GTM Superintelligence processes sales/customer call transcripts and can connect to CRMs — so we take security and privacy seriously, and we hardened the project before asking anyone to put it in front of real data.

Please report security issues privately to [email protected] (or open a GitHub security advisory). Do not file public issues for vulnerabilities. We aim to acknowledge within 2 business days.

• Transcripts stay local unless you send them somewhere: the Anthropic API (only when you run live coaching with your ANTHROPIC_API_KEY), or a CRM you explicitly connect. Running inside Claude Code needs no API key and sends nothing externally.
• Telemetry is opt-in and off by default, and never includes transcript content or PII — see docs/telemetry.md.
• CRM writes are dry-run by default. Live writes require credentials you provide and are clearly flagged.
• --redact masks obvious PII (emails, phones, SSNs, cards, URLs) before any model call. It's best-effort, not a compliance guarantee — see docs/privacy-and-pii.md.

The agent templates and YAML rubrics are configuration/prompts, not executable code — but they do instruct an LLM and can reference CRM/Slack actions. Treat any externally contributed template like untrusted input:

• Review a template's instructions and tool/trigger config before enabling it against real systems.
• Grant the least privilege needed (read-only CRM scopes where possible; scoped Slack channels).
• Be wary of prompt-injection in transcripts themselves: an agent that reads call text should not be given destructive, unconfirmed write permissions.

Never commit API keys. Use environment variables / .env (gitignored). ATTENTION_API_KEY, SF_ACCESS_TOKEN, HUBSPOT_ACCESS_TOKEN, and ANTHROPIC_API_KEY should come from your secret manager in production.