Back to skill

Security audit

AI Launch Pipeline

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a coherent AI launch monitoring pipeline, but its default screenshot step opens RSS-provided URLs in a browser without meaningful destination validation.

Review this skill before installing if you will run the full pipeline or monitor feeds you do not fully trust. Prefer using --skip-screenshot, running it in a network-restricted sandbox, or adding strict URL and redirect validation before enabling screenshot capture.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/screenshot_capture.py:37
Finding

Unrestricted Browser Navigation Enables Server-Side Request Forgery

Content
View full analysis

Vulnerability Details

File Location: scripts/screenshot_capture.py, lines 37–45; unsafe navigation occurs at lines 21–23
Vulnerability Type: Server-Side Request Forgery (SSRF) through untrusted RSS links
Risk Level: Medium

Vulnerable Code

python
def capture(url: str, out_path: str) -> str | None:
    try:
        from playwright.sync_api import sync_playwright
    except ImportError:
        print("    ⚠ playwright not installed, skipping screenshots", file=sys.stderr)
        return None

    with sync_playwright() as p:
        browser = p.chromium.launch(headless=True)
        page = browser.new_page(viewport={"width": 1920, "height": 1080})
        try:
            page.goto(url, timeout=20000, wait_until="domcontentloaded")
            page.wait_for_timeout(2000)
            page.screenshot(path=out_path, full_page=True)
            print(f"    ✓ {out_path}")
            return out_path
        except Exception as e:
            print(f"    ✗ {url}: {e}", file=sys.stderr)
            return None
        finally:
            browser.close()


def run(launches: list) -> list:
    os.makedirs(SCREENSHOT_DIR, exist_ok=True)
    results = []
    for i, item in enumerate(launches):
        url = item.get("link", "")
        if not url or not url.startswith("http"):
            continue
        print(f"  [{i+1}/{len(launches)}] Capturing: {url[:70]}")
        parsed = urlparse(url)
        fname = f"{parsed.netloc.replace('.','_')}_{int(time.time())}.png"
        out_path = os.path.join(SCREENSHOT_DIR, fname)
        path = capture(url, out_path)

Technical Analysis

RSS entry links are remotely supplied input. The implementation only tests whether a link begins with the characters http, then passes it directly to Playwright's page.goto() method.

It does not strictly allow only HTTP and HTTPS schemes, validate the resolved destination address, restrict destination ports, or reject loopback, private, link-local, reserved, and ...[truncated 1994 chars]

Remediation
View remediation

Remediation Suggestions

  1. Parse URLs with urllib.parse.urlsplit() and accept only exact http and https schemes.
  2. Require a nonempty hostname and reject malformed URLs, embedded credentials, unexpected ports, and ambiguous numeric host representations.
  3. Resolve the hostname before navigation and reject every IPv4 and IPv6 address classified as loopback, private, link-local, multicast, reserved, unspecified, or otherwise non-global.
  4. Explicitly deny known metadata destinations, including link-local metadata addresses and provider-specific metadata hostnames.
  5. Revalidate every redirect target rather than validating only the initial URL.
  6. Configure Playwright request interception so document, frame, script, image, stylesheet, and other subresource requests are blocked when their resolved destinations are unsafe.
  7. Prefer an allowlist of expected public domains when the monitored feed set is controlled.
  8. Run the browser in a sandboxed environment with outbound network policy that denies private and metadata ranges. Network-level egress restrictions should supplement application validation.
  9. Keep screenshot capture opt-in or use --skip-screenshot until destination validation is implemented.
  10. Add tests covering loopback, RFC 1918, IPv6 local addresses, DNS rebinding, redirects to private addresses, alternate IP representations, and malicious subresources.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (11)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill claims to provide a four-stage end-to-end launch-monitoring pipeline, but the analyzed content indicates the implemented behavior is narrower than advertised. This mismatch is dangerous because operators may grant trust, automation, or elevated execution based on documented capabilities and outputs that are inaccurate, which can hide missing controls, unreviewed code paths, or deceptive packaging.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
86% confidence
Finding

The skill advertises executable pipeline behavior with network access, environment-variable use, and file writes, but it does not declare any explicit tool scope or permissions boundary. In an agent ecosystem, this increases the chance of over-broad execution privileges and makes it harder for reviewers or policy engines to constrain what the skill may access or modify.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This code sends a query derived from each launch item's title to an external search service, which may disclose user- or system-provided data to a third party. Although the script logs that it is 'Searching', it does not clearly warn that the data is being sent off-host to DuckDuckGo.

Content

No source excerpt is available for this finding.

Tainted flow: 'out_path' from os.environ.get (line 50, credential/environment) → open (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/product_search.py (reported line 51)May include surrounding context.

python
time.sleep(1)  # rate limit

    out_path = os.path.join(DATA_DIR, "enriched_launches.json")
    with open(out_path, "w") as f:
        json.dump(enriched, f, indent=2, ensure_ascii=False)
    print(f"  Saved {len(enriched)} enriched items → {out_path}")
    return enriched

Tainted flow: 'SEEN_FILE' from os.environ.get (line 13, credential/environment) → open (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/rss_monitor.py (reported line 31)May include surrounding context.

python
def save_seen(seen: set):
    os.makedirs(DATA_DIR, exist_ok=True)
    with open(SEEN_FILE, "w") as f:
        json.dump(sorted(seen), f)

Tainted flow: 'out_path' from os.environ.get (line 109, credential/environment) → open (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/rss_monitor.py (reported line 110)May include surrounding context.

python
# Write output
    os.makedirs(DATA_DIR, exist_ok=True)
    out_path = os.path.join(DATA_DIR, "raw_launches.json")
    with open(out_path, "w") as f:
        json.dump(unique, f, indent=2, ensure_ascii=False)
    print(f"  Saved {len(unique)} new launches → {out_path}")
    return unique

Tainted flow: 'out_path' from os.environ.get (line 44, credential/environment) → open (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/screenshot_capture.py (reported line 51)May include surrounding context.

python
time.sleep(1)

    out_path = os.path.join(DATA_DIR, "screenshot_results.json")
    with open(out_path, "w") as f:
        json.dump(results, f, indent=2, ensure_ascii=False)
    print(f"  Captured {sum(1 for r in results if r.get('screenshot_path'))} screenshots")
    return results

Tainted flow: 'trends_path' from os.environ.get (line 54, credential/environment) → open (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/trend_analyzer.py (reported line 55)May include surrounding context.

python
# Write JSON
    trends_path = os.path.join(ANALYSIS_DIR, "trends.json")
    with open(trends_path, "w") as f:
        json.dump(trends, f, indent=2)

    # Write Markdown report

Tainted flow: 'report_path' from os.environ.get (line 60, credential/environment) → open (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/trend_analyzer.py (reported line 61)May include surrounding context.

python
# Write Markdown report
    report = generate_report(trends)
    report_path = os.path.join(ANALYSIS_DIR, "launch_analysis_report.md")
    with open(report_path, "w") as f:
        f.write(report)

    print(f"  Trends → {trends_path}")

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

This markdown file describes an end-to-end automated workflow and lists outputs later, but it does not clearly warn users up front that running the pipeline will create JSON files, screenshots, and analysis reports on disk. For markdown files, user-facing warnings should disclose behaviors that affect user data or system state, especially for one-click automated workflows.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
76% confidence
Finding

This code persists processed data to disk, but the surrounding documentation only mentions enriching launch items and does not state that results are saved to a local file. For safety-sensitive auditing, file writes should be disclosed via comments, docstrings, CLI help, or other visible user-facing documentation.

Content

No source excerpt is available for this finding.