Book review

The engine's daily loop, run inside Claude. Produces the digest; runs playbooks only when asked.

Step 1: Scope the book

Read cs-engine/stack-profile.md. Resolve "my book" as accounts in ~~crm owned by the operator's crm_owner_id (ask once if unknown). Accept a narrower scope when the user gives one ("enterprise only", "renewing this quarter", "the ten biggest"). Cap a single run at 40 accounts; for larger books, run by segment or ARR tier and say so: a digest that takes an hour is not a digest.

Step 2. Tier 0: score everything from the CRM alone (1–3 calls)

List the book's accounts with their properties in as few calls as the CRM allows: ARR, renewal date, seats, owner, champion/decision-maker flags, last activity date, contraction flags, open renewal opportunity. Build a provisional Customer 360 for each with only account, contract, relationship (from CRM contacts), opportunities, cadence, lifecycle, and actions_last_30d; every other block gets its "unavailable" default and flag from field-sources.md. Score and evaluate triggers on these:

python3 ${CLAUDE_PLUGIN_ROOT}/src/orchestrator.py --input cs-engine/accounts/ --triggers-only --out cs-engine/runs/<today>

(--input accepts a directory; it reads every customer_360.json beneath it.) This pass is seconds, not minutes. Reuse any customer_360.json less than one day old instead of rebuilding it.

Step 2b. Tier 1: deep pull only where it matters

For accounts that fired any trigger at Tier 0, or are inside 120 days of renewal, or are in the top 10 by ARR, pull usage, support, and sentiment from their connectors (per field-sources.md) and re-score. Everything else stays at Tier 0 and is labeled CRM-only score in the digest. Never deep-pull more than 15 accounts in one run; say when the cap was hit and which accounts were left at Tier 0.

Without Python, score and evaluate by hand using docs/02_METRICS_TAXONOMY.md §B1 and docs/03_PLAYBOOKS_MASTER_INDEX.md §H, and say so.

Step 3: Portfolio notes

Compute with src/models/metrics_calculator.py (or by hand): book ARR, ARR-weighted health, band counts, renewals in 30/60/90/120 days with ARR, accounts over 85% utilization (expansion-ready) and under 50% (contraction risk), QBR/EBR due in 30 days, and the capacity flag (book > 115% of the segment target with ≥ 2 red accounts) per docs/02_METRICS_TAXONOMY.md §D1. When ~~data warehouse exposes mart_retention_nrr_grr, add the segment's TTM NRR/GRR.

Step 4: The digest

Write cs-engine/runs/<today>/digest.md and reply with it. Write it for a CEO who knows none of these accounts (${CLAUDE_PLUGIN_ROOT}/skills/customer-360/references/writing-standard.md): every line says who the customer is in value terms, what is happening in plain words, and what we will do. In this order:

1. Headline: one sentence on the book and what needs action: "31 customers worth $4.1M a year; two need action today, together $420,000 a year."
2. One line on how to read it: health is 0 to 100, built from product use, support, relationships and contract; 75 and up is healthy, under 50 is at risk.
3. Act today (P0: red accounts, champion departures, aged P1s, ECRI ≥ 60, renewals ≤ 120d with health < 65), This week (P1: TTV stalls, feature-depth erosion, standard renewal defense), Coming up (P2: EBRs due in 30 days), Room to grow (P3: utilization ≥ 85%, gate hits, PQL, with a rough size when obvious). The criteria in parentheses are for you; the reader sees only the plain headings. One line per account: **<account>** ($<ARR> a year, renews <date>): <what is happening, in plain words, with the one number that proves it>. → <what we will do>.
4. Since last time: accounts that changed band, new urgent tickets, main contacts who left or were replaced (diff against the previous runs/<date>/digest.json).
5. The whole book: the numbers from step 3 in one or two sentences.
6. What I could not see: missing connections, and how many accounts are scored with less information because of it.

Keep codes out of the lines; the playbook skill name goes in the details only when the user runs it.

Then one AskUserQuestion (multi-select): which flagged accounts to run the full playbook on now. Run only the selected ones, each through its skill. If the stack profile records pre-approved playbooks for scheduled runs, run those automatically for P0 items only and leave the rest as the digest.

Scheduling

If the user wants this daily or weekly, set it up as a scheduled task with this skill as the instruction ("run the cs-claude-engine book review for <operator>, digest only, post to <channel> if ~~chat is connected"). In a scheduled run with nobody present: never execute CRM writes or communications; write the digest, post it to ~~chat when permitted, and stop.

Rules

• A digest line is one line a stranger to the account can understand. The playbook skills carry the detail.
• Never rank by ARR alone; rank by priority, then ARR within priority.
• Keep connector calls proportionate to the book: batch reads where the CRM allows (list companies with properties in one call; list open tickets filtered by account ids in one call).
• Save runs/<today>/digest.json (machine-readable: per-account score, band, fired playbooks, reasons) for the next run's diff.