The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate_brief.sh contains a command injection vulnerability in the prompt_input function. It uses eval "$var_name='$input'" to assign interactive user input to variables. An attacker providing input with single quotes and shell metacharacters (e.g., value' ; command ; ') can execute arbitrary code in the shell environment.
[DYNAMIC_EXECUTION]: The skill uses eval for dynamic variable assignment from unvalidated user input. This pattern is inherently risky as it allows for the execution of arbitrary shell logic constructed at runtime from external sources.
[INDIRECT_PROMPT_INJECTION]: The skill ingests raw user input for fields like project description and goals and writes them directly into a generated Markdown document. This allows for the inclusion of malicious instructions that could target downstream processes or agents reading the brief.
Ingestion points: User input via read -r in scripts/generate_brief.sh.
Boundary markers: None present in the generated output to delimit untrusted content.
Capability inventory: File system write access and shell command execution.
Sanitization: None; the script does not validate or escape user-supplied strings before use in eval or file writing.

design-brief-generator