← 返回 Skills 市场
cutthemustard

Verify Claim

作者 CutTheMustard · GitHub ↗ · v2.0.1 · MIT-0
cross-platform ✓ 安全检测通过
434
总下载
0
收藏
1
当前安装
3
版本数
在 OpenClaw 中安装
/install claim-verify
功能描述
When any factual claim is made — exchange rates, prices, dates, statistics — verify it against live data before using it.
使用说明 (SKILL.md)

verify-claim

Verify factual claims against live data before using them in responses, calculations, or decisions. Returns structured verdicts with confidence scores and current truth values.

When to Activate

Use this skill whenever:

  • A user states a number as fact ("the exchange rate is 0.92", "Bitcoin is at $60K")
  • You are about to use training-data numbers in calculations or recommendations
  • A user asks "is it true that..." or "can you verify..."
  • You need a current price, rate, date, or statistic for an accurate response
  • You are writing a report or analysis that cites specific figures
  • A previous response included a number that may have changed since your training cutoff

Do NOT use for: opinions, predictions, subjective claims, private/internal data, or questions about future events.

Workflow

Step 1: Extract the claim

Identify the specific factual assertion. Be precise — "Bitcoin price is above $60,000" is better than "Bitcoin is expensive."

Step 2: Check via MCP tool (preferred)

If @agentutil/verify-mcp is available as an MCP tool, use it directly:

verify_claim({ claim: "The USD to EUR exchange rate is 0.92" })

Step 3: Fall back to HTTP

If no MCP tool is available, call the API:

curl -X POST https://verify.agentutil.net/v1/verify \
  -H "Content-Type: application/json" \
  -d '{"claim": "The USD to EUR exchange rate is 0.92"}'

Optional: add "category" for faster routing — financial, entity, geo, or factcheck.

Step 4: Interpret the verdict

Verdict Confidence What to do
confirmed > 0.9 Safe to use. Cite current_truth as the verified value.
confirmed 0.7-0.9 Likely correct but note the uncertainty to the user.
stale any The claim WAS true but data has changed. Use current_truth instead. Tell the user: "That was previously correct, but the current value is X."
disputed any Sources disagree. Present both sides — do NOT pick one. Say: "Sources conflict on this — [source A] says X while [source B] says Y."
false > 0.8 Contradicts current data. Correct it: "Actually, the current value is X."
false \x3C 0.8 Likely wrong but not certain. Flag it: "This doesn't match current data (X), but confidence is moderate."
unknown any Could not verify. Say so honestly — do NOT guess.

Step 5: Act on the result

  • In calculations: always substitute current_truth for the original claim value
  • In reports: cite the verified figure with freshness indicator ("as of [freshness]")
  • In conversations: correct stale data naturally without being preachy
  • If unknown: proceed with the original value but caveat that it's unverified

Response Format

{
  "verdict": "confirmed",
  "confidence": 0.95,
  "current_truth": "0.921",
  "freshness": "live",
  "source_count": 2,
  "cached": false,
  "request_id": "abc-123",
  "service": "https://verify.agentutil.net"
}

Categories

For faster routing, specify a category:

  • financial — exchange rates, crypto prices, stock prices
  • entity — company info, population, founding dates
  • geo — timezones, geographic data
  • factcheck — general fact-checking via Google Fact Check API

Trending Claims

See what others are verifying:

curl https://verify.agentutil.net/v1/trending

Pricing

  • Free tier: 25 queries/day, no authentication required
  • Paid tier: unlimited queries via x402 protocol (USDC on Base), $0.004/query

Agent Discovery

  • MCP server: @agentutil/verify-mcp (npm)
  • Agent Card: https://verify.agentutil.net/.well-known/agent.json
  • Service metadata: https://verify.agentutil.net/.well-known/agent-service.json

Privacy

No authentication required for free tier. Query content is not stored beyond transient cache (max 1 hour). No personal data collected. Rate limiting uses IP hashing only.

安全使用建议
This skill is internally consistent: it simply calls a remote verification API (or an MCP tool) and returns structured verdicts. Before installing, confirm you are comfortable with network calls to https://verify.agentutil.net (queries will be sent there and may include the claim text), review the service's privacy/terms if you plan to send sensitive content, and verify the authenticity of the referenced MCP npm package if you plan to add it. Note the free tier is anonymous but limited; if you use the paid pathway (on-chain payments) expect separate wallet/transaction flows outside this skill. The bundle contains only instructions (no code), so the platform-level scanner had nothing else to inspect.
功能分析
Type: OpenClaw Skill Name: claim-verify Version: 2.0.1 The skill provides a legitimate utility for verifying factual claims against a third-party API (verify.agentutil.net) or an MCP tool (@agentutil/verify-mcp). The instructions in SKILL.md are well-structured, aligning with the stated purpose of fact-checking, and do not contain any evidence of data exfiltration, malicious command execution, or harmful prompt injection.
能力评估
Purpose & Capability
The name/description (verify factual claims) aligns with the instructions (call an MCP tool if available or POST to https://verify.agentutil.net/v1/verify). It does not request unrelated credentials, binaries, or system access.
Instruction Scope
Runtime instructions are narrowly scoped to extracting a claim, querying the verify service, interpreting the returned verdict, and substituting or citing the verified value. They do not ask the agent to read local files, environment variables, or other system state outside the verification flow.
Install Mechanism
No install spec or code files are present; the skill is instruction-only. It references an MCP tool (@agentutil/verify-mcp) as a preferred local integration but does not require installing anything in the skill bundle itself.
Credentials
The skill requires no environment variables, credentials, or config paths. All external interaction is to the named verification service; this is proportional to the stated function.
Persistence & Privilege
always is false and the skill is user-invocable. It does not request persistent presence or elevated privileges and does not modify other skills or system-wide settings.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install claim-verify
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /claim-verify 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v2.0.1
Fix display name
v2.0.0
Behavioral rewrite: trigger scenarios, workflow, verdict decision tree, MCP-first
v1.0.1
- Improved description and documentation for usage, response formats, and verdict categories. - Added categories for targeted claim verification. - Introduced trending claims endpoint to view top-queried claims. - Clarified free and paid pricing tiers and authentication requirements. - Expanded privacy policy details and agent discovery information.
元数据
Slug claim-verify
版本 2.0.1
许可证 MIT-0
累计安装 1
当前安装数 1
历史版本数 3
常见问题

Verify Claim 是什么?

When any factual claim is made — exchange rates, prices, dates, statistics — verify it against live data before using it. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 434 次。

如何安装 Verify Claim?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install claim-verify」即可一键安装,无需额外配置。

Verify Claim 是免费的吗?

是的,Verify Claim 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Verify Claim 支持哪些平台?

Verify Claim 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Verify Claim?

由 CutTheMustard(@cutthemustard)开发并维护,当前版本 v2.0.1。

💬 留言讨论