The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess.run to execute commands for git, npm, and task-master-ai. Most calls use the list format which prevents shell injection vulnerabilities.
Evidence: Found in script.py, mcp_server/taskmaster.py, and mcp_server/lib.py.
[EXTERNAL_DOWNLOADS]: The skill uses an installation script hosted on the author's public GitHub repository. This script clones the repository and copies files to the local skills directory.
Evidence: install.sh and README.md reference https://raw.githubusercontent.com/anombyte93/prd-taskmaster/main/install.sh.
[REMOTE_CODE_EXECUTION]: The skill's primary installation method is a curl | bash pattern from a remote source controlled by the author.
Evidence: Pattern detected in README.md and install.sh.
[PROMPT_INJECTION]: The skill ingests Project Goals and PRD content provided by the user. These inputs are interpolated into agent instructions during discovery and expansion phases without strict boundary markers, representing an indirect prompt injection surface.
Evidence: phases/DISCOVER.md and companion-skills/expand-tasks/script.py ingest untrusted content into prompts.
[COMMAND_EXECUTION]: The skill includes functionality to dynamically generate and write utility scripts (for time tracking, rollbacks, and security audits) to the project's local directory and grants them executable permissions.
Evidence: cmd_gen_scripts function in script.py writes multiple shell and Python utilities.

prd-taskmaster-v2