The Agent Skills Directory

[SAFE]: The skill demonstrates best-in-class security practices by providing the agent with explicit instructions and reference scripts to validate its own operations. It includes detailed logic for preventing path traversal (rejecting patterns like '../' and access to sensitive directories like '.ssh'), detecting hardcoded secrets in generated documentation, and flagging potential SQL or shell injection patterns in verification criteria.
[COMMAND_EXECUTION]: The skill instructs the agent to utilize standard shell utilities such as grep and file to perform the aforementioned security validations on generated requirements and output paths. These commands are used defensively to ensure the integrity and safety of the output.
[PROMPT_INJECTION]: The instructions contain a directive to override specific task-level output path instructions in favor of direct user confirmation. While technically a behavioral override, this is implemented as a safety constraint to ensure the user retains full control over the file system and to prevent automated processes from overwriting files without explicit consent.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect injection surface as it ingests untrusted data from project source code and existing requirements files.
Ingestion points: Project source code files (Python, JS, etc.) and existing requirements Markdown files.
Boundary markers: The agent is instructed to scan for and flag injection patterns within the verification criteria of requirements.
Capability inventory: The agent performs file read/write operations and shell-based validation checks (grep).
Sanitization: The skill provides regex patterns to detect and replace hardcoded secrets with placeholders and to identify common injection vectors in the processed content.

sync-req