The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill parses Product Requirements Documents (PRDs) from .taskmaster/docs/prd.txt to generate actionable tasks. This represents a high-risk ingestion point for untrusted data. If a PRD contains malicious instructions, the skill's capability to execute commands via npx and modify the project structure could be exploited.
[Unverifiable Dependencies] (MEDIUM): The skill requires the installation of task-master-ai via NPM. The underlying repository (eyaltoledano/claude-task-master) is not a recognized trusted source, posing a risk of supply chain attack or execution of malicious code during the npm install or npx execution phases.
[Data Exposure & Exfiltration] (HIGH): The setup instructions prompt users to provide ANTHROPIC_API_KEY and PERPLEXITY_API_KEY in environment variables. Because the skill has the ability to read local files (PRDs, Git configuration) and perform network requests (via Perplexity for 'fresh information gathering'), there is a risk of credential exfiltration if the tool is compromised.
[Remote Code Execution] (HIGH): The recommended MCP configuration uses npx -y task-master-ai. This command downloads and executes the latest version of the package at runtime, which bypasses version pinning and allows for arbitrary code execution from an untrusted publisher.

task-master