The Agent Skills Directory

[COMMAND_EXECUTION]: The helper scripts scripts/create_bug_report.sh and scripts/generate_test_cases.sh contain a shell command injection vulnerability. The function prompt_input uses the eval command to dynamically assign user input to variable names using the pattern eval "$var_name=\"$input\"". This allows for arbitrary code execution if the input contains shell metacharacters such as backticks, dollar signs for command substitution, or semicolons.
Evidence found in scripts/create_bug_report.sh (lines 27-38) and scripts/generate_test_cases.sh (lines 28-39).
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data including feature requirements and Figma design specifications, which could be used to influence the agent's output or logic.
Ingestion points: Feature descriptions, requirement strings, and Figma design URLs provided by the user or extracted via MCP.
Boundary markers: The skill templates in SKILL.md do not utilize explicit delimiters or instructions to ignore potential injection patterns within the ingested requirements.
Capability inventory: The skill can write files to the local disk via its bash scripts and perform network operations via the Figma MCP integration.
Sanitization: No evidence of input sanitization or validation was found before interpolation into templates or shell variable assignments.

qa-test-planner