The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes 'Master Prompts' in references/master_qa_prompt.md and references/llm_prompts_library.md that utilize role-playing ('senior QA engineer with 20+ years of experience') and authoritative directives ('CRITICAL INSTRUCTIONS', 'MANDATORY RULES') to override standard agent behavior and force adherence to the skill's specific logic.\n- [PERSISTENCE]: In references/day1_onboarding.md, the skill provides instructions to the agent to modify user shell profile files like ~/.bashrc or ~/.zshrc to persist environment variables. Modifying shell startup files is a classic persistence technique used to maintain changes across sessions.\n- [COMMAND_EXECUTION]: The skill's autonomous execution and onboarding workflows require the agent to execute shell commands such as git clone, pnpm install, and docker exec using external or user-provided input, which could lead to arbitrary command execution if the inputs are compromised.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to read and process external data from CSV tracking files and Markdown test case specifications autonomously. Without boundary markers or input sanitization, these files serve as attack surfaces for indirect prompt injection.\n
Ingestion points: tests/docs/templates/TEST-EXECUTION-TRACKING.csv and category-specific Markdown test case files.\n
Boundary markers: None found; the LLM is instructed to read the files directly.\n
Capability inventory: Access to system commands (git, docker, pnpm) and local Python script execution.\n
Sanitization: No sanitization or validation of the content within the tracking CSVs or test case documentation is performed.

qa-expert