Back to skill

Security audit

seedance-product-video

Security checks for vulnerabilities and agentic risk

Overview

The skill is a mostly coherent product-video wrapper, but it should be reviewed because it directs installation of an unpinned external skills repo and contains confusing out-of-scope people/dialogue guidance.

Install only if you are comfortable sending selected product images, prompts, and generation metadata to Ofox and potentially incurring charges after approving a cost table. Avoid letting an agent run the unpinned `npx skills add ofoxai/skills` fallback automatically; prefer a pinned, reviewed copy of the required core skill. Do not use this product-only skill for clips involving people, voices, or dialogue despite the included experimental notes.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:1445
Finding
Unpinned External Dependency Installation Through npx## Vulnerability Details **File Location**: `SKILL.md`, lines 1445–1459 **Vulnerability Type**: Unpinned third-party dependency installation **Risk Level**: Medium ### Vulnerable Code ```text ## If the script isn't found ``` bash: ../ofox-video-core/references/ofox-video.sh: No such file or directory ``` This means `ofox-video-core` isn't installed alongside this skill — not that anything is broken. This skill delegates all execution to it and reaches it by relative path. Fix: `npx skills add ofoxai/skills` (the whole repo). Say that plainly rather than relaying the raw path error, which names neither the missing skill nor the fix. ``` ### Technical Analysis The project does not include the `ofox-video-core` implementation or the delegated `ofox-video.sh` script. Instead, it directs the agent to execute an npm-distributed command through `npx` to install the external `ofoxai/skills` repository. Neither the `skills` CLI nor the repository is pinned to a reviewed version, release tag, commit hash, checksum, or signature. The command also installs the whole external repository rather than only the required `ofox-video-core` component. The code ultimately executed can therefore differ from the code available at audit time. Because the bundled files repeatedly instruct the agent to delegate generation, polling, downloading, API-key handling, and local-file processing to the missing script, the unaudited dependency becomes part of the trusted execution path. ### Attack Path 1. The agent attempts to use the skill while `../ofox-video-core/references/ofox-video.sh` is absent. 2. Following the documented recovery procedure, the agent runs `npx skills add ofoxai/skills`. 3. `npx` retrieves and executes a mutable version of the `skills` package. 4. That package obtains and installs mutable content from the external `ofoxai/skills` source. 5. A compromised package, account, release, dependency, or upstream reposito ...[truncated 1148 chars]
Remediation
## Remediation Suggestions 1. Pin the `skills` npm package to an exact reviewed version rather than invoking an unspecified current release through `npx`. 2. Pin `ofoxai/skills` to an immutable commit hash or cryptographically signed release. 3. Verify downloaded artifacts with a published checksum or signature before installation or execution. 4. Install only the required `ofox-video-core` component instead of the whole repository. 5. Prefer bundling the reviewed `ofox-video.sh` implementation and required reference files with the artifact. 6. Require explicit user approval before downloading or executing installation code. 7. Run the dependency with least privilege and a restricted environment containing only `OFOX_API_KEY` and the files required for the current operation. 8. Restrict filesystem writes to a dedicated output directory and restrict network access to documented API endpoints. 9. Re-audit the exact pinned dependency version, including its transitive dependencies and API-key handling, before treating it as trusted.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Context-Inappropriate Capability

Medium
Confidence
91% confidence
Finding
The skill explicitly preserves a documented out-of-scope path for presenter-led, dialogue-enabled video generation even though the manifest says not to use it for people/dialogue. That contradiction can cause operators or downstream agents to route human-containing media through a workflow not designed or safety-reviewed for people, moderation, consent, or dialogue handling, increasing the chance of policy bypass and misuse.

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The file says the people/dialogue exclusion still stands, but immediately provides actionable evidence that such clips were successfully produced with this skill's execution layer. In practice, this undermines the restriction and normalizes off-label use, making it easier for an agent to justify sending human images, spoken content, or dialogue workflows through a product-only pipeline.

Missing User Warnings

Low
Confidence
71% confidence
Finding
The skill instructs use of an API key and remote submission to Ofox while deferring the actual safety/privacy warning to another skill's document. That indirection can cause users or calling agents to miss that prompts, images, and possibly product photos are transmitted to a third-party service, which is a data-handling and privacy risk in a media-generation context.

Static analysis

No suspicious patterns detected.