The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill reads local files to retrieve authentication data for external platforms.
Evidence: The process loads ~/.claude-marketing/brands/{slug}/profile.json which is explicitly stated to contain "connected platform credentials".
[COMMAND_EXECUTION]: The skill executes a local script to manage user approvals for high-risk actions.
Evidence: The skill invokes python "${CLAUDE_PLUGIN_ROOT}/scripts/approval-manager.py" with brand and action arguments during the execution gate process.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from diverse external sources to generate dashboard specifications.
Ingestion points: The skill ingests marketing data from Google Analytics, Google Ads, Meta Ads, CRM systems, and email platforms via connected agents.
Boundary markers: An explicit "Execution gate" is implemented requiring a typed yes from the user before proceeding with external operations.
Capability inventory: The skill performs subprocess calls to approval-manager.py and utilizes specialized agents to map data sources to dashboard widgets.
Sanitization: No specific sanitization or filtering logic is described for the content retrieved from external marketing platforms before it is incorporated into the dashboard design specification.

live-dashboard