send-email-campaign
Warn
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes multiple Python scripts via shell commands where variables are interpolated directly into the command string. Evidence:
python "${CLAUDE_PLUGIN_ROOT}/scripts/approval-manager.py" --brand {slug} ...andpython "${CLAUDE_PLUGIN_ROOT}/scripts/send-time-optimizer.py" --history {file}. If these variables contain shell metacharacters, it could lead to arbitrary command execution. - [DYNAMIC_EXECUTION]: Runtime command construction occurs in steps 5 and 9 where parameters are assembled from brand context and user-provided inputs to form command-line arguments.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from brand-specific files and user content. 1. Ingestion points:
~/.claude-marketing/brands/{slug}/profile.jsonand brand guidelines (SKILL.md). 2. Boundary markers: Absent. 3. Capability inventory: Shell script execution, file writing, and network operations (SKILL.md). 4. Sanitization: No explicit validation or escaping of ingested data is mentioned before use in logic or shell execution.
Audit Metadata