The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses dynamic path manipulation by calling sys.path.insert(0, os.path.join("tools", "supabase")) within its instructions to load the internal SupabaseClient module. While used for local organization, dynamic modification of the Python search path can be a vector for module hijacking.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from an external Supabase database and interpolates it into email templates, creating a potential vector for malicious data to influence agent behavior.
Ingestion points: Lead metadata including title, company, industry, and icp_segment are retrieved from the people table in Supabase and loaded into the agent's context.
Boundary markers: The skill lack specific delimiters or instructional guards to isolate lead data from the generated prompt content, increasing the risk of the agent obeying instructions embedded in the database fields.
Capability inventory: The skill can manage email campaigns via Smartlead MCP tools (mcp__smartlead__*), write files to the local system (CSV exports), and perform write operations to the Supabase database (outreach_log and people tables).
Sanitization: No sanitization, validation, or escaping logic is described for the data retrieved from the database before it is used in email generation or tool parameter construction.

cold-email-outreach