Back to skill

Security audit

Subclip Remotion Layer

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because its documentation includes examples flagged for broad network access and other workflow boundaries, while part of the scan was incomplete.

Consider the suspicious verdict before installing. In particular, review whether the examples that load remote resources fit your needs and whether those requests and resource checks are acceptable. The AIG scan completed with no findings, but Skillspector covered 95.2% of components and reported partial analysis, including parser limits and unresolved references, so the scan evidence is incomplete.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (11)

Context Leakage

High
Category
Data Exfiltration
Confidence
75% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · references/authoring-contract.md (reported line 97)May include surrounding context.

md
Cesium's asset loader. True elevation-terrain or photorealistic flyovers are not available in the layer runtime.

Use `arcHeight: 0` for normal surface routes; raised arcs are opt-in and may be visible above the horizon. Keep globe depth testing enabled. Read [globe-rendering.md](globe-rendering.md) for sampling and capture readiness. In asynchronous React code, obtain `delayRender`, `continueRender`, and `cancelRender` from `useDelayRender()`, and use `useRemotionEnvironment()` for the active capture context. A global delay handle does not block `renderStillOnWeb`'s scoped render. Release readiness only after the ready UI commits and render-critical assets/canvas drawing finish, and release each abandoned task's own handle during cleanup.

Lottie animation JSON must arrive as a validated prop and be passed directly to `<Lottie animationData={...} />`; direct `fetch()` is unavailable. For a public Lottie JSON URL, use `import_public_lottie_json` first. It saves a checked project-local

Unbounded Output

Medium
Category
Output Handling
Confidence
60% confidence
Finding

Output size or generation rate is not bounded. Unbounded output enables denial-of-service through resource exhaustion, log flooding, or context-window stuffing.

Content

Scanner excerpt · references/globe-rendering.md (reported line 7)May include surrounding context.

md
cHeight: 0` in the Cesium component, Motion Graphics template, and `plan_globe_route`. The renderer adds 1 km of clearance to avoid z-fighting. Extra elevation is opt-in; a high arc can remain visible beyond the limb because that part of the route really is above the horizon. Do not diagnose that as an incorrect coordinate or fix it with an overlay mask.

Each geodesic leg is sampled at intervals no greater than 25 km, including every waypoint, so straight GPU segments follow the curved surface without cutting through it. Route reveal uses cumulative route distance and shares its exact endpoint and altitude with the moving marker. Pins and labels keep depth testing enabled; the host also checks ellipsoid line of sight for markers. Preserve occlusion during preview, seeking, and export.

For an old saved map, `update_runtime_layer` with `mergeBindings: true` and `props: {arcHeight: 0}` changes only the elevation. Read its source before refreshing `templateId: "map-explainer"`: a customi

The caption example implements standalone project composition setup despite documenting a host-managed embedded-layer workflow.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

In references/official/remotion-best-practices/remotion-captions/display-captions.md, L008 and L021 state that standalone setup is unavailable and Subclip manages dependencies and execution. However, L045-L066 supply a standalone composition wrapper and public-folder media reference, rather than a component that uses Subclip's existing composition and imported asset bindings; this contradicts the documented execution boundary.

Content

No source excerpt is available for this finding.

The subtitle-loading example relies on standalone public-folder and direct-network loading instead of the declared Subclip resource boundary.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The example directs the agent to load subtitle resources directly and explicitly permits arbitrary remote URLs, bypassing the documented managed-resource workflow. If followed through a network-capable execution path, this introduces unvalidated external requests and unbounded response parsing; the stated lack of layer fetch support limits exploitability but does not make the guidance appropriate.

Content

No source excerpt is available for this finding.

The metadata example introduces arbitrary URL fetching beyond the layer-authoring purpose.

Medium
Category
Excessive Agency
Confidence
85% confidence
Finding

The example fetches a props-controlled URL and incorporates the response into rendering props without resource validation or response-size limits. This conflicts with the managed layer boundary and could permit unintended network access if adapted into a network-capable host or tool; the supplied code does not demonstrate a sandbox escape.

Content

No source excerpt is available for this finding.

The effects example changes standalone renderer configuration rather than remaining within layer authoring.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

L039-L041 imports the Remotion CLI configuration API and changes the Chromium OpenGL renderer. This is host render-environment configuration, not React/TSX layer creation using the pinned runtime, and conflicts with the file's stated reliance on Subclip-managed execution at L017-L019.

Content

No source excerpt is available for this finding.

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · references/official/remotion-best-practices/remotion-markup/gifs.md (reported line 96)May include surrounding context.

Control what happens when the animation finishes:

tsx
// Loop indefinitely (default)
<AnimatedImage
  src={props.animationUrl}
  width={500}

The Lottie-loading workflow fetches a third-party asset without a privacy notice.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The substantive issue is not merely a missing privacy notice: the example loads mutable third-party JSON directly into Lottie without the required validated project-local import. In a network-capable context, this exposes request metadata and bypasses asset checks; the documented prohibition on layer fetch reduces immediate exploitability.

Content

No source excerpt is available for this finding.

The font-loading example introduces external font requests without disclosure.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The best-practice workflow instructs loading a font through @remotion/google-fonts, which can retrieve font resources from an external font service. It does not disclose that this can expose the requesting system's IP address and request metadata, even though text measurement itself is local.

Content

No source excerpt is available for this finding.

The sound-effect example instructs remote media loading without a privacy notice.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The example binds an external URL directly to an approved media component instead of first importing and verifying the asset through Subclip's library. Where accepted by the host, this causes third-party requests and introduces a mutable external dependency; no transmission of project contents or credentials is shown.

Content

No source excerpt is available for this finding.

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · references/official/REMOTION-LICENSE.md (reported line 35)May include surrounding context.

md
### Warranty notice

The software is provided "as is", without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose, and non-infringement. In no event shall the author or copyright holders be liable for any claim, damages or other liability, whether in an action of contract, tort or otherwise, arising from, out of or in connection with the software or the use or other dealings in the software.

### Support