Back to skill

Security audit

whisperx

Security checks for vulnerabilities and agentic risk

Overview

The skill's audio transcription purpose is coherent, but it tells agents to install multiple unpinned external skills with auto-accept, so users should review it before use.

Review and pin the `skills` CLI and PrunaAI skill references before running the install commands, avoid auto-accept where possible, and run setup in a restricted environment. Only provide a Replicate token and audio files you are comfortable sending to the external transcription provider.

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:14
Finding
Unpinned Third-Party Skill Installation Through npx## Vulnerability Details **File Location**: `SKILL.md`, lines 14-21; the same installation pattern is repeated at lines 31-33 and 78-87 **Vulnerability Type**: `T08: Insecure Dependencies` **Risk Level**: Medium ### Vulnerable Code ```markdown Install and load these skills before generating (skip if already in context via `@pruna`): | Skill | Description | Install | | --- | --- | --- | | `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` | | `pruna-api` | Use before any Pruna or Replicate HTTP call — credentials, upload/poll/download, parallel batches, and agent safety. | `npx skills add PrunaAI/pruna-skills@pruna-api -y` | Or install the full suite once: `npx skills add PrunaAI/pruna-skills@pruna -y` ``` ### Technical Analysis The skill instructs users or agents to invoke an unversioned `npx` package and install skills from repository references that are not pinned to audited immutable commit hashes. No checksum, signature, lockfile, or equivalent integrity verification is specified. The `-y` option automatically accepts installation prompts, reducing the opportunity to inspect the resolved package, version, and source before execution. Because both the package resolved by `npx` and the referenced repository content may change independently after this skill has been reviewed, the effective installed dependency is not fully represented by the audited project. No malicious dependency is present in the reviewed artifact, and exploitation requires an upstream compromise or malicious update. Nevertheless, the installation mechanism creates a supply-chain trust boundary without adequate integrity controls. ### Attack Path 1. An attacker compromises the publishing account, package registry entry, repository, mutable reference, or distribution infrastructure ...[truncated 1556 chars]
Remediation
## Remediation Suggestions 1. Pin the `npx` package to a specific, audited version rather than allowing the registry to resolve its current release. 2. Pin `PrunaAI/pruna-skills` to an immutable audited commit hash or signed release instead of mutable skill or branch references. 3. Publish and verify cryptographic checksums or signatures for downloaded skill artifacts before installation. 4. Remove `-y` from documented commands where practical so that the resolved source and version can be reviewed before installation. 5. Use a lockfile, trusted internal mirror, or vendored reviewed dependency to make installation reproducible. 6. Run installation in a sandbox with restricted filesystem, environment-variable, credential, and network access. 7. Document the expected package owner, exact versions, commit identifiers, and verification procedure. 8. Re-audit dependencies whenever their pinned versions are intentionally updated.
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (9)

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding
The skill instructs the agent to run `npx skills add ...` without pinning the `skills` CLI itself to a specific version or immutable reference. That creates a supply-chain risk: future or compromised releases of the CLI could change behavior and execute unintended code during install/use.

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding
This line references `npx skills add ...` in a way that depends on whatever version of the `skills` tool resolves at runtime. If the package is updated maliciously or unexpectedly, an agent following the skill could fetch and execute unreviewed code.

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding
The installation command uses `npx skills` without fixing the tool version, so the executed code is not reproducible or stable over time. In agent workflows, this is more dangerous because automated systems may execute setup commands with limited human review.

Rp1

Medium
Category
MCP Rug Pull
Confidence
85% confidence
Finding
This 'When NOT to use' section still recommends an unpinned `npx skills add ...` command, carrying the same supply-chain exposure as the primary install path. Even though it is ancillary guidance, an agent may still execute it automatically when redirecting to another skill.

Rp1

Medium
Category
MCP Rug Pull
Confidence
85% confidence
Finding
The command on this line invokes an unpinned package-management path through `npx`, which can execute newly resolved code from the registry. Because the skill is designed for automated agent use, this increases the chance of silent execution of compromised dependencies.

Rp1

Medium
Category
MCP Rug Pull
Confidence
85% confidence
Finding
An unpinned `npx skills add` example appears here for `video-editing`, again exposing users to registry drift or package compromise. The contextual risk is meaningful because this skill encourages chaining into other skills, expanding the trusted code surface.

Rp1

Medium
Category
MCP Rug Pull
Confidence
85% confidence
Finding
This follow-on install instruction again relies on unpinned `npx skills`, allowing code execution from a non-immutable dependency source. In a skill ecosystem where agents may dynamically install follow-on capabilities, this can lead to broader compromise if the toolchain is hijacked.

Rp1

Medium
Category
MCP Rug Pull
Confidence
85% confidence
Finding
The line uses a floating `npx`-based install flow for another skill, which is a classic software supply-chain weakness. It is not obviously malicious, but it permits future code changes outside the reviewed skill content to affect agent behavior.

Rp1

Medium
Category
MCP Rug Pull
Confidence
85% confidence
Finding
This final `npx skills add` reference is also unpinned and therefore subject to registry tampering, typo-squatting, or unexpected upstream releases. The skill context does not reduce the risk; if anything, agent automation and cross-skill chaining make it more operationally dangerous.

Static analysis

No suspicious patterns detected.