The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on several external Node.js scripts located in a scripts/ directory (including onboarding.js, generate-slides.js, add-overlay.js, post-to-tiktok.js, check-analytics.js, and daily-report.js) that are not included in the analyzed fileset, leading to unverifiable local execution.
[COMMAND_EXECUTION]: Instructions direct the agent to install system-level libraries using brew or apt-get and Node.js dependencies via npm install canvas, which involves elevated shell operations.
[DATA_EXFILTRATION]: The skill prompts the user for and stores sensitive credentials, including API keys for OpenAI, Stability AI, Replicate, Postiz, and RevenueCat, into a local configuration file (tiktok-marketing/config.json), creating a high-value target for data exposure.
[EXTERNAL_DOWNLOADS]: The skill encourages the use of an unverified third-party skill manager (clawhub) to download and install additional functionality (clawhub install revenuecat), which introduces potential supply chain risks.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection during the 'Phase 2: Competitor Research' step in references/onboarding-phases.md. The agent is instructed to ingest untrusted data from TikTok and App Store search results without defined boundary markers or sanitization protocols. The agent possesses capabilities for subprocess execution and file system writes, which could be exploited by malicious instructions embedded in researched content.
[PROMPT_INJECTION]: The instructions utilize highly directive and mandatory language (e.g., 'IMMEDIATELY start a conversation', 'ALWAYS portrait aspect ratio') to override default agent reasoning during onboarding and content generation.

tiktok-app-marketing