The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/package.json file contains several dependencies with non-existent or future version numbers that do not match current releases in the public NPM registry. This includes next@16.0.5, react@19.2.0, openai@6.9.1, lucide-react@0.555.0, and @anthropic-ai/sdk@0.71.0. These unverifiable versions present a dependency confusion risk and will likely lead to installation failures or the potential execution of malicious code if published by an attacker to the public registry.
[INDIRECT_PROMPT_INJECTION]: Untrusted meeting transcripts are ingested and processed by the AI model without adequate protection mechanisms.
Ingestion points: Meeting transcripts are read from a file path provided as a command-line argument in scripts/cli.ts.
Boundary markers: Absent. The buildExtractionPrompt and buildTermPrompt functions in scripts/src/lib/extract-service.ts directly append raw transcript chunks to the LLM instructions without delimiters or protective framing.
Capability inventory: The skill possesses network access to the Cerebras AI API and file-writing capabilities for the markdown output.
Sanitization: Absent. The AI's response is parsed via regex and JSON.parse without structural validation or content filtering for malicious instructions returned by the model.
[EXTERNAL_DOWNLOADS]: Communicates with the Cerebras AI API (https://api.cerebras.ai/v1) using the OpenAI Node.js SDK to perform transcript analysis. This is a standard operation for the skill's stated purpose.

transcript-analyzer