marketing-pipeline-content-automation
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple external sources (TechCrunch, a16z, Twitter/X, LinkedIn) via a news scraper and directly feeds this data into LLM prompts (Claude and OpenAI) without explicit sanitization or boundary markers.
- Ingestion points: The
scrapeNewsForTopicfunction insrc/lib/crawler/news-scraper.tsfetches data from RapidAPI sources. - Boundary markers: None detected in the provided code snippets; content is interpolated directly using
JSON.stringify(request.researchData). - Capability inventory: The skill can write files to the local system (video rendering via Remotion) and potentially post to social media platforms (Facebook, LinkedIn).
- Sanitization: No evidence of filtering or escaping external content before LLM interpolation.
- [COMMAND_EXECUTION]: The documentation includes standard build and rendering commands (
npm run dev,npx remotion render) that execute local scripts. - [EXTERNAL_DOWNLOADS]: The installation instructions direct the user to clone a repository from
github.com/pennydinh/marketing-pineline-share.git. This is an external, non-vendor source. - [CREDENTIALS_UNSAFE]: The skill requires multiple sensitive API keys (OpenAI, Anthropic, RapidAPI, Facebook, LinkedIn) for operation. While it correctly recommends using a
.env.localfile, the wide access required increases the impact of potential data exfiltration if the environment is compromised.
Audit Metadata