growth-marketing-os-prompts
Warn
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSPERSISTENCECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill clones a repository from an external, untrusted source (https://github.com/growthack88/growth-marketing-os.git) to provide marketing assets.- [PERSISTENCE]: The function
install_claude_skillcopies files from the downloaded repository into the Claude Desktop configuration directory (~/Library/Application Support/Claude/skills). This allows for the installation of new, potentially malicious agent instructions that persist across sessions.- [COMMAND_EXECUTION]: The skill uses thegit clonecommand to fetch remote content from an external repository.- [INDIRECT_PROMPT_INJECTION]: The skill processes content from an external source and uses it directly as system instructions for LLM calls without sanitization. - Ingestion points: Files located in
growth-marketing-os/prompts/andgrowth-marketing-os/skills/within the cloned repository. - Boundary markers: None. The logic extracts raw content from markdown files.
- Capability inventory: Includes file-writing capabilities (
shutil.copy) and network-based LLM interactions. - Sanitization: None. The extracted content is treated as trusted instruction data.
- [DATA_EXPOSURE]: The
install_claude_skillfunction is vulnerable to path traversal because theskill_nameargument is not sanitized before being used to construct a file path. A malicious input could potentially read and copy sensitive files from the local system into the configuration directory.
Audit Metadata