Back to skill

Security audit

RSS个性化追踪摘要

Security checks for vulnerabilities and agentic risk

Overview

This RSS personalization skill is coherent in purpose, but it exposes click-tracking data and controls through an unauthenticated public tunnel.

Review carefully before installing. This skill should not be run as-is on a real host or with real users unless the public tunnel is opt-in, admin routes are private or authenticated, raw IP and URL logging is minimized and protected, redirect destinations are validated, SSH host verification is restored, and configuration updates are made explicit and reversible.

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
rss-redirect-server.js:29
Finding

Unauthenticated Public Disclosure and Deletion of Click-Tracking Data

Content
View full analysis
${originalUrl}\n`); res.writeHead(302, { 'Location': originalUrl, 'Cache-Control': 'no-store, no-cache, must-revalidate', 'Pragma': 'no-cache' }); res.end(); return; } if (pathname === '/health') { res.writeHead(200, { 'Content-Type': 'text/plain' }); res.end('OK\nTunnel: ' + TUNNEL_URL); return; } if (pathname === '/clicks') { res.writeHead(200, { 'Content-Type': 'text/plain', 'Access-Control-Allow-Origin': '*' }); try { res.end(fs.readFileSync(LOG_FILE, 'utf8')); } catch (e) { res.end(''); } return; } if (pathname === '/clear') { fs.writeFileSync(LOG_FILE, ''); res.writeHead(200, { 'Content-Type': 'text/plain' }); res.end('Cleared'); return; } if (pathname === '/newclicks') { // Return clicks recorded after the supplied timestamp. const since = parsedUrl.query.since || '0'; const sinceTime = new Date(parseInt(since)); try { const data = fs.readFileSync(LOG_FILE, 'utf8'); const lines = data.trim().split('\n').filter(l => { const ts = l.match(/^\[(\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2})\]/); return ts && new Date(ts[1]) > sinceTime; }); res.end(lines.join('\n')); } catch (e) { res.end(''); } return; } ``` ```javascript server.listen(PORT, '0.0.0.0', () => { console.log(`Server on port ${PORT}`); startTunnel(); }); ``` ### Technical Analysis The service records each visitor's source IP address and decoded destination URL in `/tmp/rss-clicks.log`. The `/clicks` and `/newclicks` endpoints disclose that information without authentication, while `/clear` permits unauthenticated deletion. The server ...[truncated 1529 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
rss-redirect-server.js:17
Finding

Unvalidated Redirect Destination Enables Open Redirects and Denial of Service

Content
View full analysis
${originalUrl}\n`); res.writeHead(302, { 'Location': originalUrl, 'Cache-Control': 'no-store, no-cache, must-revalidate', 'Pragma': 'no-cache' }); res.end(); return; } ``` ### Technical Analysis Base64 is reversible encoding and provides no authenticity or integrity. Any remote user can generate a `/go/` path containing an arbitrary decoded value. The service inserts that value directly into the HTTP `Location` header without parsing the URL, restricting its scheme, or checking it against an expected destination set. This makes the public service an open redirect that can be used to disguise phishing destinations behind the Skill's tunnel domain. Depending on client behavior, unsafe or unexpected schemes may also be accepted. Malformed decoded values, including values unsuitable for an HTTP header, can cause Node.js header validation to throw an exception at `res.writeHead`. That operation is outside the decoding `try/catch`. An uncaught exception can terminate the Node.js process, resulting in denial of service. ### Attack Path 1. The attacker selects a malicious or phishing destination. 2. The attacker Base64URL-encodes that destination. 3. The attacker creates a link in the form `https:///go/ ...[truncated 788 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
start.sh:33
Finding

Remote Tunnel Output Is Interpolated into Executable Python Source

Content
View full analysis
/tmp/serveo.log 2>&1 & sleep 6 if grep -q "Forwarding HTTP" /tmp/serveo.log; then TUNNEL_URL=$(grep "Forwarding HTTP" /tmp/serveo.log | awk '{print $NF}') echo "✓ Tunnel started: $TUNNEL_URL" # Update configuration. python3 -c " import json with open('/root/.openclaw/workspace/rss-weights.json') as f: w = json.load(f) w['redirect_base'] = '$TUNNEL_URL' with open('/root/.openclaw/workspace/rss-weights.json', 'w') as f: json.dump(w, f, ensure_ascii=False, indent=2) " echo "✓ Configuration updated" else echo "❌ Tunnel failed to start" cat /tmp/serveo.log exit 1 fi ``` ### Technical Analysis `TUNNEL_URL` is extracted from output produced by an external SSH service and inserted directly into a double-quoted `python3 -c` program. The value is placed inside a Python single-quoted string without escaping. If the extracted value contains a single quote followed by valid Python syntax, it can terminate the intended string and inject additional Python statements. Because the entire value is treated as source code rather than data, a malicious tunnel endpoint can potentially execute arbitrary Python with the privileges of the user running `start.sh`. The SSH invocation also disables strict host-key checking. This weakens first-connection authenticity and increases the possibility of trusting an unintended endpoint or attacker-controlled key. The code does not validate that the extracted value is a correctly formed HTTPS URL under an expected hostname. ### Attack Path 1. An attacker controls or successfully impersonates the external tunnel service, or otherwise causes malicious content to appear in `/ ...[truncated 1087 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
rss-redirect-server.js:81
Finding

Unverified Third-Party Reverse Tunnel Exposes Excessive Local Functionality

Content
View full analysis
{ const out = d.toString(); fs.appendFileSync(LOG_TUNNEL, out); process.stdout.write(out); // Extract the URL. const m = out.match(/Forwarding HTTP traffic from (https:\/\/[^\s]+)/); if (m) { fs.writeFileSync('/tmp/tunnel_url.txt', m[1]); } }); tunnel.stderr.on('data', (d) => fs.appendFileSync(LOG_TUNNEL, d.toString())); tunnel.on('exit', (code) => { console.log(`Tunnel exited with code ${code}, restarting in 5s...`); setTimeout(startTunnel, 5000); }); fs.writeFileSync(PID_FILE, tunnel.pid.toString()); return tunnel; } server.listen(PORT, '0.0.0.0', () => { console.log(`Server on port ${PORT}`); startTunnel(); }); ``` ### Technical Analysis The process automatically creates a reverse SSH tunnel to a third-party service while disabling strict host-key checking. It then restarts that tunnel indefinitely after every exit. The tunnel forwards the complete HTTP server, not only the intended redirect route. Public ingress is needed for the declared click-redirect feature. However, forwarding health, log-reading, log-deletion, and incremental-log endpoints is unnecessary. This exceeds the minimum network exposure required by the Skill. The tunnel provider is also positioned to observe connection metadata and redirect requests. Since original destinations are encoded only with Base64URL, they are readily recoverable and should not be treated as confidential. The static pre-scan warn ...[truncated 1508 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
Findings (19)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill explicitly describes recording user clicks and adjusting future content based on those clicks, but it provides no user-facing notice, consent flow, or retention policy. Because the system also logs identifiable metadata such as source IP and target URL, this creates a meaningful privacy and compliance risk rather than a purely theoretical concern.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The design stores timestamp, source IP, and destination URL in a plaintext file under /tmp, which is commonly accessible to local processes and may persist long enough for accidental disclosure. This creates a concrete privacy and data-leak risk because the log directly links a user's network identity to reading activity.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill documentation highlights a readable click log as an operational artifact, encouraging routine access to detailed user activity data. In context, this increases the likelihood of oversharing, weak permissions, and long-term retention of sensitive browsing behavior.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The specified log format explicitly captures identifiable user activity details, including time, source IP, and target content URL. This is dangerous because it creates a durable mapping of who accessed what and when, which can be sensitive personal data and attractive for misuse or unauthorized access.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code automatically opens a reverse SSH tunnel to serveo.net, exposing the local HTTP service to the public internet without authentication or operator approval. In this skill's context, that materially increases risk because sensitive click logs and destructive endpoints become remotely reachable, turning a local utility into an externally accessible service.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The system is intentionally designed to collect click-derived behavioral data and use it to profile user interests for future summaries. Even if the feature is core functionality, the plain-language instructions normalize collection of behavioral telemetry without safeguards, making unintended disclosure or misuse of preference data more likely.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The documented design exposes a localhost redirect service to the public internet through a serveo.net SSH tunnel, yet the skill includes no warning about external reachability, abuse potential, or data leaving the host via a third-party tunneling provider. This increases the chance that an operator unintentionally publishes a tracking endpoint and associated logs without understanding the security boundary change.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The output format requires '中文内容概要' and the examples are fixed to Chinese, but the document does not provide a language choice or explain why a Chinese-only output policy is required. This is a natural-language locale constraint that should be optional or explicitly justified.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The operational workflow instructs the system to read historical click logs before generating new outputs, meaning previously collected user activity data is repeatedly reused for profiling. Reuse increases the privacy impact of any initial overcollection and makes stale or sensitive behavior data more likely to influence outputs beyond what users expect.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The /clicks endpoint returns the full click log with no authentication and explicitly enables cross-origin access via Access-Control-Allow-Origin: *. Because the same service is also publicly exposed through the reverse tunnel, external parties can read visitor IPs and visited URLs, creating a clear confidentiality/privacy issue.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The /clear endpoint allows anyone who can reach the service to erase the log with a simple unauthenticated request. In context, the reverse tunnel makes this remotely accessible, enabling loss of audit data and easy tampering with whatever tracking or monitoring the tool is intended to perform.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

Starting a reverse SSH tunnel subprocess without clear disclosure is risky because it silently creates an outbound connection that publishes the local service externally. While the lack of warning is partly a transparency issue, here it directly contributes to hidden network exposure and amplifies the impact of the unauthenticated endpoints.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

This manifest-style JSON heavily encodes Chinese-language terms as the primary keyword set, alongside a few English brand aliases, but provides no natural-language indication that the skill is China-specific or that users can choose another language/locale. Under the policy, locale-specific behavior should be explicitly justified or offered as an opt-in rather than silently baked into the configuration.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script presents itself as a simple RSS system launcher, but it also exposes a local service to the public internet via a Serveo reverse tunnel and then rewrites local configuration to use that public URL. That behavior materially changes the trust boundary and network exposure of the system, creating a risk of unintended external access, data leakage, and abuse if the redirected service is reachable by untrusted parties.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

Disabling SSH host key checking removes protection against man-in-the-middle attacks when connecting to the tunneling host. An attacker able to intercept DNS or network traffic could impersonate the remote SSH endpoint, capture tunneled traffic, or manipulate the session without the user's knowledge.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script silently overwrites /root/.openclaw/workspace/rss-weights.json with a new redirect_base value derived from the public tunnel URL. Unannounced modification of application configuration can redirect traffic, persist unintended exposure, and surprise operators who expected a read-only startup action.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The inline Chinese comment states 'return newly added clicks (added after the last query),' which implies server-side statefulness based on the previous request. In reality, the endpoint simply uses the since query parameter and does not remember any prior query, so the documentation actively misdescribes the behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

The inline comment is written in Chinese with no indication that language choice is configurable or intentional for a region-specific skill. This can violate a language/locale policy when the skill otherwise appears to target a general audience and does not offer user choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script's comments and terminal output are written entirely in Chinese, which imposes a specific language on users without offering a choice. This matches the language/locale policy concern for natural-language content in code files.

Content

No source excerpt is available for this finding.