The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of go run to execute scripts located in the local repository's ../../scripts/cmd/ directory.
[COMMAND_EXECUTION]: Potential for command injection exists in multiple phases where user-provided input is passed to shell commands as arguments.
Evidence: In Phase 2 (SKILL.md), the variable $ANSWER is passed to go run ... --details="answer=$ANSWER".
Evidence: In Phase 5 (SKILL.md), discovery learnings are passed to go run ... --body="<key learnings from discovery>".
Evidence: The $SLUG variable is used as an argument across almost all script execution calls throughout the discovery process.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing untrusted data from external researcher sub-agents and user responses without sufficient sanitization.
Ingestion points: Market research findings (Phase 1) and user answers (Phase 2).
Boundary markers: No delimiters or boundary markers are instructed to distinguish system commands from data when variables are interpolated into shell commands or templates.
Capability inventory: Shell command execution via go run, file system write access for saving value propositions, and Git repository management (commit/add).
Sanitization: There is no instruction to sanitize, escape, or validate the content of user-provided data before it is used in executable command strings or written to the filesystem.

product-discovery