The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run a Node.js installation script (maestro-install.js) and shell commands (ls, git) to set up the environment and detect project structure. These scripts are provided by the skill author and managed within the plugin's own directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the local repository (such as package.json, framework configurations, and local SKILL.md files) to determine how to scaffold the orchestrator, which represents a potential surface for indirect injection from untrusted codebases.
Ingestion points: Reads package.json, project framework configs, and custom SKILL.md files located in the project's .claude/skills/ directory.
Boundary markers: The skill explicitly requires the agent to print a proposed mapping and use AskUserQuestion to get user consent before attaching discovered local skills to the orchestrator.
Capability inventory: Executes local Node.js scripts and shell commands; writes configuration and scripts to .claude/scripts/, modifies .claude/settings.json, and updates the project's .gitignore.
Sanitization: Discovered data is filtered through a "best-fit" relevance logic that only maps skills to specific predefined agent roles, and the final mapping is subject to manual user review.
[SAFE]: The skill implements a security-enhancing hook (bash-validation.sh) designed to block unauthorized access to sensitive .env files by project agents, preventing accidental credential exposure during automated tasks.

maestro-install