Overwing
概要
作者による説明(英語の原文)
Guardrails for LLM output: pass / fail / review verdicts and one recommended action. Hosted or npm.
PopNx MCPの3つの柱
この3つを1つのスコアにまとめることはしません。意見、測定、人気はそれぞれ別のものだからです。
- 利用者の声iレビューはまだありませんレビューは今後のステップで、アカウント機能と一緒に始まります
- 私たちの基本チェックi3項目中3項目に合格公開情報からチェック。コードはまだスキャンしていません
- 利用状況iとても新しい↓ 106/週 · ★ –
できること
このサーバーのやさしい解説はまだ書いていません。以下はレジストリから読み取った事実です。
- 接続方法: パソコンでもクラウドでも
- トランスポート: stdio, Streamable HTTP
- カテゴリ: 検索・AIの記憶
- 入力が必要なもの:OVERWING_API_KEY, OVERWING_AGENT_KEY, Authorization
インストール
使っているAIアプリを選んで、設定をコピーしてください。
インストールの前に
- 基本チェック:3項目中3項目に合格
- 作者のサーバーで動きます。あなたが頼んだ内容は、作者のサーバーを通ります。
- できること:41件(ツール) evaluate, evaluate_batch, list_rule_sets, get_rule_set …
- 自分で入力するもの:
OVERWING_API_KEY,OVERWING_AGENT_KEY,OVERWING_BASE_URL,Authorization
「秘密」と表示された値は、あなたのアカウントの鍵です。絶対に人に教えたり、チャットに貼り付けたりしないでください。
必要なもの
どのアプリを使っていますか?
上でアプリを選んでください。選んだアプリを覚えておくので、次からはボタンに「…にインストール」と表示されます。
このアドレスはリクエストの header にキーが必要ですが、コネクタを追加するメニューでは設定できません。代わりにClaude Code、Cursor、VS Codeを使ってください。
または、パソコンで動かす
Claude Desktop → Settings → Developer → Edit Config を開き、claude_desktop_config.json にこれを貼り付けます
{
"mcpServers": {
"overwing": {
"command": "npx",
"args": [
"-y",
"overwing-mcp"
],
"env": {
"OVERWING_API_KEY": "your-value",
"OVERWING_AGENT_KEY": "your-value",
"OVERWING_BASE_URL": "your-value"
}
}
}
}ターミナルでこれを実行します
claude mcp add --transport http overwing https://overwing.ai/mcp --header 'Authorization: your-value'または、パソコンで動かす
claude mcp add overwing --env OVERWING_API_KEY=your-value --env OVERWING_AGENT_KEY=your-value --env OVERWING_BASE_URL=your-value -- npx -y overwing-mcp動かす前に、空欄の値を入力してください。
または、設定を自分でコピーする
プロジェクトの .cursor/mcp.json、または ~/.cursor/mcp.json に貼り付けます
{
"mcpServers": {
"overwing": {
"url": "https://overwing.ai/mcp",
"headers": {
"Authorization": "your-value"
}
}
}
}または、パソコンで動かす
または、設定を自分でコピーする
プロジェクトの .cursor/mcp.json、または ~/.cursor/mcp.json に貼り付けます
{
"mcpServers": {
"overwing": {
"command": "npx",
"args": [
"-y",
"overwing-mcp"
],
"env": {
"OVERWING_API_KEY": "your-value",
"OVERWING_AGENT_KEY": "your-value",
"OVERWING_BASE_URL": "your-value"
}
}
}
}ブラウザがアプリを開いてよいか確認し、そのあとアプリが設定を表示して確認を求めます。キーの値は空欄のままなので、自分で入力してください。
または、設定を自分でコピーする
プロジェクトの .vscode/mcp.json に貼り付けます
{
"servers": {
"overwing": {
"type": "http",
"url": "https://overwing.ai/mcp",
"headers": {
"Authorization": "your-value"
}
}
}
}または、パソコンで動かす
または、設定を自分でコピーする
プロジェクトの .vscode/mcp.json に貼り付けます
{
"servers": {
"overwing": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"overwing-mcp"
],
"env": {
"OVERWING_API_KEY": "your-value",
"OVERWING_AGENT_KEY": "your-value",
"OVERWING_BASE_URL": "your-value"
}
}
}
}ブラウザがアプリを開いてよいか確認し、そのあとアプリが設定を表示して確認を求めます。キーの値は空欄のままなので、自分で入力してください。
このアドレスはリクエストの header にキーが必要ですが、コネクタを追加するメニューでは設定できません。代わりにClaude Code、Cursor、VS Codeを使ってください。
選んだアプリは、この端末のCookieに保存されます。アカウントは不要です。
自分で入力する値
- Overwing organization key (ow_live_...). Optional: evaluate and atlas_lookup work without it, 10 a day each. Needed for higher limits, rule sets and Tower setup. Get one at https://overwing.ai/login…
- Optional Overwing Tower agent key (ow_agent_...). Without it, the tower_create_agent tool mints one for the session
OVERWING_BASE_URLenvOverride the API base URL for self-hosted deployments (default https://overwing.ai)Authorization秘密 HTTP headerOptional. "Bearer ow_live_..." lifts the free limits; evaluate and atlas_lookup work without it, 10 a day each
この設定は、作者のレジストリ登録情報から作成しています。READMEと違う場合は、READMEに従ってください。
できること
ツールの名前と説明は、サーバーが2026年11月8日に返したそのままの内容です。書いたのは作者で、私たちではありません。プレーンテキストとして表示しています。
evaluateWorks with no API key: 10 evaluations a day, inputs up to 2,000 characters, and the text is not stored. With a key: 250 a day and up, inputs up to 100,000 characters, and your own rule sets. Score any text (typically an LLM's output) against an Overwing rule set. The text can be
evaluate_batchScore up to 50 texts against one rule set in a single call. Each item counts as one evaluation. Returns a summary plus per-item verdicts; items can fail independently. Needs an organization key.
list_rule_setsList the prebuilt and custom rule sets available to this organization. Needs an organization key.
get_rule_setFetch a rule set with its full rule definitions. Use 'content-safety' as a worked example when writing your own. Needs an organization key.
create_rule_setCreate a custom rule set with 1 to 25 rules. Each rule is a choice (pick one option), score (position on an ordered scale), or noul (yes/no) question with a fail condition, optional review threshold, weight, and action (block, redact, or review) that callers should take when it f
get_evaluationFetch a stored evaluation by id, including the original input and per-rule results. Needs an organization key.
list_evaluationsList recent evaluations, newest first, with optional verdict and rule set filters. Use next_cursor to page. Needs an organization key.
get_usageDaily usage, remaining quota for today, and plan limits. Needs an organization key.
whoamiIdentify the organization and plan behind the API key on this request, the key's scope, the organization's data settings (store_inputs, retention_days), and whether billing is set up. Needs an organization key.
create_accountCreate an Overwing account for yourself, with no email: an organization and an API key, returned once. Nothing is sent to anyone. Use it when you have no key and nobody has given you one; do not create a second account if you already have a key (whoami says which account a key be
prove_domainBegin proving that your account controls a domain. For an account with no email the domain stands in for one: it raises the limits to the normal free tier, opens full Beacon reports, and lets a lost key be replaced. Returns one value to publish at the domain, as a DNS TXT record
verify_domainLook for the value prove_domain asked for, at the domain. Found: the domain is the account's. Not found: an error saying what was looked for; DNS changes can take a few minutes, and calling again is safe. Needs an organization key.
list_plansPublic plan catalog: prices, daily limits, and per-minute burst limits. No API key needed.
atlas_lookupSay what a User-Agent string claims to be and whether the claim can be trusted, from the Overwing Atlas registry of AI crawlers, fetchers and browser agents. Returns the claimed agent, operator, purpose class, verification method (Web Bot Auth signature, user-agent string only, o
atlas_agentsBrowse or search Overwing Atlas, the registry of AI crawlers, fetchers and browser agents: name, operator, user-agent tokens, Web Bot Auth key directory, robots.txt behaviour, and (with Atlas Pro or Team) purpose class, verification, evasion flags and traffic shares. Filter by fr
atlas_register_agentAdd an agent or crawler you operate to the Overwing Atlas registry, free, so a site that looks up its User-Agent learns who runs it. Give the agent's name, the operator (the company or person running it), the operator's domain, and the token the User-Agent carries (the product na
atlas_verify_registrationLook for the proof at the operator's domain: the DNS TXT record or the file that atlas_register_agent asked for. Found: the entry is published in the registry, or held for a person when the operator name already belongs to someone. Not found: an error saying what was looked for;
atlas_list_registrationsThe agents this organization has registered in Overwing Atlas, newest first, each with its status (pending_verification, pending_review, published, rejected) and, while unverified, the value to publish at the domain. Needs an organization key.
atlas_withdraw_registrationWithdraw a registration. A published entry leaves the registry, so lookups stop naming the agent; an unfinished request is abandoned. This cannot be undone: register again to put it back. Needs an organization key.
atlas_summaryPublic numbers from Overwing Atlas: registry counts by purpose and verification, published browser-agent traffic shares, sector field-scan headlines (e.g. how many OSINT sites carry AI-crawler rules or any agent-payable surface), and the summary of the Agent Consumers report on w
beacon_startOverwing Beacon answers one question about a site: is this product reachable by agents? It checks robots.txt rules for AI agents, llms.txt, the sitemap, the MCP server card, hosted MCP endpoint and MCP Registry listing, the A2A agent card, OpenAPI discovery, and how the home page
beacon_reportThe report for a check started with beacon_start. The first call runs the check. With an Overwing key on the connection: score (0 to 100), verdict (yes, partly, no), the find / read / use answers, every check with what was found and a fix when it did not pass, and top_fixes ranke
feedbackSend one message to the people who run Overwing: a verdict you think was wrong, an error you could not act on, a payment you could not make, a tool that did not do what its description says. Read by a person, not answered automatically; give a contact if you want an answer. No ke
beacon_sampleA real Overwing Beacon report, free: the check of overwing.ai itself, refreshed daily. Read it to see exactly what a full report holds before starting a check. No key needed.
preflight_paymentCall this with the 402 you were given, before signing an x402 payment. It reads the terms (network, asset, amount, payee, timeout, resource) and answers allow, review or refuse with reasons: the terms against your policy, the payee against x402-list's directory (verified delivery
preflight_checkCall this before signing any Solana transaction. Overwing Preflight simulates the unsigned transaction against the chain as it is now, works out what it would take from the wallet you name (SOL, tokens, and control of its token accounts) and answers allow or refuse against your p
preflight_verdictThe public record of one Preflight verdict: its decision, reason codes, the programs the transaction would run, the transaction's digest, our signature, and every landed transaction reported against it. Never the wallet, the amounts or the transaction. No key needed.
preflight_reportAfter a transaction that Preflight allowed has landed, report its signature. Preflight reads the landed transaction from the chain and says whether the allow held: not_a_miss, or miss when it took more from the wallet than the policy permitted. The outcome is published. A covered
preflight_recordEverything Preflight has said and how it held up: counts of verdicts and misses, every miss with what the chain showed, the latest verdicts, and what the guarantee's reserve holds. Read it before deciding whether to rely on Preflight. No key needed.
preflight_overviewThe reference for Overwing Preflight: every policy field, every reason code a refusal can carry, what a verdict does and does not promise, the price, and the terms of the guarantee. Read it once before the first preflight_check. No key needed.
tower_load_templateSet up Overwing Tower for this organization by loading the starter workflow: email purchase order to order entry, with create_order, update_order and cancel_order against a mock IBM i system, and a starter policy. Idempotent. Needs an organization key. Returns a sample input you
tower_create_agentCreate a scoped agent identity and its key. Needs an organization key. Scope it to the operations the agent needs (for example create_order) rather than * where you can. The key (ow_agent_...) is returned once, in this result, and this server does not keep it: store it, and send
tower_list_agentsList this organization's Tower agents with scopes, status and last use. Keys are never returned. Needs an organization key.
tower_revoke_agentRevoke an agent. Its key stops working at once and cannot be restored. Needs an organization key.
tower_capabilitiesList the operations this agent is allowed to call, each with the JSON Schema its input must match and its compensating operation. Call this first; build inputs from the schema rather than guessing. Needs an agent key.
tower_decideAsk Tower how it would rule on an operation without doing anything: auto (would execute), review (a person must approve), or reject. Returns the score, the reason, and each policy question's answer. No side effects. Needs an agent key.
tower_submit_actionAsk Tower to perform an operation on the legacy system. Tower decides: executed (done), pending (a person must approve; poll tower_get_action, do not resubmit), or rejected (do not retry unchanged). Always send an idempotency_key that is stable for this business request, such as
tower_get_actionStatus and result of an action: pending, approved, executed, failed, compensated, or rejected. Use it to poll an action that is waiting on human review. Needs an agent key.
tower_compensateRun the compensating operation for an executed action, for example cancel the order that create_order made. Runs once; repeating it returns the first outcome. Needs an agent key.
tower_get_receiptFetch one signed receipt by id or by sequence number: the payload, its hash, the previous link's hash, and the Ed25519 signature. Needs an agent key.
tower_verify_receiptsRecompute every hash and check every signature over a range of this organization's receipts. Reports the first break, if any. Defaults to the whole chain (up to 5,000 links per call). Needs an agent key.
基本チェックi
誰でも見られる情報をチェックしており、コードは読んでいません。当てはまらない項目は数に入れていません。 · 2026年11月9日にチェック
- ✓ 合格
作者とリポジトリが一致
レジストリ名はドメイン overwing.ai で確認済みです(レジストリがドメインの所有を確認しています)
- … 未チェック
今もメンテナンスされている
このリポジトリはまだGitHubから読み取っていません
- … 未チェック
ライセンスがある
このリポジトリはまだGitHubから読み取っていません
- ✓ 合格
エンドポイントが応答する
262ミリ秒で応答(HTTP 200)
- – 対象外
OAuthでサインイン
サインイン不要
- ✓ 合格
ツールの説明に隠れた指示がない
41件のツールの説明を読み、AIへの隠れた指示は見つかりませんでした
検索・AIの記憶のほかのサーバーとの比較i
- このカテゴリのほかの5749件のうち0%よりスターが多い
- このカテゴリのほかの5749件のうち59%より最近レジストリに公開
- 基本チェックの結果が、このカテゴリのほかの5749件のうち100%と同じかそれ以上
レジストリでこのカテゴリにある有効なすべてのサーバーと比較しています。スターは、公開者がリポジトリの所有者である場合だけ数えます。
よくある質問
Overwingは安全ですか?
当てはまる基本チェック3項目のうち3項目に合格しました。コードは読んでいません。キーやトークンを追加する前に、サーバーがどんなアクセスを求めているかを確認し、必要な権限だけを持つキーを使ってください。
どんなキーやコードが必要ですか?
入力が必要なもの:OVERWING_API_KEY, OVERWING_AGENT_KEY, Authorization
パソコンで動きますか、それともクラウドですか?
どちらでも使えます。パソコンにインストールしても、クラウド版につないでもかまいません。
今もメンテナンスされていますか?
GitHubリポジトリの最終更新は–、最新バージョン(0.15.0)がレジストリに届いたのは26 日前です。
どのアプリで使えますか?
Claude Desktop、Claude Code、Cursor、VS CodeなどのMCP対応アプリで使えます。URLでつながるので、リモートMCPサーバーに対応したアプリ(開発者モードのChatGPTなど)でも使えます。
代わりに使えるものは?
同じカテゴリには:Context7, Codebase Memory, Exa
代わりの候補
同じカテゴリで、ほかの作者が作ったサーバー。スターの多い順。
カードの見方:↓ 週あたりのダウンロードi★ GitHubスターi3つの柱:レビュー · チェック · 利用状況i更新=最新バージョンをレジストリに公開した日i
Codebase Memory
Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.
Exa
Connect AI agents to Exa for web search, content fetching, and multi-step research.
作者のREADME
READMEを開く
作者のリポジトリから書かれたとおりにコピーしたものです。PopNx MCPは内容を確認・推奨していません。HTML、画像、スクリプトは取り除いています。リンクはこのサイトの外へ移動します。
Guardrails for LLM output, as MCP tools.
Score any text for safety and quality. Get pass / fail / review verdicts with calibrated confidence, usually in under 500 ms.
简体中文 · 日本語 · 한국어
Overwing is an API that checks what your model said before it ships. This package exposes it to any MCP-capable agent: Claude Desktop, Claude Code, Cursor, Windsurf, VS Code, OpenAI's Agents SDK, and anything else that speaks the Model Context Protocol.
- Real verdicts, not vibes. Every rule returns a typed answer, a probability, and a confidence.
failmeans a rule matched;reviewmeans it was unsure;passmeans neither. - Prebuilt `content-safety` and `outbound-message` rule sets: toxicity, personal data, confidential leaks, self-harm, sexual content, severity. Or write your own rules in plain language.
- Built for agents. Sign up, pay, evaluate, rotate keys, and cancel, all as JSON. No CAPTCHA, no browser required. See overwing.ai/llms.txt.
Try it without installing anything: paste text into the console at overwing.ai.
Hosted, no install
The same 41 tools are served at https://overwing.ai/mcp (Streamable HTTP). Point any client that takes a remote MCP URL at it:
claude mcp add --transport http overwing https://overwing.ai/mcp
No key is needed for evaluate, atlas_lookup, atlas_summary, list_plans, create_account, the three beacon_ tools, and preflight_verdict, preflight_report, preflight_record and preflight_overview. For the rest, send Authorization: Bearer ow_live_... on the connection.
Install
It works with no API key: evaluate and atlas_lookup each run 10 times a day free, and text sent without a key is not stored. For more, get a key at overwing.ai/login, or let your agent make its own account with the create_account tool. That account has no email: nothing is sent to anyone, and the key is returned once. The same call from a shell:
curl -X POST https://overwing.ai/api/v1/signup
Put the key in OVERWING_API_KEY. An account with no email starts at 50 evaluations a day; proving a domain (prove_domain, verify_domain) raises it to the normal free limits and makes a lost key recoverable.
Claude Code
claude mcp add overwing -e OVERWING_API_KEY=ow_live_... -- npx -y overwing-mcp
Claude Desktop, Cursor, Windsurf, VS Code (any JSON-configured client)
{
"mcpServers": { "overwing": { "command": "npx", "args": ["-y", "overwing-mcp"], "env": { "OVERWING_API_KEY": "ow_live_..." } } } }
Set OVERWING_BASE_URL to point at a self-hosted deployment. Requires Node 20+.
For Overwing Tower, set OVERWING_AGENT_KEY=ow_agent_... to operate as an existing agent. Without it, tower_create_agent mints a key with the organization key and keeps it in memory for the session.
evaluate, atlas_lookup, atlas_summary and list_plans work with no key at all.
For Overwing Preflight, preflight_check needs OVERWING_API_KEY; the other preflight_ tools need no key. No Preflight tool takes a private key, and none should ever be given one.
The text can be in any language. It was tested on 2026-09-29 in Spanish, Portuguese, French, German, Japanese, Simplified Chinese, Korean, Arabic and Hindi: a small test, not a benchmark. Results come back in English.
Tools
| Tool | What it does |
|---|---|
evaluate | Score one text against a rule set. Returns the verdict, a recommended_action (block, redact, review, or allow), aggregate score, confidence, latency, and per-rule results with each rule's action. Takes an optional context object (recipient, channel, ownership) that context-aware rule sets such as outbound-message read, and store: false to run the check without keeping the text. |
evaluate_batch | Score up to 50 texts in one call, with a summary and per-item verdicts and recommended actions. |
list_rule_sets · get_rule_set · create_rule_set | Browse the prebuilt set or define your own rules: yes/no questions, classifications, or scored scales. |
get_evaluation · list_evaluations | Read stored results, filter by verdict or rule set, page with a cursor. |
create_account · prove_domain · verify_domain | An account for the agent itself, with no email: create_account returns a key and sends nothing to anyone. Proving a domain (a DNS TXT record or a file) takes the place of the email: it raises the limits, opens full Beacon reports, and makes a lost key recoverable. |
atlas_lookup | Overwing Atlas: say what a User-Agent string claims to be and whether the claim can be trusted (Web Bot Auth, spoofable string, or unattributable). Works with no key, 10 a day; 100 a day with a free key. |
atlas_agents · atlas_summary | Search the registry of AI crawlers, fetchers and browser agents; get traffic shares, sector field-scan headlines, and the agent-spending summary. |
atlas_register_agent · atlas_verify_registration · atlas_list_registrations · atlas_withdraw_registration | Add an agent you operate to the Atlas registry, free, so a lookup of its User-Agent names you. Register it, publish the value you are given at the operator's domain (a DNS TXT record or a file), then verify. Needs OVERWING_API_KEY. |
beacon_start · beacon_report · beacon_sample | Overwing Beacon: is a site reachable by agents? Checks robots.txt, llms.txt, the MCP server card, endpoint and Registry listing, the A2A agent card and OpenAPI, and how the home page reads to a model. A check is free: beacon_start returns an id and beacon_report runs the check and returns the report. With OVERWING_API_KEY it is the full report, saved to your dashboard; with no key it is the summary (the score, the three answers and the first fix). beacon_sample is a real report in full. |
feedback | Tell the people who run Overwing what went wrong: a wrong verdict, an error you could not act on, a payment you could not make. No key needed. |
preflight_check | Overwing Preflight: should the agent sign this Solana transaction? Give it the unsigned transaction in base64 and a policy: the wallet to protect, the most SOL that may leave it (max_sol_out, fees included, or max_sol_out_lamports), and optionally max_token_out, min_token_in, allowed_programs and allow_delegation. It simulates the transaction and answers allow or refuse with every reason. Sign only on allow; a refusal, an error or anything else means do not sign. A verdict is valid for about two minutes. It never takes a private key. One check counts as one evaluation. Needs OVERWING_API_KEY. |
preflight_payment | Should the agent sign this x402 payment? Reads the 402's terms and checks the payee against x402-list's directory and the server's own manifest, and the price against the record. allow, review or refuse with reasons. No key needed. |
preflight_verdict · preflight_report · preflight_record · preflight_overview | Preflight's public side, with no key: read one published verdict, report the transaction that landed after a verdict (by its signature) to learn whether the verdict held, read the record of every verdict and miss, and read the overview of policy fields, reason codes, limits and prices. |
tower_load_template · tower_create_agent · tower_list_agents · tower_revoke_agent | Overwing Tower setup, with the organization key: load the starter order-entry workflow, then mint a scoped agent identity. The new agent key is used for the rest of the session. |
tower_capabilities · tower_decide · tower_submit_action · tower_get_action · tower_compensate | Overwing Tower operations, with the agent key: see which operations you may call and their input schemas, ask how a request would be ruled, submit it (executed, pending human review, or rejected), poll it, undo it. |
tower_get_receipt · tower_verify_receipts | Read a signed receipt, or recompute every hash and signature in the chain. |
get_usage · whoami · list_plans | Today's quota, the org behind the key, and the public plan catalog. |
The overwing://guide resource returns the full plain-text API guide.
Data handling and security
What this server does on your machine: it reads three environment variables (OVERWING_API_KEY, OVERWING_AGENT_KEY, OVERWING_BASE_URL) and makes HTTPS requests to the Overwing API. It reads no files, runs no commands, and talks to no other host. It has two runtime dependencies (@modelcontextprotocol/sdk, zod), and the whole server is one file: src/index.ts.
What happens to text you evaluate:
- It is sent to the Overwing API, and from there to TypeSafe, whose Jev model produces the verdict. It is not used to train models.
- Without a key the text is never stored.
- With a key the text, context and verdict are stored so you can read them back with
get_evaluation, until you delete them. To change that: - pass
store: falsetoevaluateorevaluate_batchto keep no text or context for that call; - set
store_inputs: falseon the organization (PATCH /api/v1/org, or the dashboard) to keep none at all; - set
retention_dayson the organization to delete evaluations after that many days. - A Solana transaction you check with
preflight_checkis sent to the Overwing API with its policy, and simulated against a Solana node. What is published is the verdict: its id, decision, reason codes, programs and the transaction's digest. Never the wallet, the amounts or the transaction. No private key is ever sent, and the tool does not accept one. - Traffic is encrypted in transit (TLS) and the database is encrypted at rest (AES-256).
Give your agent a restricted key. A key with scope evaluate can run checks and read rule sets, and cannot read stored evaluations, change rule sets, manage keys, webhooks or billing, or delete anything:
curl -X POST https://overwing.ai/api/v1/api-keys \
-H "Authorization: Bearer ow_live_..." -H "Content-Type: application/json" \ -d '{"name":"agent","scope":"evaluate"}'
Overwing is a young service and has not had an independent security audit. The full picture, including subprocessors and what we have not done, is at overwing.ai/security; the privacy policy is at overwing.ai/privacy. To report a vulnerability, see SECURITY.md.
Example
Ask your agent:
Check this reply before I send it: "Reach me at dana@example.com or 555-0142 to sort out the refund."
It calls evaluate and gets back:
Verdict: FAIL score=0.82 confidence=0.97 251ms
toxicity: pass (answer="safe", confidence=1) pii_detected: fail (answer=true, confid
冒頭だけを表示しています。続きはリポジトリで読めます