dar-arbiter

dar-arbiter is an agent for Claude Code from Kaos599/Deep-Ass-Research. It costs 17 tokens per session (424 once invoked), scanned A, original, MIT.

A decision-making agent for claims that remain disputed during a verification process. It reviews the original sources and earlier reviewers' objections, then records whether each claim is confirmed, rejected, or still unresolved.

In plain words
What is it for?
Use it to rule on contested research or fact-checking claims, update their status, record the decision in a verification log, and create follow-up items for open questions.
Why use it?
It prevents uncertain claims from being silently accepted or removed when earlier checks disagree. Unresolved claims remain marked as disputed and are sent back for further investigation.

Agent for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: model in frontmatter.

Part of the deep-ass-research plugin — 1 command, 7 agents shipped together

Good fit Use it to rule on contested research or fact-checking claims, update their status, record the decision in a verification log, and create follow-up items for open questions.

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/kaos599/deep-ass-research/dar-arbiter
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.

Clone the repo
git clone --depth 1 https://github.com/Kaos599/Deep-Ass-Research

Made for: Claude Code.

Or install deep-ass-research, the plugin that ships this one along with the rest of its 1 command, 7 agents.

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 dar-arbiter

README.md
[![agentmods](https://agentmods.dev/badge/agents/kaos599/deep-ass-research/dar-arbiter.svg)](https://agentmods.dev/agents/kaos599/deep-ass-research/dar-arbiter)
Your own site
<a href="https://agentmods.dev/agents/kaos599/deep-ass-research/dar-arbiter"><img src="https://agentmods.dev/badge/agents/kaos599/deep-ass-research/dar-arbiter.svg" alt="Measured on agentmods" height="20"></a>
Per session 17 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 424 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
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 $0.00017 $0.00424
Opus 5 $0.00009 $0.00212
Sonnet 5 $0.00003 $0.00085
Haiku 4.5 $0.00002 $0.00042

Measured 7d ago against content hash 35d72f448870, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-07, from the pricing page.

Security

Grade A, and why

dar-arbiter 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 7d 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.

adapters/claude-code/plugin/agents/dar-arbiter.md · 30 lines

What it actually says

Role: Arbiter (Verify phase, contested claims only)

You are the Arbiter. You only see claims the skeptics could not cleanly resolve — CONTESTED (1-of-3 refuted) or DISPUTED (2-of-3 refuted) — plus their raw sources and the skeptics' verdicts. You make the final call. Cost control: you do not touch CONFIRMED claims.

What to do

For each contested claim:

  1. Read the cached raw/ source(s) and the skeptics' disconfirming / scope_correction notes.
  2. Decide one of:
    • CONFIRM — the refutation doesn't hold; restore the claim (optionally with a corrected scope/confidence).
    • KILL — the refutation holds; set status: refuted. Do not delete the note — it stays as recorded disconfirming evidence.
    • OPEN_QUESTION — genuinely unresolved with current sources; downgrade to status: disputed and push a lead back onto _open-threads.md so it's reported honestly, not hidden.
  3. You may do one FETCH only to break a true tie. Otherwise rule on the evidence at hand.
  4. Update the claim note's status and append your ruling to _verify-log.md.

Output

{
  "rulings": [
    {"claimId":"S-201","ruling":"CONFIRM|KILL|OPEN_QUESTION","new_status":"verified|refuted|disputed","corrected_claim":"… or null","reasoning":"…"}
  ]
}

Bias: when truly uncertain, prefer OPEN_QUESTION over a confident CONFIRM/KILL — an honest open question is a valid outcome; a falsely-confident verdict is the failure mode DAR exists to prevent.

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. 7d ago First seen · 30 lines · 17 tokens per session scan A 35d72f448870

Subscribe to this mod's changes

dar-arbiter is an agent published in the GitHub repository Kaos599/Deep-Ass-Research (4 stars, last pushed 29d ago), licensed MIT. It adds 17 tokens to every session and 424 once invoked, about $0.0001 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.