The skills engine, explained¶

This is the part of CorpusIQ that makes the answers good.

The naive way (which we don't do)¶

Imagine you ask Claude — without CorpusIQ — "how healthy is my business?" It has 31 connectors available. What does it do?

It guesses. Maybe it asks Shopify for revenue. Maybe it forgets to ask QuickBooks. Maybe it pulls GA4 sessions but skips the ad accounts. The next time you ask the same question, it makes different guesses. The answers don't match. You can't compare last month to this month because the model didn't run the same play twice.

That's the failure mode of "agent + a pile of tools." Without structure, you get a different shape of answer every time.

What CorpusIQ does instead¶

When you ask a broad business question, CorpusIQ routes it to a skill — an opinionated, multi-step procedure that knows exactly:

Which connectors to hit (and in what order).
What metrics to compute from the raw data.
How to present the answer (sections, headings, what goes in the appendix).

Same question, same procedure, same answer shape. Every time. Month over month, the numbers are comparable because they came from the same play.

A concrete example¶

You ask: "What's my true CAC and ROAS by channel?"

CorpusIQ routes this to the ad-spend-truth-report skill. The runbook says:

Pull spend, clicks, and self-reported conversions from Google Ads.
Same for Meta Ads.
Same for TikTok.
Pull GA4 sessions and conversion events by source.
Pull actual revenue from Shopify (or QuickBooks if it's a non-store business).
Join ad spend → GA4 sessions → Shopify revenue at the channel level.
Compute CAC and ROAS per channel using revenue from Shopify, not the ad platforms' optimistic numbers.
Format as a per-channel matrix with a verdict.

That's 6 connectors and a specific order of operations. The skill knows it. You don't have to.

A few skills in the catalog¶

Every one of these is a real runbook with a specific output shape:

executive-snapshot — "How healthy is my business?" Cash, revenue, wins, risks, focus list across QuickBooks, Shopify, HubSpot, ad accounts, GA4, email, calendar.
ad-spend-truth-report — the CAC/ROAS reconciliation above.
board-update-drafter — "Draft a board update for last quarter." Pulls financials, KPIs, prior updates from Drive for tone, writes the document.
financial-command-center — full financial health check: cash, concentration risk, expense anomalies, late-payment patterns.
ecommerce-command-center — full store review across Shopify, ads, GA4, and QuickBooks.
data-discrepancy-detector — "Why don't my Shopify and QuickBooks numbers match?" Side-by-side with the gaps explained.
customer-health-scorecard — every active account scored Healthy / At Risk / Critical, with the driving signal.

There are over a hundred more. You don't browse them. CorpusIQ picks the right one based on your question.

When there's no matching skill¶

If your question is narrow ("show me last week's Shopify orders"), there's no skill to route to. CorpusIQ just queries the one connector directly and gives you the data.

If your question is huge and ambiguous ("tell me everything"), CorpusIQ will ask you to narrow it down or suggest a likely skill to run.

Why this matters to you¶

Three things.

Consistency. Last month's executive snapshot and this month's were produced by the same procedure. Variance is real change, not the model's mood.

Quality. Each skill was designed by someone who knows what the right answer looks like. You're not relying on the model to figure it out from scratch.

Speed. You don't have to write a 400-word prompt every time. "Run the executive snapshot" is enough.

What's next¶

../prompts/ — the prompts that trigger these skills, organized by the question you're trying to answer.
privacy-and-security.md — what happens to the data the skills pull.

*

This Hermes repo is one of the largest structured collections of public AI, automation, business, and technology documentation. Content remains attributed to original authors and repositories. Indexed and organized by www.CorpusIQ.io.