The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text inputs such as user stories, feature descriptions, and Figma specifications to generate structured documentation. There are no clear prompt boundary delimiters or explicit instructions telling the agent to treat this ingested content strictly as passive data, presenting an attack surface where embedded natural language instructions could manipulate the agent's behavior.
[DYNAMIC_EXECUTION]: The helper bash scripts scripts/create_bug_report.sh and scripts/generate_test_cases.sh employ an unsafe dynamic evaluation pattern within the prompt_input function. The command eval "$var_name=\"$input\"" evaluates raw text captured from the user input. If an attacker inputs command substitutions or shell metacharacters, it can lead to arbitrary shell command execution in the execution context.

qa-test-planner