The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection (Category 8). It processes untrusted data and uses it across multiple automated steps to generate source code, documentation, and interface prototypes.\n
Ingestion points: The SOFTWARE_NAME parameter provided by the user and files found within local reference directories (e.g., reference/style.md).\n
Boundary markers: The instructions in Step 1 of the SKILL.md require the agent to distinguish between verified facts and expanded settings in the specification document, providing a conceptual boundary.\n
Capability inventory: The skill has extensive capabilities including writing numerous files, executing shell commands via local Python scripts (calling cloc and playwright), and interacting with the Google Gemini API.\n
Sanitization: There is no evidence of explicit sanitization or escaping of the SOFTWARE_NAME input before it is interpolated into prompts or used in file paths.\n- [COMMAND_EXECUTION]: The skill executes local system commands through Python scripts to facilitate its core functions.\n
The validate_outputs.py script invokes the cloc utility using subprocess.run to verify code line counts.\n
The screenshot_html_prototypes.py script utilizes the Playwright library to automate a headless browser for capturing UI screenshots.\n
These executions are legitimate, scoped to the skill's operational directory, and use safe argument passing methods.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of several standard third-party libraries and tools, including playwright, google-genai, python-docx, pillow, and cloc. These are reputable packages from well-known services and organizations.\n- [SAFE]: The skill implements a secure approach to secret management by reading the GEMINI_API_KEY from a local environment file (~/aj-skills/.env), which is consistent with industry best practices for AI agent skills and does not constitute unsafe credential exposure.

aj-copyright-writer