Generating plain-English description…
Plain-English summary of the active rules — what agents can do, what gets blocked, and what needs a human.
For agents matching `langgraph-prospect-*` in the `dev` environment, only these tools are allowed: `read_customer_profile`, `score_icp_fit`, `draft_outreach_email`, and `send_outreach_email`. The only approved models are `gpt-5-mini` and `claude-sonnet-5`, and each request may use up to 3,000 tokens; if usage keeps exceeding limits, the system will warn at 80% and then deny once the overage rules are hit (including a hard deny after repeated excesses within the 24-hour UTC window). `send_outreach_email` requires approval before it can be used. PII handling is allowed, but prompt injection and data exfiltration attempts should trigger warnings, and time-window enforcement is deny-based in UTC.