The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git rev-parse --show-toplevel to resolve the root directory of the plugin. This is a standard and safe use of command execution to determine the environment context.
[DATA_EXPOSURE]: The skill requires access to marketing exports and outcome truth data (orders/leads). It includes explicit instructions not to expose sensitive IDs and only persists data to authorized paths after explicit authorization.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data in the form of ad account exports and CRM data.
Ingestion points: Reads normalized account/portfolio evidence sets from campaign, query, placement, and audience exports (SKILL.md).
Boundary markers: The instructions mandate a typed conversation header from a runbook and specific score states, creating a structured format for output.
Capability inventory: The skill has capabilities for file writing (memory/audits/) and shell execution (git rev-parse), but these are restricted by explicit authorization requirements.
Sanitization: The skill uses a validation script (validate-audit-artifact.py) before persisting artifacts and includes guardrails against high-risk platform data.
[CREDENTIALS_UNSAFE]: The skill correctly instructs users to provide normalized data but does not request or store raw API keys or credentials. It mentions AARON_SKILLS_ROOT and CLAUDE_PLUGIN_ROOT environment variables for path resolution, which is standard practice.

ad-account-auditor