antfleet/bench — benchmark mirror index

The single home for AntFleet's public benchmark mirrors. Each bench-* repo is a standalone, faithful snapshot of an external project that AntFleet ran through its two-model unanimous security-review methodology. They live as separate repos on purpose — the review is only credible if it runs against the real, unmodified source at a pinned commit. This repo is the catalog, not a monorepo.

Public receipts pin to the upstream repo (e.g. github.com/mudler/LocalAI/blob/<sha>/...), not to the mirror. The mirror is AntFleet's immutable copy of exactly what was reviewed, in case upstream force-pushes or disappears.

Status: all mirrors archived (2026-07-27)

All 23 mirrors below are archived (read-only). They were forked solely to run the AntFleet two-model audit; antfleet[bot] was installed for benchmark purposes only, and no further PRs will be opened against them. Archiving declutters the org's active repo list while preserving everything the mirrors exist for:

• they stay public at their existing URLs,
• PR history and review comments are intact,
• they remain fully cloneable / fetchable at the reviewed SHA, so repro-verify (which only ever reads — clone/checkout/fetch, never push) is unaffected.

Archiving only blocks new pushes, new PRs, new issues, and webhook deliveries — none of which these mirrors need. Reversible per repo: gh api -X PATCH repos/AntFleet/bench-<name> -F archived=false.

Naming convention (house rule)

• Every benchmark mirror is named bench-<upstream-repo>. Prefix, never suffix.
• The prefix makes all mirrors sort as one contiguous block in the org repo list.
• Non-mirror bench products keep product names and are not in this index (e.g. open-evmbench = public leaderboard; aeon-template = productized template).
• Repo description format: AntFleet benchmark mirror of <owner>/<repo> — two-model security review methodology.

See HOUSE-RULES.md for the full org tidy-up rules.

Mirrors (23)

Mirror	Upstream	Visibility	Status
bench-agent-autonomopoly	Liquid-Protocol-Ops/agent-autonomopoly	public	archived
bench-agent-openhuman	tinyhumansai/openhuman	public	archived
bench-agentfloat	ronkenx9/agentfloat-hook	public	archived
bench-antseed	AntSeed/antseed	public	archived
bench-bankrskills	BankrBot/skills	public	archived
bench-berry	levyonchain/Berry-Juicer-Public	public	archived
bench-bitterbot-desktop	Bitterbot-AI/bitterbot-desktop	public	archived
bench-claude-mem	thedotmack/claude-mem	public	archived
bench-doppler	whetstoneresearch/doppler	public	archived
bench-hermes-desktop	fathah/hermes-desktop	public	archived
bench-localai	mudler/LocalAI	private	archived
bench-miroshark	aaronjmars/MiroShark	public	archived
bench-mythos-router	thewaltero/mythos-router	public	archived
bench-openclaude	Gitlawb/openclaude	public	archived
bench-orlixai	tylerbroqs/orlixai	public	archived
bench-potatopad	itsfriedpotato/potatopad	public	archived
bench-reppo-cli	Reppo-Labs/reppo-cli	public	archived
bench-reppo-polyagent	Reppo-Labs/reppo-polyagent	public	archived
bench-roboco	rennf93/roboco	public	archived
bench-shh	privashh/shh	public	archived
bench-synthethic	AISynthetics/synthetic-users	public	archived
bench-virtuals-acp-cli	Virtual-Protocol/acp-cli	public	archived
bench-virtuals-acp-node-v2	Virtual-Protocol/acp-node-v2	public	archived

All 23 mirror descriptions follow the § 1 format AntFleet benchmark mirror of <owner>/<repo>.

Not in this index (bench-named but not mirrors)

Repo	What it actually is
open-evmbench	Public leaderboard for AI smart-contract vulnerability work — a product, not a mirror.
aeon-template	Productized two-model-consensus review template (was aeon-bench). Renamed to stop it reading as a mirror.