The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a local Python script scripts/check-output.py and instructs the agent to execute it to validate deliverable consistency. The script performs regex-based checks on the output content and does not involve network requests or sensitive file access.
[INDIRECT_PROMPT_INJECTION]: The skill architecture is designed to ingest data from external sources such as CRM exports, job postings, and company watchlists. This data is used to trigger outbound actions (sequencers, CRM tasks, Slack alerts), creating an attack surface for indirect prompt injection.
Ingestion points: External data enters via the Source table from CRM exports (L04 Stale Opportunity), job postings (L03 Hiring Signal), and external watchlist sources (L01 Funding Signal).
Boundary markers: The skill recommends using "Gates before sequencer" such as human_approved flags and LeadMagic verify status to separate automated processing from outbound actions.
Capability inventory: The workflow enables automated routing to external systems including HubSpot/Salesforce (CRM), Smartlead (Sequencers), Slack, and n8n webhooks.
Sanitization: The skill outlines a "Human review queue" for mid-tier scores and a "Suppression check" to filter data against existing customer lists and opt-out flags.

clay-loops-toolkit