The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and strictly 'enforce' content from external brand files such as profile.json, restrictions.md, and messaging.md. These files are intended to provide brand voice and compliance guardrails but represent an indirect prompt injection surface where malicious instructions embedded in the data could influence agent behavior.
Ingestion points: Brand profile and guideline files located in ~/.claude-marketing/brands/{slug}/ and skills/context-engine/.
Boundary markers: The instructions lack explicit boundary markers or 'ignore embedded instructions' warnings when reading these files.
Capability inventory: The skill can execute Python scripts and perform localized file reads.
Sanitization: No sanitization of the brand-supplied content is specified before it is integrated into the agent's context.
[COMMAND_EXECUTION]: The skill utilizes shell commands to run local Python utilities for tracking and benchmarking purposes. Specifically, it executes python "${CLAUDE_PLUGIN_ROOT}/scripts/campaign-tracker.py" and python scripts/benchmark_book.py. These commands involve the interpolation of a {slug} or other parameters, which could present a vulnerability surface if the values are derived from unsanitized user input.
[EXTERNAL_DOWNLOADS]: The skill references several external URLs for API documentation and announcements (e.g., developers.google.com, ads-developers.googleblog.com). These sources belong to well-known and trusted technology organizations (Google, Meta, LinkedIn) and do not pose a security risk.

paid-advertising