io.github.farssaadeldin7-ux/customer-sales-support

Mechanical checks on one drafted article with the evidence quoted: length over 400 words, a title without a question, missing "Applies to", owner or review date, preamble phrases before the answer, failure talk with no literal error string, policy-looking values in a procedure article, region words under an "all customers" scope, plus the template sections present and the extracted first two sentences. Facts only — whether the first two sentences answer the intent is a judgement it does not make, and set-level failures (near-duplicates, stale outranking) are outside a single-article lint. Requires a paid plan.

The article architecture: the template, the style rules with their reasons, the policy-versus-procedure split and its one-question test, the five anti-patterns with their rewrites, the maintenance rules, and the four retrieval failure modes. Reference data only — use article_lint to check a draft against the mechanical subset. Omit section to get everything.

Open the Stripe billing portal to change plan, update the card, or cancel the Customer Sales Support subscription.

The escalation and evaluation rules: the hard-trigger table with its detection cues, the refund-threshold guidance, the handover format and its rules, the grounding contract, the regression-set composition, the four outcomes, the metric definitions with formulas and targets, and the rollout gates. Reference data only. Omit section to get everything.

Screen one customer message, plus any supplied conversation state, against the hard-trigger table's literal detection phrases, quoting every match. Literal matching only: it misses paraphrase, reads no sentiment, and a message can require escalation without matching anything here — no match is not clearance. For exercising the trigger list against real ticket text, not a replacement for the agent's own trigger implementation. Requires a paid plan.

Store and activate a Customer Sales Support licence key on this machine. Use when the user pastes a licence key, or after a purchase completes.

Check the Customer Sales Support licence on this machine: plan, included capabilities, remaining quota and seat usage. Call this when a paid tool reports a licence problem, or when the user asks what their plan includes.

List the available Customer Sales Support plans, prices and what each one includes.

The recorded regression runs on this machine, newest first, each with the arithmetic difference from the previous run in percentage points. Counting and subtraction only — it does not say whether a movement is significant. Requires a paid plan; the licence check is the only thing this tool sends anywhere.

Compute the metrics for one scored regression run by the stated formulas — containment, accuracy on contained, false-containment, over-escalation — check the set's composition minimums, flag internally inconsistent cases, and read the two rollout gates a regression run can assess. Arithmetic over outcomes a person has already assigned: it does not judge whether any answer was correct. Set record to true to add the run to the local history. Requires a paid plan.

Begin a Customer Sales Support purchase. Returns a Stripe Checkout link. Call list_plans first if the user has not chosen a plan.

Audit a recorded ticket taxonomy against the method's stated thresholds: the 500-ticket volume floor, top-20 coverage of 70%, shares summing to 100%, the 0.5% article threshold, the policy-change rule, missing resolution paths, unclassified intents and the 15% multi-intent limit. Computes the containment ceiling by the step-2 arithmetic — static share plus automatable account-specific share — and the ordered article backlog. It does not re-classify intents and names no realistic containment figure, only the ceiling. Requires a paid plan.

The intent-classification tables: the three kinds and their agent behaviour, the three-way test with its edge cases, the clustering method and thresholds, the two worked taxonomies (e-commerce and B2B SaaS, illustrative volumes), the recording format and the symptoms of a wrong taxonomy. Reference data only — it does not classify your intents. Omit section to get everything.