The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates multiple local PowerShell and Bash scripts to manage project structure, such as setup-plan.ps1 and create-new-feature.sh. These scripts perform operations including Git branch management, directory creation, and file copying from templates.
[INDIRECT_PROMPT_INJECTION]: The workflow ingests user-provided feature descriptions via $ARGUMENTS and parses implementation plans to update AI agent context files (e.g., CLAUDE.md, .cursor/rules/specify-rules.mdc). This ingestion surface could potentially be used to poison the agent's long-term instructions for the project.
Ingestion points: User-provided strings in $ARGUMENTS in SKILL.md; project specification data in plan.md parsed by update-agent-context.ps1.
Boundary markers: Absent. The skill does not use specific delimiters or cautionary instructions when passing user input into the planning workflow or context updates.
Capability inventory: The skill has capabilities for local script execution, comprehensive file system write access for project documentation, and modification of configuration files used by various AI agents.
Sanitization: The scripts utilize regex filtering and character escaping (e.g., sed sanitization in Bash and [Regex]::Escape in PowerShell) to prevent shell command injection when processing branch names and technology stack metadata.

speckit-plan-zh