Back to skill

Security audit

video-add-content-cards

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent video-overlay workflow with disclosed local file writes and browser previews, but users should understand its runtime dependencies.

Use this skill for trusted video projects where you are comfortable with it reading and writing project files under work/, review/, and cache/. Be aware that it opens local HTML pages, may trigger browser network requests from bundled examples, and relies on external tools/packages such as ffmpeg and npx hyperframes.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (25)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The declared description suggests a creative authoring skill for producing transcript-timed video graphics in HyperFrames HTML. This code does not generate HTML, transcript-timed assets, or card content from a video project. Instead, it is a review-application tool: it ingests an existing plan and review export, validates them, marks selected cards as approved, updates metadata like placement and visual treatment, optionally validates chart data, and writes the resulting JSON back to disk. The primary purpose is materially different from the declared authoring purpose, and it includes undeclared file mutation behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The declared description says this skill is for authoring transcript-timed HyperFrames HTML graphics such as titles and lower-thirds. The supplied code does not generate those graphics. Instead, it assembles a review interface from an existing card plan, validates timeline/card metadata, computes source-frame timestamps, extracts JPEG frames from the source video, and embeds review payload data into a content-card review HTML template. Its primary purpose is review-page generation for candidate cards, not direct creation of the described graphics. The frame extraction and review/selection workflow are materially different undeclared capabilities, so this is a mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description is about authoring transcript-timed video graphics as HyperFrames HTML. The actual code chunk does not generate cards, HTML, titles, lower-thirds, or any video-related creative output. Instead, it contains unit tests for concurrency and file-locking behavior in a review-application script, ensuring leases are maintained during validation and atomic writes to cards-plan.json. This is a materially different primary purpose from the declared functionality, so it is a clear mismatch.

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 · examples/index-apex.html (reported line 8)May include surrounding context.

html
<title>video-add-content-cards — apex composition</title>
    <script src="https://cdnjs.cloudflare.com/ajax/libs/gsap/3.12.5/gsap.min.js"></script>

    <!-- apex font: Titillium Web (free "race telemetry" stand-in — the real
         broadcast face is proprietary and its brand guide forbids third-party
         association). Weights enumerated (what the CSS uses): 600/700/900 +
         italics 600/900. Swap the family + weights to re-theme. -->

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 · examples/index-editorial.html (reported line 8)May include surrounding context.

html
<title>video-add-content-cards — editorial composition</title>
    <script src="https://cdnjs.cloudflare.com/ajax/libs/gsap/3.12.5/gsap.min.js"></script>

    <!-- editorial font: Archivo (grotesque sans). Google Fonts <link>, weights
         enumerated (only what the CSS uses: 500/600/800). Swap the family +
         weights to re-theme. A proprietary face needs a self-hosted @font-face. -->
    <link rel="preconnect" href="https://fonts.googleapis.com">

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 · examples/index-teal.html (reported line 8)May include surrounding context.

html
<title>video-add-content-cards — teal composition</title>
    <script src="https://cdnjs.cloudflare.com/ajax/libs/gsap/3.12.5/gsap.min.js"></script>

    <!-- teal theme: font "serif" = system Georgia/Times (no webfont <link>).
         The family is explicit on --serif. -->

    <style>

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill requires file reads, file writes, and shell execution but does not declare any explicit tool scope or permission boundary. That makes the effective privilege surface broader and less reviewable, increasing the chance that an agent can execute unexpected filesystem or OS actions while processing untrusted project content.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

Using npx hyperframes without a pinned version allows the runtime to fetch whatever package version is current at execution time. This creates a supply-chain risk where a malicious or compromised upstream release could execute code on the host during rendering or dependency resolution.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill instructs the agent to launch local HTML files via native OS handlers (open, xdg-open, Start-Process). Opening browser-rendered local content expands the attack surface beyond card authoring because local HTML can trigger browser behavior, external network access, or unsafe handler interactions on the host.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Requiring the agent to launch a generated review page through native OS commands grants an unnecessary capability for this workflow and creates a host-interaction boundary crossing. If the generated HTML or referenced assets are influenced by project data, this can become a vehicle for local script execution in the browser context or unintended application launching.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% 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 · SKILL.md (reported line 224)May include surrounding context.

xdg-open review/03-content-cards/content-cards-review.html

text

If the command fails, diagnose and retry it; do not ask the user to locate the file. The page
shows each source frame, card ID, program time, editable copy, visual treatment, and placement.
Chart layouts also show editable structured data and the point evidence references. Review
every value against its displayed evidence before approving it.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest describes a skill for authoring selective transcript-timed titles, lower-thirds, statistics, comparisons, lists, quotes, chapter cards, or calls to action as HyperFrames HTML graphics for video projects. This example additionally embeds three standalone financial/business chart scenes (bar, pie, and line charts about revenue), which broadens behavior beyond the specifically described content-card types and domain framing.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest describes a skill for adding selective transcript-timed titles, lower-thirds, statistics, metric spotlights, comparisons, lists, quotes, chapter cards, and calls to action as HyperFrames HTML graphics. This example expands into generic bar, pie, and line chart compositions with business revenue data, which is a broader visualization capability than the stated card-oriented overlay scope.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file loads executable JavaScript from cdnjs and fonts from Google-hosted domains at runtime, creating an external supply-chain and privacy dependency. If those third-party assets are modified, blocked, or monitored, the composition can execute unreviewed code or leak access metadata without any disclosure to the user.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The manifest describes a skill for adding transcript-timed titles, lower-thirds, statistics, comparisons, lists, quotes, chapter cards, and calls to action as HyperFrames HTML graphics for video projects. These lines add bar, pie, and line chart scenes with generic revenue/business data, which expands the delivered behavior from selective content cards into a broader chart/infographic template set not reflected in the manifest description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The manifest describes selective transcript-timed titles, lower-thirds, statistics, metric spotlights, comparisons, lists, quotes, chapter cards, and calls to action authored as HyperFrames HTML graphics. This example also adds full bar, pie, and line chart scenes with business-style revenue visualizations, which are a broader graphics capability not reflected in the stated description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest describes a skill for adding selective transcript-timed titles, lower-thirds, statistics, metric spotlights, comparisons, lists, quotes, chapter cards, and calls to action as HyperFrames HTML graphics. This example additionally implements bar, pie, and line chart scenes with corporate revenue content, which expands the behavioral surface into general chart rendering not stated in the manifest description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest describes a skill for authoring selective transcript-timed video overlays such as titles, lower-thirds, statistics, comparisons, lists, quotes, chapter cards, and calls to action. This example extends that scope by adding standalone bar, pie, and line chart compositions about FY2025 revenue, which are generic data-visualization scenes not tied to the stated transcript-timed content-card purpose.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · reference/ux-implementation-plan.md (reported line 23)May include surrounding context.

md
- Modify `skills/video-add-content-cards/scripts/build_review_page.py`: extract frames and fill
  the template.
- Delete `skills/video-add-content-cards/scripts/open_gallery.py`: native OS commands replace it.
- Modify `skills/video-add-content-cards/SKILL.md`: new interview order and automatic open steps.
- Modify `tests/test_content_cards_ux.py`: template, extraction, payload, and UI checks.
- Modify `tests/test_skill_contracts.py`: workflow contract.

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_review_page.py (reported line 92)May include surrounding context.

python
"2",
        str(output),
    ]
    subprocess.run(command, check=True, capture_output=True)
    if not output.is_file() or output.stat().st_size == 0:
        raise RuntimeError(f"ffmpeg did not create review frame: {output}")

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The file pulls executable JavaScript from cdnjs and fonts from Google-hosted origins at runtime, creating a supply-chain and privacy exposure. If those third-party resources are tampered with, blocked, or logged, the example may execute untrusted code or leak request metadata when opened in a browser.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The comment says this is a worked example with 'all 16 cards packed back-to-back at 6s each (96s)'. However, the document continues with three more scenes at 96–114 seconds for bar, pie, and line charts, so the documentation no longer matches what the file actually includes.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
94% confidence
Finding

This example loads third-party JavaScript from cdnjs and fonts from Google-controlled endpoints, which creates a supply-chain and privacy exposure. Anyone rendering the file will make outbound requests and trust remote content at runtime; if the CDN/library is compromised or the environment assumes offline/local-only rendering, this can lead to tracking or malicious script delivery.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The inline documentation states that the teal theme uses no per-element stagger and that rows appear with their cards as single-unit fade scenes. Later code contradicts this by animating bar fills, bar labels, pie sectors, legend items, line points, and axis labels with GSAP stagger values, which is the opposite of the documented motion model.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The design explicitly instructs the agent to open local HTML artifacts on the user's behalf using native OS commands, without a confirmation step or warning. Even though the targets are intended to be local review files, automatic launching of browser content can surprise users, trigger execution of active HTML/JS in local files, and reduce the user's ability to notice if the opened path is not what they expected.

Content

No source excerpt is available for this finding.