The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The interactive script for generating PRD documents contains a significant command injection vulnerability. Specifically, the prompt_input function in scripts/generate_prd.sh uses the eval command to process user-supplied strings: eval "$var_name='$input'". This allows an attacker or a malicious user to provide input containing shell metacharacters (e.g., single quotes and semicolons) to break out of the string literal and execute arbitrary commands on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from users, creating a surface for indirect prompt injection attacks. Every Category 8 finding must document the following evidence chain:
Ingestion points: User input is collected through the discovery questions defined in SKILL.md and via the interactive input prompts in scripts/generate_prd.sh.
Boundary markers: Absent. The generated markdown document does not utilize delimiters or specific instructions to ensure the agent distinguishes between system requirements and user-provided data.
Capability inventory: The skill has the capability to execute local shell scripts (generate_prd.sh, validate_prd.sh) and write files to the local file system.
Sanitization: Absent. No sanitization, escaping, or validation is performed on the user-provided context before it is used in the vulnerable shell command or written to the output file.

prd-generator