The Agent Skills Directory

[Dynamic Execution] (HIGH): The scripts scripts/create_bug_report.sh and scripts/generate_test_cases.sh use eval to assign user-provided input to variables.
Evidence: eval "$var_name=\"$input\"" in scripts/create_bug_report.sh (line 35) and scripts/generate_test_cases.sh (line 33).
Risk: This is a classic shell injection vulnerability. If the input contains shell metacharacters or subshell executions (e.g., $(command)), they will be executed by the shell during the eval call. This allows an attacker who can influence the data processed by the agent to achieve Remote Code Execution (RCE).
[Indirect Prompt Injection] (LOW): The skill's primary function is to process external descriptions of bugs and test cases, creating a significant surface for indirect prompt injection.
Ingestion points: Data ingested via read in scripts/create_bug_report.sh and scripts/generate_test_cases.sh.
Boundary markers: None present; the scripts treat all input as trusted data.
Capability inventory: Subprocess execution and file-system write access.
Sanitization: No sanitization or validation is performed on the input before it is passed to the eval command, making the script highly vulnerable to adversarial data.

qa-test-planner