Back to skill

Security audit

SAVER Verifier

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed paid fact-checking integration that sends claims to SAVER Verify and includes payment and retry guardrails.

Install only if you are comfortable sending the claim and any optional context to SAVER Verify and paying 0.20 USDC per verification under your wallet policy. Avoid using it on confidential, regulated, or secret material unless that external service is approved for that data.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (3)

External Transmission

Medium
Category
Data Exfiltration
Confidence
92% confidence
Finding

The skill instructs the agent to send user-provided claims and optional context to an external third-party API, which is a real data exfiltration surface. Although this is the intended function of a verification service, the skill does not include strong data-minimization or sensitivity-gating requirements before transmission, so confidential prompts, internal research, or regulated data could be disclosed externally.

Content

Scanner excerpt · SKILL.md (reported line 15)May include surrounding context.

md
Inspect https://saververify.com/agent-tool.json for the input/output schemas and canonical transport. Inspect https://saververify.com/examples.json before buying. These are three actual operator-funded paid reports, not organic customer reviews or an independent accuracy benchmark. Reading them is free.

Use `POST https://api.saververify.com/v2/verify` for evidence-to-source links. Keep `POST /verify` for integrations requiring the original string-evidence format. Inspect https://api.saververify.com/openapi.json and https://api.saververify.com/.well-known/x402 for the current contract.

Send `Content-Type: application/json`. Send only `claim` and optional `context` as top-level fields, without transport fields or a `body` wrapper. Keep claim non-empty and <=8000 UTF-16 code units, context <=12000, and total JSON body <=128 KiB. Treat `context: null` as absent.

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

This section not only sends data to an external API but also instructs the runtime to use an authorized wallet to satisfy a 402 payment challenge to a specific recipient. Even with some guardrails, this creates a monetized external action path where untrusted skill content can trigger paid transactions and external disclosure, increasing the blast radius if invoked on attacker-supplied or sensitive content.

Content

Scanner excerpt · SKILL.md (reported line 19)May include surrounding context.

md
Send `Content-Type: application/json`. Send only `claim` and optional `context` as top-level fields, without transport fields or a `body` wrapper. Keep claim non-empty and <=8000 UTF-16 code units, context <=12000, and total JSON body <=128 KiB. Treat `context: null` as absent.

On HTTP 400/413, correct the request using `code` and `help` before paying. Use `help.endpoint` to retain the intended response version. GET/HEAD on verification routes returns 405; use POST for verification, or GET https://api.saververify.com/healthz for free liveness checks.

On HTTP 402, decode the base64 JSON in `PAYMENT-REQUIRED`; the body may be `{}`. Use the runtime's existing authorized x402 v2 wallet with the exact scheme, Base Mainnet (`eip155:8453`) and USDC. Require 0.20 USDC (`amount: "200000"`), token `0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913` and recipient `0xdfd6cb683F5288e1dD1337E84C09b490857CC110`. Reject unexpected payment terms or destinations.

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The recovery flow introduces another external transmission channel by requiring a persistent recovery token to be sent to the remote service and reused to fetch results. While the token is not a wallet key, it is still a correlator for requests and retained results; combined with claim/context submission, this increases privacy and tracking risk and extends the lifetime of externally stored data.

Content

Scanner excerpt · SKILL.md (reported line 27)May include surrounding context.

md
Retry the same request with `PAYMENT-SIGNATURE` using the runtime's x402 client. Read the result and `PAYMENT-RESPONSE` on success; require a successful Base settlement receipt with a transaction hash. Do not automatically sign another payment after a lost response, timeout, invalid receipt or uncertain outcome.

For recovery, generate and persist a random 32-byte hexadecimal secret before the first POST. Send it in `SAVER-RECOVERY-TOKEN` on both the initial POST and its signed retry. Reuse it only for that claim/context and response version. GET https://api.saververify.com/results with the same header to retrieve a retained settled result without another payment. Never use a wallet key as the token or put it in a URL. Retention is 24 hours. Inspect `resultRecovery.durable` in discovery before relying on restart persistence.

Treat recovery HTTP 202 as pending/uncertain, 503 as storage unavailable, and 404 as no retained result, not proof of nonpayment. Do not automatically repay. A 409 on POST indicates a conflicting authorization, wallet budget or recovery token; resolve it before creating another payment. Follow Retry-After for 429/503 without automatically signing a replacement.