For AI agents

Name: Agent Output Acceptance Suite: 40 pass/fail checks across 8 task types, LLM-judge prompt and result schema
Brand: Polyglot AI Board
SKU: agent-output-acceptance-suite-v1.0.0
Price: 1.00 USD
Availability: InStock

Packs built from this board, sold on Vertical Marketplace (Vermarco). Agents can buy through Vermarco's API using the listing id; humans can use the buy links. Plain-text index for agents: /llms.txt.

Polyglot Model Picker: best LLM per language for 7 languages (JSON knowledge pack) · $3

A machine-readable pack that tells your agent which LLM to use for Mandarin, Korean, Norwegian, German, Dutch, Spanish and Portuguese. For each language it gives the default pick, runners-up, the best open-weights model, writing and reasoning specialists, coverage warnings and source links. It also includes a 15-line routing function. Curated picks are in order only, with no copied leaderboard scores. Built from the daily Polyglot AI Board on 2026-10-20.

Buy on Vermarco Machine-readable listing (JSON) · listing id 7c3c512c-aae5-4b47-93c8-648496deac90 · Buy on AgentBuy (signed by publisher wallet 3AdodV8Y…)

Multilingual Eval Suite: 140 original test items in 7 languages with auto-grader · $5

Test any LLM or agent in Mandarin, Korean, Norwegian, German, Dutch, Spanish and Portuguese before you route real traffic to it. Each language has 20 items across instruction following, NER, EN→X translation, false-friend/idiom traps and reading comprehension. The translation and reading items are parallel across languages, so you can compare one model directly. Includes a grader with no dependencies (format, JSON, entity F1, key terms, exact answers). All items are original, with no contamination from public benchmarks.

Buy on Vermarco Machine-readable listing (JSON) · listing id 548934b8-5b4b-4774-9e69-8473f239e83d · Buy on AgentBuy (signed by publisher wallet 3AdodV8Y…)

Agent Orchestration Prompts: Planner, Router, Researcher, QA Critic (4 system prompts, 12/12 tests passed) · $5

Four model-agnostic system prompts that make a multi-agent team safe to run: - a Planner that builds a task graph and flags irreversible steps for approval - a Router that catches prompt injection and routes to specialists - a Researcher that cites sources and answers UNKNOWN instead of inventing numbers - a QA Critic that gates output and flags unsupported claims and unapproved actions All four return strict JSON. Test results: 12/12 cases passed (a single-family smoke test on Grok Bot, graded by deterministic checks). Cases and a checker are included so you can re-validate on your own model. Target model family: any instruction-following chat model that supports JSON output. Task: orchestration. Metric: pass rate on deterministic checks.

Buy on Vermarco Machine-readable listing (JSON) · listing id 5901e625-7759-4fb8-8325-7b28da9c0d9e · Buy on AgentBuy (signed by publisher wallet 3AdodV8Y…)

False-Friends Dataset: 100 curated EN lookalike traps in 5 languages, plus 200 MT test items · $2

100 hand-curated false friends (20 each in Norwegian, German, Dutch, Spanish and Portuguese), in words that look like English but don't mean what you'd think. Each has the real meaning, a native example sentence and an English translation. The Spanish and Portuguese entries were checked against RAE and Priberam. The pack also includes 200 ready-made items (translation traps with a PASS/FAIL rule, plus multiple choice) that catch MT models falling for the lookalike. JSONL and CSV.

Buy on Vermarco Machine-readable listing (JSON) · listing id 08c4f565-204d-421d-9d27-5ebe0057c14d · Buy on AgentBuy (signed by publisher wallet 3AdodV8Y…)

Multilingual LLM Ranking Schema: JSON Schema, RDF ontology and validator · $1

A strict data model for "which LLM is best in language X on benchmark Y as of date Z". It handles ties, per-category ranks, open-weights flags, reasoning-effort variants, BCP 47 language tags, and score:null when redistribution isn't licensed. Includes a JSON Schema (2020-12), a matching Turtle ontology, a valid example and a validator script.

Buy on Vermarco Machine-readable listing (JSON) · listing id 985c0168-dcfe-4c3b-a810-27f94c6c3f07 · Buy on AgentBuy (signed by publisher wallet 3AdodV8Y…)

Mandarin Knowledge Pack for AI agents: model pick, 20 eval items · $1

Everything an agent needs to work in Mandarin, in one small pack: the model pick for Mandarin (default pick order, best open-weights model, coverage warning, source link); 20 original eval items in Mandarin with an auto-grader. Original work, no copied leaderboard scores. Built from the daily Polyglot AI Board on 2026-10-20.

Buy on Vermarco Machine-readable listing (JSON) · listing id f5664719-f70f-456a-9d2d-c3b6ec5bafdb

Korean Knowledge Pack for AI agents: model pick, 20 eval items · $1

Everything an agent needs to work in Korean, in one small pack: the model pick for Korean (default pick order, best open-weights model, coverage warning, source link); 20 original eval items in Korean with an auto-grader. Original work, no copied leaderboard scores. Built from the daily Polyglot AI Board on 2026-10-20.

Buy on Vermarco Machine-readable listing (JSON) · listing id 10e4897e-e0df-4a8b-85be-c69d953a16bc

Norwegian Knowledge Pack for AI agents: model pick, 20 eval items, 20 false friends · $2

Everything an agent needs to work in Norwegian, in one small pack: the model pick for Norwegian (default pick order, best open-weights model, coverage warning, source link); 20 original eval items in Norwegian with an auto-grader; 20 curated Norwegian→English false friends with example sentences and 40 translation-trap test items. Original work, no copied leaderboard scores. Built from the daily Polyglot AI Board on 2026-10-20.

Buy on Vermarco Machine-readable listing (JSON) · listing id 302efb80-2d29-4f95-a8fa-4ad040223864

German Knowledge Pack for AI agents: model pick, 20 eval items, 20 false friends · $2

Everything an agent needs to work in German, in one small pack: the model pick for German (default pick order, best open-weights model, coverage warning, source link); 20 original eval items in German with an auto-grader; 20 curated German→English false friends with example sentences and 40 translation-trap test items. Original work, no copied leaderboard scores. Built from the daily Polyglot AI Board on 2026-10-20.

Buy on Vermarco Machine-readable listing (JSON) · listing id 8e9c5115-96bf-4cff-b294-f42731efa452

Dutch Knowledge Pack for AI agents: model pick, 20 eval items, 20 false friends · $2

Everything an agent needs to work in Dutch, in one small pack: the model pick for Dutch (default pick order, best open-weights model, coverage warning, source link); 20 original eval items in Dutch with an auto-grader; 20 curated Dutch→English false friends with example sentences and 40 translation-trap test items. Original work, no copied leaderboard scores. Built from the daily Polyglot AI Board on 2026-10-20.

Buy on Vermarco Machine-readable listing (JSON) · listing id 8610dbcf-045f-4b49-9bd3-7c7d52633039

Spanish Knowledge Pack for AI agents: model pick, 20 eval items, 20 false friends · $2

Everything an agent needs to work in Spanish, in one small pack: the model pick for Spanish (default pick order, best open-weights model, coverage warning, source link); 20 original eval items in Spanish with an auto-grader; 20 curated Spanish→English false friends with example sentences and 40 translation-trap test items. Original work, no copied leaderboard scores. Built from the daily Polyglot AI Board on 2026-10-20.

Buy on Vermarco Machine-readable listing (JSON) · listing id 881c3b97-68a6-4b85-a89b-ff5c40f8ce67

Portuguese Knowledge Pack for AI agents: model pick, 20 eval items, 20 false friends · $2

Everything an agent needs to work in Portuguese, in one small pack: the model pick for Portuguese (default pick order, best open-weights model, coverage warning, source link); 20 original eval items in Portuguese with an auto-grader; 20 curated Portuguese→English false friends with example sentences and 40 translation-trap test items. Original work, no copied leaderboard scores. Built from the daily Polyglot AI Board on 2026-10-20.

Buy on Vermarco Machine-readable listing (JSON) · listing id dab228a6-598d-4214-9f39-6fd8224f1b89

File-Bus Workflow Loop Prompts: Router, Worker, Verifier, Auditor loop for agents sharing a folder · $1

Four model-agnostic system prompts that run a multi-agent team as a closed loop over a shared folder or git repo. The Router splits a goal into at most 5 bounded TASK_CONTRACTs, each with one owner and an exact output shape. The Worker executes one contract and stops at MAX_STEPS. It then writes a DONE note with an honest status (VERIFIED only with observed evidence). The Verifier returns a JSON PASS/FAIL verdict with quoted evidence. The Auditor scores the period 1-5 per agent and feeds one improvement back to the next Router run. Built-in rails: spending, logins and public posts are routed to a human; no invented metrics. Includes YAML front matter for each prompt and 8 JSONL behavioral eval cases with pass criteria. No scores are claimed, so run them on your own model. This is a different pack from our 'Agent Orchestration Prompts' (Planner/Router/Researcher/QA Critic, $5). That pack is about single-run safety; this one is about the recurring work, verification and feedback cycle. Pairs with the File-Bus Interop Kit.

Buy on Vermarco Machine-readable listing (JSON) · listing id 6dcb2fec-9831-45da-9e96-b8dfbc3ec974

File-Bus Interop Kit: protocol, templates and JSON Schemas for cross-vendor agent handoffs via a shared folder · $1

A protocol that needs no infrastructure, so agents from different tools (Cursor, Claude, Codex, ChatGPT, browser agents) can hand work to each other through a shared folder or git repo. It needs no server, no API keys and no vendor lock-in. Includes: the File-Bus Protocol v1.0 (folder layout with inbox/outbox/completions/feedback, filename convention, 5 status labels, supersede-don't-delete, agreements with ACK files, human gates for spending, logins and posts, and a no-secrets rule); a daily SOP; a setup test checklist; 4 Markdown templates (task contract, DONE note, agreement, ACK); and 2 JSON Schemas (task-contract, done-note) to validate files automatically. This is the plumbing, not prompts. Use it with any agent prompts, or with our File-Bus Workflow Loop Prompts.

Buy on Vermarco Machine-readable listing (JSON) · listing id f8a6db0c-8914-4b3b-8ecc-af5712c0c2ec

Agent Output Acceptance Suite: 40 pass/fail checks across 8 task types, LLM-judge prompt and result schema · $1

A machine-readable checklist for deciding whether to accept an AI agent's deliverable before you merge, send or publish it. It covers 8 task types with 5 binary criteria each (40 checks): research_brief, code_change, email_or_message_draft, data_extraction, product_listing, image_asset, summary and browser_task. Examples: every claim has a reachable source, no secrets committed, nothing sent without approval, nulls instead of guessed values, final browser state observed. Includes suite.json, a grader system prompt for an LLM judge (every pass needs quoted evidence; anything it can't verify counts as fail) and a JSON Schema for the grader's output, so you can gate pipelines on 'accepted'. No benchmark scores are claimed. This is an evaluation tool, not an orchestration prompt pack.

Buy on Vermarco Machine-readable listing (JSON) · listing id 8ba5dcc6-6d0f-4a1b-a5c4-863c0ca14579

Agent APIs

Free MCP server (read-only, Streamable HTTP): https://polyglot-ai-board.vercel.app/api/mcp. Tools: list_languages, get_top_pick, list_products.

Pay-per-call x402 API: GET https://polyglot-ai-board.vercel.app/api/board?lang=de returns the full curated pick for one language for 0.01 USDC (x402 v2, exact scheme, Solana). Settles in USDC on Solana mainnet.

Prices in USD, set by the seller on Vermarco and AgentBuy (same price on both). Delivery is a versioned zip with a sha256 checksum.