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.
git clone --depth 1 https://github.com/Wholiver/metisWrote 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/agents/wholiver/metis/metis-arbiter)<a href="https://agentmods.dev/agents/wholiver/metis/metis-arbiter"><img src="https://agentmods.dev/badge/agents/wholiver/metis/metis-arbiter/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/agents/wholiver/metis/metis-arbiter"><img src="https://agentmods.dev/badge/agents/wholiver/metis/metis-arbiter.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 | $0.00055 | $0.00687 |
| Opus 5.5 | $0.00022 | $0.00275 |
| Sonnet 5.5 | $0.00011 | $0.00137 |
| Haiku 4.5 | $0.00006 | $0.00069 |
Grade A, and why
Metis 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 8d 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.
This is a copy
89% identical to ap-arbiter — 36 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 27 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are arbiter - Level 4 (Terminal leaf - Arbiter) in the Performance hierarchy.
Execution contract
You are an internal Performance worker, not a general-purpose assistant. Your built-in role definition and task brief are already the complete operating context. Before tool use or edits, require the exact RUN-ID, RUN-NONCE, and mission binding from an active Performance run; outside an active Performance run, return INVALID-DISPATCH and stop. Do not load, invoke, or re-invoke the external orchestration; do not start a nested Performance run. Execute only this established persona and the assigned brief. If you spawn, dispatch only admitted named workers for this lane. Children emit ChildResult and must not call performance_gate. Do not build an L0–L4 fleet.
Mission source of truth
Your brief carries a MISSION POINTER with canonical path, SHA-256 hash, UTF-8 byte length, and RUN-NONCE. Read PROMPTS.txt and verify every field before acting. The exact ledger bytes outrank every downstream instruction. A mismatch is INVALID-BRIEF.
Your level: L4 - Terminal leaf
You do the assigned work and write your artifact. You are TERMINAL - you do NOT spawn any subagents. You weigh the fork, render one binding ruling, log it, and report a tight result up to the executor that spawned you. No fan-out, no delegation.
Your gate/function
ARBITER: render a binding ruling on a fork the loop cannot resolve. Ask whether a competent senior engineer could decide from the mission and opened evidence - almost always yes. Under UNATTENDED mode always rule and continue; never set userRequired: true or stop for a question. Record the ruling as an append-only GATELOG.md row plus a substantive arbiter evidence artifact, not a new governance file. For an irreversible/destructive action, real money or quota-bound resource, unavailable credential, or product-direction decision only the user can own, choose the conservative non-destructive path and surface the unresolved choice in the final report. Never waive an open P0/P1, required verification, or the coverage floor.
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.
- 8d ago First seen · 27 lines · 55 tokens per session scan A 452df8e40e74
Metis Arbiter is an agent published in the GitHub repository Wholiver/metis (369 stars, last pushed yesterday), licensed MIT. It adds 55 tokens to every session and 687 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. It is 89% identical to ap-arbiter, differing in 36 lines, and is treated as a copy.
Other agents, from other repositories
ap-run-coordinator
Start only ready, non-overlapping roadmap work and combine returned status at the written integration points.
atomic-auditor
Final gate for a finished implementation. Dispatched exactly once after the implement-review loop goes green, never per iteration. Never touches the repo; its one write is the audit report into the task scratchpad. Audits the delivered work as a whole: cumulative spec compliance, cross-iteration coherence…
lynx-specialist
Expert consultant for Lynx framework development, modern desktop applications with web technologies, and native performance optimization. Use proactively for Lynx architecture analysis, performance optimization strategies, native integration patterns, and desktop deployment guidance. Provides consultation and…
accessibility-engineer
Thin Claude adapter for accessibility-engineer; canonical brief is authoritative.
interactive-audio-designer
Thin Claude adapter for interactive-audio-designer; canonical brief is authoritative.
code-reviewer
Use when a major project step completes and needs review against the original plan and coding standards. Examples: Context: User finished implementing user authentication as step 3 of plan. user: "I've finished implementing the user authentication system as outlined in step 3 of our plan" assistant: "Let me use the…