Back to skill

Security audit

Html Report

Security checks for vulnerabilities and agentic risk

Overview

This skill is a local HTML report template and CSS inliner with disclosed file-writing and browser-opening behavior, but users should notice the fixed Odoo-AI footer and Vietnamese template text.

Install only if you want this exact local HTML report workflow. Review generated reports before sharing, especially the Odoo-AI footer branding and Vietnamese labels, and use --no-open when running in CI or when automatic browser opening is unwanted.

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
Findings (1)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:49
Finding

Mandatory Third-Party Branding Hijacks Generated Report Output

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:49-52
Vulnerability Type: T01: Skill Instruction Hijacking
Risk Level: High

The instruction is implemented by the hard-coded branding in assets/template.html:126-133 and demonstrated again in assets/example.html:176-184.

Complete code snippets:

SKILL.md:49-52:

markdown
3. **Keep the structural wrappers:** `<header><div class="head-inner">…</div></header>`
   with an optional `.eyebrow` pill, an `<h1>`, a `.subtitle`, and a `.meta` row;
   and `<footer><div class="foot-inner">…</div></footer>` with a `.foot-brand`.
   These give the black Magazine header/footer — keep them.

assets/template.html:126-133:

html
<footer>
    <div class="foot-inner">
        <span class="foot-brand"><span class="dot"></span>Odoo-AI</span>
        <span>Generated {{TIMESTAMP}}</span>
        <span>·</span>
        <span><code>{{REPO}}</code></span>
        <span>·</span>
        <span>Nguồn: {{SOURCES}}</span>
    </div>
</footer>

Technical Analysis

The Skill requires an AI Agent to preserve the .foot-brand component in every generated report. The canonical template then assigns that component a fixed third-party identity, Odoo-AI, rather than a neutral or user-controlled placeholder.

This behavior exceeds presentation standardization by imposing persistent promotional attribution on the Agent's final output, even when the user did not request or authorize that branding. Loading and following the Skill therefore alters the Agent's output requirements in favor of the Skill publisher, matching instruction hijacking.

Attack Path

  1. A user asks the Agent to generate an HTML report.
  2. The Agent loads the html-report Skill based on its activation description.
  3. The Skill directs the Agent to begin with assets/template.html and retain the footer and .foot-brand wrapper.
  4. The template supplies the fixed `Od ...[truncated 634 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the requirement that every report retain .foot-brand.
  2. Replace the fixed Odoo-AI text in assets/template.html with an optional, user-controlled placeholder such as {{BRAND}}.
  3. Omit the branding block by default when no brand value is explicitly supplied.
  4. Clearly distinguish structural requirements from promotional attribution in SKILL.md.
  5. Include third-party attribution only when the user explicitly requests or consents to it.
  6. Update assets/example.html so that it demonstrates neutral placeholder content rather than encouraging propagation of a fixed identity.
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (6)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding

The skill instructs the agent to create files under plans/reports/ and to run a shell command (build_report.py) that writes back into the HTML file and opens it, but it declares no explicit tool scope such as allowed-tools or permissions. That mismatch can let the skill obtain broader file and shell capabilities than reviewers expect, increasing the chance of unintended file modification or command execution if the skill is auto-invoked.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The activation text uses very broad triggers such as report, analysis, audit, review, and summary, which overlap with many normal user requests unrelated to this HTML formatter. This can cause the skill to activate unexpectedly and steer the agent into creating files and invoking build steps in contexts where the user did not explicitly ask for this specific local HTML workflow.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The HTML root sets lang="vi", and the visible UI strings throughout the template are in Vietnamese. For SQP-3, forcing a specific language without user opt-in or a documented region-specific justification is a natural-language policy concern.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/build_report.py (reported line 54)May include surrounding context.

python
else:
        cmd = ["xdg-open", str(path)]
    try:
        subprocess.Popen(cmd, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
    except OSError as exc:
        print(f"WARN: could not open browser: {exc}", file=sys.stderr)

Tainted flow: 'new_html' from pathlib.Path.read_text (line 86, file read) → pathlib.Path.write_text (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/build_report.py (reported line 95)May include surrounding context.

python
file=sys.stderr,
            )
            return 1
        report.write_text(new_html, encoding="utf-8")
        print(f"==> inlined {css_file.name} ({len(css)} bytes) -> {report}")

    if not args.no_open:

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

The document declares lang="vi", and the visible content is written in Vietnamese throughout the file. This imposes a specific language/locale without any indication of user opt-in or justification for a region-specific constraint, which matches the natural-language locale policy concern.

Content

No source excerpt is available for this finding.