Back to skill

Security audit

Podcast

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly benign podcast-production guidance, but it includes advice that can lead to recording guests before they know recording has begun.

Review this skill before installing if you use it for interviews. Do not follow the pre-show recording advice unless every participant has clearly agreed in advance, recording is announced, and any pre-show material gets separate approval before transcription, upload, or publication. Also review consent, copyright, and vendor data policies before using voice cloning or uploading private source material to AI tools.

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
Findings (1)

other

Warning
Location
formats.md:58
Finding

Pre-Show Recording Without Explicit Informed Consent

Content
View full analysis

Vulnerability Details

File Location: formats.md, line 58
Vulnerability Type: Privacy and recording-consent violation
Risk Level: Medium

Vulnerable Content

markdown
### Making Guests Shine
- Let them tell stories fully (don't interrupt climaxes)
- Set up their expertise before they speak
- Use their words back to them ("You mentioned...")
- End with a question that lets them be inspiring
- **Silence after answers** — guests fill silence with deeper insights
- **"Tell me more about that"** — better than next question when interesting
- **Record pre-show chat** — often the best material before "we're recording"

Technical Analysis

The instruction recommends recording a guest's pre-show conversation before the guest believes formal recording has begun. It does not require advance disclosure, explicit consent, a visible recording indicator, or separate approval before the captured material is published.

A participant may reasonably regard pre-show discussion as private and may disclose personal information, confidential business details, credentials, legal matters, or other sensitive content that they would not disclose during an announced recording. The issue is a consent and privacy risk rather than unauthorized system access or privilege escalation, so it falls outside categories T01–T09.

Attack Path

  1. A podcast producer or agent follows the instruction and starts recording before announcing the formal session.
  2. The guest believes the pre-show discussion is private and speaks without knowledge that audio is being captured.
  3. The recording captures private, personal, or commercially sensitive statements.
  4. The producer uploads the recording to editing, transcription, cloud-storage, or AI-processing services.
  5. Pre-show material is incorporated into an episode, transcript, promotional clip, or internal archive without separate approval.
  6. The guest's statements or personal data are exposed beyond the context in which ...[truncated 650 chars]
Remediation
View remediation

Remediation Suggestions

Replace the instruction with consent-centered guidance, for example:

markdown
- **Pre-show recording:** Record pre-show discussion only after every participant has given explicit, informed consent. Clearly announce when recording begins, provide a visible or audible recording indicator, and obtain separate approval before publishing any pre-show material.

Apply the following safeguards:

  1. Obtain explicit consent from every participant before activating any recording device or service.
  2. State whether pre-show audio will be retained, transcribed, uploaded to third-party services, or considered for publication.
  3. Use a clear audible announcement or visible recording indicator when capture begins.
  4. Treat pre-show material as non-publishable by default.
  5. Obtain separate, documented approval before including pre-show content in an episode, transcript, clip, or promotional asset.
  6. Provide a process through which participants can request deletion of pre-show recordings.
  7. Restrict access to recordings and define retention and deletion periods.
  8. Verify and follow all recording-consent and privacy laws applicable to the producer and each participant.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% 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 · ai-generation.md (reported line 118)May include surrounding context.

text

### Ethical Considerations
- Don't clone voices without consent
- Don't create fake interviews with real people
- Don't present AI opinions as human expertise
- Be transparent with audience

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file recommends AI voice cloning and source-upload podcast generation workflows without any caution about consent, impersonation risk, copyright/privacy exposure, or handling of uploaded source material. In a podcast-production skill, these are realistic misuse paths: users may clone voices or upload sensitive recordings/documents to third-party services without understanding legal and privacy implications.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

The example command includes --language en, which imposes English as the transcription language. Under the policy, forcing a specific language without user opt-in is a locale-related natural-language policy concern unless the constraint is clearly documented and justified, which it is not here.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.