qa-test-planner
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMCOMMAND_EXECUTION
Full Analysis
- [Command Execution] (MEDIUM): The scripts
scripts/create_bug_report.shandscripts/generate_test_cases.shutilizeevalto assign user-provided input to internal variables. - Evidence: The line
eval "$var_name=\"$input\""in both scripts allows for command substitution. If a user provides input like$(whoami), the command will be executed by the shell during the evaluation process. - [Indirect Prompt Injection] (LOW): The skill provides templates and scripts that generate structured markdown files intended for further processing by agents or developers.
- Ingestion points:
scripts/create_bug_report.shandscripts/generate_test_cases.shcapture raw user input. - Boundary markers: The generated markdown uses standard headers but lacks explicit 'ignore embedded instructions' warnings for downstream LLM processors.
- Capability inventory: The scripts have file-write capabilities and the ability to execute shell commands via the
evalvulnerability. - Sanitization: There is no sanitization or escaping of the input before it is evaluated by the shell or written to the final markdown files.
- [Data Exposure] (LOW): The scripts allow users to specify an output directory via the first command-line argument without validation.
- Evidence:
OUTPUT_DIR="$1"allows a user to potentially overwrite files in sensitive locations if the script is run with high privileges.
Audit Metadata