mimic

Run Mimic-generated eval scenarios against a target agent skill (default `revenue-recovery`) inside this Claude Code session. Reads `.mimic/exports/mimic-scenarios.json`, sub-agents one run per scenario via the Agent tool, scores responses with hybrid (strict substring + LLM-judge-on-miss + numeric-range) checks, and prints a scored table. Use when the user says any of "run the mimic eval", "score the agent", "eval against the facts", "/mimic-eval".

Identifies cancelled and downgraded customers in the window, with prior MRR, current MRR, delta, and reasons pulled from Stripe cancellation_details and Postgres support_notes. Sub-step inside revenue-recovery.

Quantifies the MRR delta for a window and decomposes it into involuntary churn, voluntary churn, downgrades, refunds, and expansion offset. Sub-step inside revenue-recovery — do not invoke standalone unless the operator explicitly asks for the diagnosis only.

Pulls failed payment_intents from Stripe in the window, ranks retry candidates by success probability based on decline reason, and surfaces patterns (BIN clusters, expiry waves, single-bank declines). Sub-step inside revenue-recovery.

Compares every customer present in both Postgres and Stripe and surfaces drift — status mismatches, mrr_cents disagreements, orphans, and refunds in Stripe not reflected in payments. Sub-step inside revenue-recovery.

Use when the operator asks why MRR dropped, who's affected, or what to do about failed payments and churn. Investigates Postgres + Stripe in parallel, produces a recovery plan, and on explicit approval executes payment retries, refunds, and subscription changes against Stripe. Two-phase contract — never auto-executes writes.

Correlates affected customers (from churn-investigation, payment-failures, reconciliation) with internal risk_flags and support_notes to surface "we knew this was coming" findings. Promotes any customer with a relevant prior signal up one risk tier in the recovery plan.

Use when the user has an upcoming external meeting and needs context. Pulls from CRM, meeting notes, email, internal Slack, and the product DB in parallel and produces a one-screen briefing. Strictly read-only — never writes to any system.

Pulls Gmail thread state (last 5 messages + total count + unsent drafts) and Slack

Pulls per-source call/meeting counts (Granola, Gong, HubSpot calls, HubSpot meetings) plus 3-5 highlights from the most recent 2-3 calls. Surfaces split-call-history when sources disagree. Sub-step inside briefing-prep.

Pulls every deal at the account/domain from Attio and HubSpot — stage, value, line items, products, slip pattern, and per-CRM disagreements with computed deltas. Use as a sub-step inside briefing-prep — do not invoke standalone unless the user explicitly asks for deal context.

Run Mimic-generated eval scenarios against a target agent skill (default `briefing-prep`) inside this Claude Code session. Reads `.mimic/exports/mimic-scenarios.json`, sub-agents one run per scenario via the Agent tool, scores responses with hybrid (strict substring + LLM-judge-on-miss + numeric-range) checks, and prints a scored table. Use when the user says any of "run the mimic eval", "score the briefing agent", "eval the brief against the facts", "/mimic-eval".