Back to skill

Security audit

audio-prompting

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly ordinary audio-prompting guidance, but it repeatedly tells users or agents to install mutable external skills through unpinned npx commands with automatic confirmation.

Review the install commands before using them. Prefer pinning the installer and skill sources to reviewed versions, remove `-y` so you get a confirmation checkpoint, and install only the specific related skill needed for a task rather than the full suite. Treat uploads of narration, voice, or client audio as third-party data sharing and confirm cost/privacy expectations first.

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:15
Finding
Unpinned Third-Party Skill Installation Through npx## Vulnerability Details **File Location**: `SKILL.md:15-18`, `SKILL.md:28-35`, `SKILL.md:66-70`, `SKILL.md:75-86` **Vulnerability Type**: Supply-chain risk from mutable external dependencies **Risk Level**: Medium ### Vulnerable Code `SKILL.md:15-18`: ```markdown | Skill | Description | Install | | --- | --- | --- | | `audio-prompting` | Use when crafting TTS, music, or bed prompts for any generative audio model — director style, song structure, and post-production layering. | `npx skills add PrunaAI/pruna-skills@audio-prompting -y` | | `generation-diversity` | Use when writing any generative prompt — ritual seed, explicit structure, scenario axes, and quality gates before paid API calls. | `npx skills add PrunaAI/pruna-skills@generation-diversity -y` | ``` `SKILL.md:75-86`: ```markdown | Skill | Description | Install | | --- | --- | --- | | `gemini-3.1-flash-tts` | Use when someone needs spoken narration or voiceover — explainer tracks, documentary lines, or voice to pair with generated video. | `npx skills add PrunaAI/pruna-skills@gemini-3.1-flash-tts -y` | | `music-2.5` | Use when someone wants an original AI song with vocals — sung lyrics, a style prompt track, or source audio for a music video. | `npx skills add PrunaAI/pruna-skills@music-2.5 -y` | | `stable-audio-2.5` | Use when someone wants light instrumental background music — an ambient bed under dialogue or underscore for reels and explainers. | `npx skills add PrunaAI/pruna-skills@stable-audio-2.5 -y` | | `whisperx` | Use when someone needs word-level timestamps from audio — lyric alignment, cut-safe line boundaries, or caption source timing before burn-in with video-editing. | `npx skills add PrunaAI/pruna-skills@whisperx -y` | | `p-video-2` | Use when someone wants the best-quality short clip from text, images, or audio — polished B-roll, start/end frame animation, or a motion shot with stronger lip-sync. Not for full multi-scene films or talking-head-o ...[truncated 2920 chars]
Remediation
## Remediation Suggestions 1. Pin the `skills` CLI to an explicitly reviewed version rather than allowing `npx` to resolve its current release. 2. Pin every external skill to an immutable commit hash, signed release, or content-addressed artifact. 3. Record and verify cryptographic integrity hashes or signatures before installation. 4. Remove `-y` so that installation requires an explicit user confirmation and review checkpoint. 5. Avoid recommending installation of the full suite when only one component is required; install the smallest necessary dependency set. 6. Review the complete dependency tree, installation lifecycle scripts, and skill instructions before approving a version. 7. Run installation in a sandbox or least-privileged environment without production credentials or sensitive mounted directories. 8. Maintain a lock file or allowlist documenting approved CLI versions, skill revisions, source repositories, and integrity values.
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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (20)

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
The skill repeatedly instructs users to run `npx skills add ...` commands without pinning the `skills` package itself to a specific version. Because `npx` resolves the latest available package by default, a compromised or breaking upstream release could execute unexpected code during installation, creating a supply-chain risk.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
This install command relies on `npx skills` without pinning the `skills` executable version. An attacker who compromises the upstream package or publishes a malicious update could cause arbitrary code execution when a user follows the documented command.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
The documented `npx skills` invocation is unpinned, so it will fetch and run whatever version is current at execution time. That makes the skill documentation a delivery path for supply-chain compromise if the package or its dependencies are tampered with.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
The command at this line invokes an unversioned package through `npx`, which can execute arbitrary upstream code from the latest release. In documentation intended for direct copy/paste, this creates a meaningful and avoidable software supply-chain risk.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
Using `npx skills` without a fixed version leaves the executed package mutable over time. If the package ecosystem is compromised, users following this skill could run attacker-controlled code locally.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
This is another unpinned `npx skills` command, exposing users to arbitrary changes in the fetched executable. The danger is elevated because the file presents these commands as normal installation steps across multiple related skills, increasing the chance of copy/paste execution.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
The installation example depends on an unpinned `npx` package resolution. This enables supply-chain attacks or accidental breakage from upstream updates, potentially resulting in code execution on the user's workstation.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
Because `npx skills` is not version-locked, the actual code executed is uncontrolled by this document and may change between runs. In a skill file that encourages installation of adjacent tools, that broadens the trust boundary and makes compromise more consequential.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
This line again instructs execution of `npx skills` with no version pinning, creating a standard package-resolution supply-chain weakness. A hostile or hijacked release could execute code during install before the user has any opportunity to inspect it.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
The unpinned `npx skills` install example leaves the fetched executable subject to upstream changes and compromise. This is a real risk because these examples are framed as ready-to-run commands, not abstract references.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
This command is vulnerable for the same reason: it executes an unversioned package via `npx`, exposing consumers to malicious or unstable upstream releases. Documentation that normalizes such commands increases practical exploitability through user copy/paste behavior.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
The line contains another unpinned `npx skills` invocation, meaning the executed code is determined by current registry state rather than a reviewed fixed artifact. That is a textbook supply-chain exposure in install documentation.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
The command uses floating package resolution through `npx`, which can silently change behavior or become malicious over time. Since this skill is operational guidance for AI tooling, the context does not mitigate the risk; it still leads users to run untrusted code locally.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
This install command presents an unpinned executable package invocation. If the package, dependency tree, or registry path is compromised, users may execute attacker-supplied code directly from the documentation.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
The security issue here is not the referenced downstream skill name but the unversioned `npx skills` runner. That creates a reusable attack surface across every copy/paste install line in the file.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
Another unpinned `npx skills` example exposes users to arbitrary upstream changes at execution time. The repeated pattern throughout the skill suggests poor supply-chain hygiene rather than a one-off mistake, increasing confidence this is a genuine weakness.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
Because the `skills` package is not pinned, following this documented command may run different code over time, including compromised releases. The context is user-facing installation guidance, so the risk is more practical than a theoretical code smell.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding
The 'full suite' install command also relies on unpinned `npx skills`, magnifying the blast radius because it fetches a broader tool set through a mutable installer. A compromise here could impact more components and more user environments.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The skill instructs users to upload audio to Pruna and use external paid generation services, but it does not explicitly tell users that their audio content will be transmitted to third-party providers or may incur charges. This creates a real privacy and consent risk, especially if narration audio contains sensitive voice data, client material, or unpublished content.

Natural-Language Policy Violations

Low
Confidence
82% confidence
Finding
The guidance says that when locale is open, the agent should initiate clarification intake in the first reply. This imposes locale handling as a required policy without stating that the user may choose to proceed without locale-specific adaptation or opt into a preferred locale, which can be a natural-language locale policy concern.

Static analysis

No suspicious patterns detected.