Back to skill

Security audit

Fraud Filter

Security checks for vulnerabilities and agentic risk

Overview

The skill’s payment-reputation purpose is coherent, but its local dashboard and sync code expose high-impact configuration, reporting, and code-execution risks that should be reviewed before installation.

Review carefully before installing. The skill may be useful for payment endpoint reputation checks, but only run the dashboard in a trusted local environment, avoid custom trust_db_url values until the injection bug is fixed, do not rely on the dashboard for offline mode, and inspect queued reports before enabling network participation or flushing them.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (4)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/sync-trust-db.sh:62
Finding

Arbitrary JavaScript Execution Through Trust Database URL Interpolation

Content
View full analysis
&2; exit 1 ;; esac done if [ -n "$CUSTOM_URL" ]; then CDN_URL="$CUSTOM_URL" elif [ -f "$CONFIG_PATH" ]; then CDN_URL=$(node --input-type=module -e " import { readFileSync } from 'node:fs'; try { const c = JSON.parse(readFileSync('${CONFIG_PATH}', 'utf-8')); console.log(c.trust_db_url || 'https://api.fraud-filter.com/trust.json'); } catch { console.log('https://api.fraud-filter.com/trust.json'); } ") else CDN_URL="https://api.fraud-filter.com/trust.json" fi ``` ```bash if node --input-type=module -e " import { writeFileSync } from 'node:fs'; const res = await fetch('${CDN_URL}', { headers: { 'User-Agent': 'fraud-filter-skill/1.0' }, signal: AbortSignal.timeout(15_000), }); if (!res.ok) throw new Error('HTTP ' + res.status); const data = await res.json(); if (!data.endpoints) throw new Error('missing endpoints field'); writeFileSync('${TEMP_PATH}', JSON.stringify(data)); console.log('Valid: ' + Object.keys(data.endpoints).length + ' endpoints'); " 2>&1; then ``` ### Technical Analysis `CDN_URL` is populated either from the command-line `--url` argument or from the configurable `trust_db_url` field. It is subsequently interpolated directly into a JavaScript program passed to `node -e`. Shell quoting does not make this safe because the interpolation occurs while constructing JavaScript source. A value containing a single quote can terminate the `fetch()` string literal and inject arbitrary JavaScript statements. Node.js exposes filesystem, process, and child-process APIs, so successful injection can perform actions ...[truncated 1538 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
server/server.js:201
Finding

Unauthenticated Local API Allows Cross-Site Configuration Changes and Report Exfiltration

Content
View full analysis
{ const chunks = []; req.on("data", (c) => chunks.push(c)); req.on("end", () => resolve(Buffer.concat(chunks).toString())); req.on("error", reject); }); } ``` ```js if (req.method === "POST") { const body = await readBody(req); let data; try { data = JSON.parse(body); } catch { json(res, { error: "invalid JSON" }, 400); return; } ``` Sensitive operations are directly exposed: ```js if (pathname === "/api/reports/flush") { const result = await submitPendingReports(); json(res, result); return; } ``` ```js if (pathname === "/api/config") { const config = loadConfig(); const allowedKeys = [ "trust_db_url", "report_endpoint", "sync_interval_hours", "participate_in_network", "on_block", "on_caution", ]; for (const key of allowedKeys) { if (key in data) { config[key] = data[key]; } } saveConfig(config); const { install_id, ...safeConfig } = config; json(res, safeConfig); return; } ``` The configurable endpoint is then trusted during report submission: ```js if (!config.participate_in_network) { return { submitted: 0, failed: 0, skipped: 0, reason: "network_participation_disabled" }; } ``` ```js const response = await fetch(config.report_endpoint, { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(body), }); ``` ### Technical Analysis Binding to `127.0.0.1` limits direct network exposure but does not authenticate browser-originated reques ...[truncated 2684 chars]
Remediation
View remediation
` and `localhost:`. - Add `Cache-Control: no-store` to report and configuration responses. - Restrict `report_endpoint` to HTTPS and require explicit confirmation before allowing a non-default host. - Separate configuration changes from report transmission and require a fresh confirmation or token for every flush. - Consider using a Unix domain socket or authenticated local IPC instead of an unauthenticated HTTP management API. - Add request-body size limits and rate limits. - Do not rely solely on lack of CORS headers or browser Private Network Access behavior. ]]>

T09 · Insecure Skill Coding Practices

Error
Location
server/index.html:379
Finding

Stored DOM XSS Through Unescaped Remote Trust Database Fields

Content
View full analysis
No endpoints found'; return; } el.innerHTML = `

Showing ${endpoints.length} of ${total} endpoints

${endpoints.map(e => ` `).join('')}
ScoreHostnameReportsSuccessMedian $Price RangeLast FailureWarnings
${scoreBadge(e.score)}${esc(e.url_hint || '?')}${e.report_count || 0}${e.success_rate != null ? (e.success_rate * 100).toFixed(1) + '%' : '-'}$${(e.median_price_usd || 0).toFixed(4)}$${(e.price_p10_usd || 0).toFixed(4)} – $${(e.price_p90_usd || 0).toFixed(4)}${e.last_failure || '-'}${(e.warnings || []).map(w => `${esc(w)}`).join('')}
`; } ``` The score helper places its input directly into markup: ```js function scoreBadge(score) { if (score == null) return '?'; const cls = score >= 70 ? 'score-high' : score >= 40 ? 'score-mid' : 'score-low'; return `${score}`; } ``` ### Technical Analy ...[truncated 2284 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
server/server.js:295
Finding

Standalone Dashboard Ignores the Hotlist Synchronization Opt-Out

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (61)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Running a local HTTP dashboard with configuration and flush endpoints expands the attack surface beyond the advertised trust-checking function. If those APIs can modify enforcement settings or trigger report submission, local attackers or other processes may influence payment policy or exfiltrate queued data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

Running a local HTTP dashboard with configuration and flush endpoints expands the attack surface beyond the advertised trust-checking function. If those APIs can modify enforcement settings or trigger report submission, local attackers or other processes may influence payment policy or exfiltrate queued data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Running a local HTTP dashboard with configuration and flush endpoints expands the attack surface beyond the advertised trust-checking function. If those APIs can modify enforcement settings or trigger report submission, local attackers or other processes may influence payment policy or exfiltrate queued data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Running a local HTTP dashboard with configuration and flush endpoints expands the attack surface beyond the advertised trust-checking function. If those APIs can modify enforcement settings or trigger report submission, local attackers or other processes may influence payment policy or exfiltrate queued data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Running a local HTTP dashboard with configuration and flush endpoints expands the attack surface beyond the advertised trust-checking function. If those APIs can modify enforcement settings or trigger report submission, local attackers or other processes may influence payment policy or exfiltrate queued data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

Running a local HTTP dashboard with configuration and flush endpoints expands the attack surface beyond the advertised trust-checking function. If those APIs can modify enforcement settings or trigger report submission, local attackers or other processes may influence payment policy or exfiltrate queued data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

Running a local HTTP dashboard with configuration and flush endpoints expands the attack surface beyond the advertised trust-checking function. If those APIs can modify enforcement settings or trigger report submission, local attackers or other processes may influence payment policy or exfiltrate queued data.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

Running a local HTTP dashboard with configuration and flush endpoints expands the attack surface beyond the advertised trust-checking function. If those APIs can modify enforcement settings or trigger report submission, local attackers or other processes may influence payment policy or exfiltrate queued data.

Content

No source excerpt is available for this finding.

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · server/index.html (reported line 178)May include surrounding context.

html
<div class="tab" data-tab="settings">Settings</div>
</div>

<!-- Endpoints Tab -->
<div id="tab-endpoints" class="tab-content active">
  <div class="search-bar">
    <input type="text" id="searchQuery" placeholder="Search by hostname (e.g. stockdata.xyz)" onkeyup="if(event.key==='Enter')searchDb()">

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · test/test.js (reported line 222)May include surrounding context.

js
assert.ok(w.includes("limited_data"));
  });

  it("returns no warnings for healthy endpoint", () => {
    const w = deriveWarnings({
      success_rate: 0.95, report_count: 100,
      price_p10_usd: 0.01, price_p90_usd: 0.02,

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
95% confidence
Finding

The skill declares no explicit tool scope or permissions while its documented behavior clearly relies on shell execution, network access, and local file/process management. That mismatch prevents users and policy systems from understanding or constraining the skill’s capabilities, increasing the chance of unintended outbound access, execution, or data handling.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The documentation is internally inconsistent about whether failures are only queued locally or automatically reported back to the network. In a skill that handles payment outcomes, this ambiguity affects user consent, privacy expectations, and compliance decisions.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Conflicting statements about local-only queuing versus automatic network reporting create a deceptive or at least unreliable disclosure surface. Users cannot accurately judge what transaction outcome data may leave the system.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

Automatic queuing of transaction outcome data, even locally, is a privacy-relevant behavior that should be disclosed prominently before use. In the payment context, endpoint URLs, prices, and failure reasons may reveal sensitive operational or business information.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · TECHNICAL.md (reported line 57)May include surrounding context.

md
For negative outcomes, the plugin always queues a local signal automatically:

- **post_payment_failure**: Queued locally without confirmation — paid and received nothing is unambiguous. Submission to the network requires explicit opt-in.
- **pre_payment_failure**: Queued locally. Technical false positives (timeouts, misconfigured agents) are filtered downstream by deduplication and the 3-reporter hotlist threshold.

With `participate_in_network: false` (the default), failure reports accumulate in `data/pending-reports.jsonl` for local review and are never sent anywhere. The user decides when and whether to submit.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/signal-format.md (reported line 67)May include surrounding context.

Example:

text
Input:  https://API.StockData.xyz/report/AAPL?key=abc
Normal: https://api.stockdata.xyz/report/AAPL
Hash:   sha256:<hex>

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

This line documents posting anonymous outcome signals to an external service endpoint, which is a real external transmission path. Given the skill metadata says network reporting is opt-in, documenting unconditional submission increases the risk of unauthorized outbound communication, privacy leakage, and implementation drift toward silent telemetry.

Content

Scanner excerpt · references/signal-format.md (reported line 80)May include surrounding context.

Anonymous outcome signal

Sent to POST https://api.fraud-filter.com/reports:

json
{

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The documentation explicitly says anonymous outcome signals are sent to a remote endpoint, which conflicts with the stated skill behavior that reporting is opt-in and locally queued by default. This kind of documentation/behavior mismatch is security-relevant because implementers may wire in automatic outbound reporting, causing unexpected data egress and privacy violations even if the payload is pseudonymous.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · README.md (reported line 47)May include surrounding context.

md
# Outcomes: success, post_payment_failure, pre_payment_failure
#
# Examples:
#   report.sh https://api.example.com/data post_payment_failure 0.05 --reason "Needed AAPL price. Got HTTP 200 with empty data array."
#   report.sh https://api.example.com/data post_payment_failure 0.05 --skill stock-research --reason "Paid for structured stock data. Received garbled JSON."
#   report.sh https://api.example.com/data success 0.03 --reason "Needed current AAPL price. Advertised real-time quotes. Got price, volume, and 52-week range in clean JSON."

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 114)May include surrounding context.

md
# Outcomes: success, post_payment_failure, pre_payment_failure
#
# Examples:
#   report.sh https://api.example.com/data post_payment_failure 0.05 --reason "Needed AAPL price. Got HTTP 200 with empty data array."
#   report.sh https://api.example.com/data post_payment_failure 0.05 --skill stock-research --reason "Paid for structured stock data. Received garbled JSON."
#   report.sh https://api.example.com/data success 0.03 --reason "Needed current AAPL price. Advertised real-time quotes. Got price, volume, and 52-week range in clean JSON."

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 115)May include surrounding context.

md
# Outcomes: success, post_payment_failure, pre_payment_failure
#
# Examples:
#   report.sh https://api.example.com/data post_payment_failure 0.05 --reason "Needed AAPL price. Got HTTP 200 with empty data array."
#   report.sh https://api.example.com/data post_payment_failure 0.05 --skill stock-research --reason "Paid for structured stock data. Received garbled JSON."
#   report.sh https://api.example.com/data success 0.03 --reason "Needed current AAPL price. Advertised real-time quotes. Got price, volume, and 52-week range in clean JSON."

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · scripts/report.sh (reported line 10)May include surrounding context.

sh
# Outcomes: success, post_payment_failure, pre_payment_failure
#
# Examples:
#   report.sh https://api.example.com/data post_payment_failure 0.05 --reason "Needed AAPL price. Got HTTP 200 with empty data array."
#   report.sh https://api.example.com/data post_payment_failure 0.05 --skill stock-research --reason "Paid for structured stock data. Received garbled JSON."
#   report.sh https://api.example.com/data success 0.03 --reason "Needed current AAPL price. Advertised real-time quotes. Got price, volume, and 52-week range in clean JSON."

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · scripts/report.sh (reported line 11)May include surrounding context.

sh
# Outcomes: success, post_payment_failure, pre_payment_failure
#
# Examples:
#   report.sh https://api.example.com/data post_payment_failure 0.05 --reason "Needed AAPL price. Got HTTP 200 with empty data array."
#   report.sh https://api.example.com/data post_payment_failure 0.05 --skill stock-research --reason "Paid for structured stock data. Received garbled JSON."
#   report.sh https://api.example.com/data success 0.03 --reason "Needed current AAPL price. Advertised real-time quotes. Got price, volume, and 52-week range in clean JSON."

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · scripts/report.sh (reported line 12)May include surrounding context.

sh
# Outcomes: success, post_payment_failure, pre_payment_failure
#
# Examples:
#   report.sh https://api.example.com/data post_payment_failure 0.05 --reason "Needed AAPL price. Got HTTP 200 with empty data array."
#   report.sh https://api.example.com/data post_payment_failure 0.05 --skill stock-research --reason "Paid for structured stock data. Received garbled JSON."
#   report.sh https://api.example.com/data success 0.03 --reason "Needed current AAPL price. Advertised real-time quotes. Got price, volume, and 52-week range in clean JSON."

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · server/index.html (reported line 194)May include surrounding context.

html
# Outcomes: success, post_payment_failure, pre_payment_failure
#
# Examples:
#   report.sh https://api.example.com/data post_payment_failure 0.05 --reason "Needed AAPL price. Got HTTP 200 with empty data array."
#   report.sh https://api.example.com/data post_payment_failure 0.05 --skill stock-research --reason "Paid for structured stock data. Received garbled JSON."
#   report.sh https://api.example.com/data success 0.03 --reason "Needed current AAPL price. Advertised real-time quotes. Got price, volume, and 52-week range in clean JSON."