Playbooks
21 ready-made investigations. Each one chains several tools into a single finished answer with a concrete deliverable at the end — a ranked fix list, a budget thesis, a punch list you can ship this week.
What a playbook actually is
A structured prompt that already knows the tool surface. Without one, a model tends to call a single tool, summarise, and stop — which produces a shallow answer. A playbook tells it which tools to chain, in what order, and what the output has to contain before it is finished.
They reach your assistant as MCP resources, at clickcontext://playbooks/{id}. A client that supports resources can list them and load one when your question matches. You can also just name one:
Run the "Did my last theme deploy break anything?" playbook. I shipped on the 14th.
Plan gating
Four sampler playbooks are readable on every plan, including Free — they are marked Free below. The rest of the library unlocks on Growth. The list itself stays open on every plan, so your assistant can always see what exists.
The library
Find the money
Locate revenue that is leaking right now, and what it costs per day.
End-to-end funnel audit: finds the biggest revenue leaks, identifies why, and produces a ranked fix list.
Identifies sessions that came close to converting but didn't, finds the pattern, and proposes a recovery play.
Pins the worst funnel drop-off to a specific page element and converts it into a cost-per-day figure.
Ranks products by abandoned cart value (add-to-carts that didn't convert × price), then explains why the top one leaks.
Finds the element visitors stare at longest without clicking — the friction a click heatmap can't see — and whether it's blocking a high-intent moment.
Field-level form abandonment: completion rates with confidence intervals and the exact field visitors give up on.
Strategic levers
Reframes and budget decisions: where to invest, and what the data really means.
Compares new vs returning visitors on revenue per session and revenue share, then turns it into a budget reallocation thesis.
Plots conversion by number of products viewed per session, finds the non-linear jump, and quantifies the prize of moving the single-product pool up one step.
Ranks every page by attention-per-visit against the traffic it receives, surfacing high-attention pages being starved and low-engagement pages being over-fed.
Runs statistically-corrected conversion correlations and forces an explanation of any backwards-looking result instead of dismissing it as noise.
Cross-references the highest-value visitor cohort with unresolved JS errors, rage hotspots and funnel friction — and reframes the result as a retention risk.
The flagship report. Produces 5–7 findings that each require joining behaviour + commerce + traffic, ranked by revenue impact, each with a specific action.
Paid traffic
Stop wasting ad spend, and find out what the campaigns are actually doing.
Finds which ad campaigns waste budget by sending bouncing traffic vs. which ones drive engaged sessions.
Finds products that convert normally from organic but collapse from a paid campaign — an ad-to-page mismatch, not a bad product.
Health & regressions
Catch breakage and dead surfaces before they bleed a week of revenue.
Compares last 3 days vs. previous 14-day baseline. Catches regressions in errors, click patterns, and conversion before they bleed revenue for a week.
Detects high-traffic 'pages' that are actually broken or encoded URLs — intent-carrying sessions silently bouncing off dead ends.
Compares devices on time + frustration + revenue (not just conversion) to tell a real defect apart from naturally weaker traffic.
Core Web Vitals audit at p75 with Google's grading, tied to the pages where speed actually costs money.
Product & discovery
Catalogue, search and merchandising signals — demand you are not capturing.
Deep-dive on a single product page: traffic, intent signals, click frustration, layout issues, and a prioritized PDP fix list.
Surfaces unmet demand by analyzing site search queries that returned no results — direct signal of products customers want but you don't carry.
Crosses variant/swatch click demand with inventory and disabled-state markup to surface sold-out colours still being clicked and the highest-demand variant being fumbled.
Writing your own
Nothing stops you from writing a prompt that chains the tools yourself. The two things the built-in playbooks do that home-made prompts usually miss:
- They name the tool sequence.“Run the funnel segmented by device, then take the worst segment and pull the DOM snapshot and element clicks for the page where it drops.”
- They end with a concrete deliverable.“Give me a punch list of fixes I can ship this week, ranked by revenue impact.” Without that, models resolve to an open-ended summary.
Both patterns are worth stealing. See Getting good answers for the phrasing that steers hardest.