The Agent Skills Directory

[COMMAND_EXECUTION]: The troubleshooting section in references/sdd_install.md provides a shell script for Linux users that executes sudo dpkg -i to install a downloaded package. While intended for legitimate tool installation, the use of elevated privileges for external binaries is a high-risk operation.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download and installation of the specify-cli and git-credential-manager from GitHub repositories. Although these are well-known resources, they involve executing code and binaries fetched from remote servers.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests and processes user-generated project artifacts to perform analysis and task tracking.
Ingestion points: The analysis scripts in the scripts/ directory read and parse contents from .speckit/features/ files, such as tasks.md, spec.md, and plan.md (e.g., in scripts/phase_summary.sh and scripts/analyze-requirements.py).
Boundary markers: There are no explicit boundary markers or instructions to the agent to ignore embedded commands within the processed project files.
Capability inventory: The skill possesses the ability to execute shell commands (bash) and Python scripts, and it can modify the project directory structure.
Sanitization: The analysis logic relies on regex and simple string matching, which does not provide robust protection against maliciously crafted content in the input files.

sdd