersmediumAnti-refusal

Omitting safety caveats hides risk from the user and is a common jailbreak preamble.

one line, offer the concrete fix (build it, or recommend the value). Do not lecture past those three beats, and do not start a round on a known-bad input.
commands/gauntlet.md · 194 lines

How it starts

The opening of the file, as written. The whole thing — 194 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the gauntlet orchestrator. Your job is to prove, or make true, this claim: a median-skill dev given ONLY what an outside contributor gets can set up, build one small feature, and submit it, unaided. The artifact under convergence is the CONTRIBUTOR SURFACE (onboarding docs, contributor docs, helper scripts, the task-card template), never the probe agent. When a round fails, the surface gets fixed.

Three kit shapes compose here; keep their roles straight:

  • Loop: the bounded-revise engine (Evaluator-Optimizer lineage; worked sibling: /kit:test-plan-review-team). Rounds, two-tier scan, severity-aware convergence, hard cap, honest halt.
  • Goal: the seed card is a goal contract (outcome + acceptance criteria + verification command + termination-on-blocker), and the probe agent runs under it. A gauntlet run therefore also validates the repo's card template itself.
  • Eval: the persisted run record is an eval artifact, metrics (severity trajectory, rounds-to-unaided-pass, intervention count), seed data (the card), and a defended verdict, in the lab-report spirit.

Inputs (confirm before round 1; ask for any missing one)

Input What it is Default
Target repo the repo whose surface converges current repo
Surface globs the files the reviser may touch CONTRIBUTING.md, docs/onboarding*, README, scripts/onboard-*, scripts/preview-*, the card template
Tier 1 command the repo's deterministic onboarding suite e.g. ONBOARDING_TESTS=1 npx vitest run test/onboarding
Clean-room recipe how a fresh env is built from committed state e.g. test/onboarding/run.sh (docker build from git archive HEAD)
Seed card ONE small real task in the repo's card template, <= 1 agent-day, low-slop type (test backfill, doc-drift fix, small reproduced bug) pick one; record it in the run dir
Submission checker deterministic validator of the probe's output e.g. test/onboarding/gauntlet/check-submission
Probe model the synthetic dev Sonnet, DELIBERATELY not frontier: a smarter model succeeds despite bad docs and destroys the signal
Probe credentials the ONLY secret the clean room gets one spend-capped model API key; never CF / 1P / GitHub credentials
Round cap hard stop 3
Runner host where rounds physically run kit.toml [gauntlet] runner_host: "local", or an ssh alias (an always-on host suits long campaigns and the resume scenario); run-remote.sh ships committed state, runs there, pulls the record back. The probe key resolves ON the runner host from gauntlet.probe_key_ref, never traveling over ssh

Read the full file on GitHub · 194 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 · 194 lines · 111 tokens per session scan B f8906507a856

Subscribe to this mod's changes

gauntlet is a command published in the GitHub repository dwarvesf/dwarves-kit (11 stars, last pushed 2d ago), licensed MIT. It adds 111 tokens to every session and 3,045 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it B with 1 finding (strips warnings and disclaimers). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.