The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to operate the specify CLI. It facilitates the execution of various SDD commands (e.g., /speckit-plan, /speckit-tasks) which interact with the local filesystem and development environment.
[DYNAMIC_EXECUTION]: The toolkit supports extensions and workflows that can execute scripts or shell commands. The authoring documentation explicitly warns users that shell steps run unsandboxed and provides guidance on avoiding raw string interpolation to prevent command injection.
[EXTERNAL_DOWNLOADS]: The skill allows for the installation of the specify-cli via uv from official GitHub repositories. It also supports installing extensions and bundles from external URLs and catalogs, with documentation emphasizing the need to vet untrusted sources and catalogs before installation.
[INDIRECT_PROMPT_INJECTION]: The workflow system handles external data through inputs.
Ingestion points: Inputs defined in workflow.yml files (referenced in references/authoring.md).
Boundary markers: Documentation recommends using enum constraints to limit input values.
Capability inventory: Subprocess calls via shell steps and extension scripts (references/authoring.md).
Sanitization: The guide explicitly warns that shell interpolation is raw string substitution and advises against direct interpolation of untrusted agent/prompt output.

spec-kit