AI agents (MCP)

Playbook skills

The 22 playbooks the AmICited MCP server ships as skill resources: which tools to call, in which order, and the traps that make a confident answer wrong.

Open in ChatGPT Open in Claude

Tools are what an agent can reach. Skills are how to use them for a real question. With over 300 tools, the hard part is rarely finding one: it is knowing that “why did traffic drop” needs six reports in a particular order, or that an empty ecommerce report can mean “not connected” rather than “zero revenue”.

The AmICited MCP server ships these playbooks as MCP resources. Each one was written by giving agents hard questions against the live server with no playbook, logging every dead end and wrong assumption, and turning those failures into instructions. Every tool name a skill mentions is checked against the live tool registry on each build, so a skill never points at a tool that no longer exists.

How agents find them#

Skills are served at skill://<name>/SKILL.md. Listing them costs a name and one line each; the body is fetched only when an agent wants it.

The server’s instructions, sent to every client on connect, tell the agent to read skill://using-amicited/SKILL.md first. That skill explains the find, inspect, run loop and lists the others.
Many clients never list resources on their own. If your agent seems unaware of a skill, name it: “Read skill://ai-visibility-audit/SKILL.md, then audit example.com.”
Skills are read-only text. Reading one does not count toward the hourly tool-call cap.

Conventions#

Skill	Purpose
`using-amicited`	How to drive the endpoint: finding the right tool among 300+, invoking it, and the conventions every other skill assumes. Read first.
`report-windows`	Date windows and comparison periods across reports: the different argument styles, the equal-length rule, and how to avoid confidently wrong numbers.

AI visibility and competitors#

Skill	Purpose
`ai-visibility-audit`	Diagnose why a brand is or is not cited in AI answers: which engines actually ran, what the headline number hides, and four checks that explain a gap.
`competitor-gap-analysis`	Find where competitors win AI answers and search that you do not, and which gaps to attack first.
`competitor-content-gaps`	Find competitor sitemap sections, vocabulary and publishing gaps without overstating incomplete freshness data.
`ai-fanout-queries`	Find the searches an AI assistant ran against your site (the ones carrying a `site:` operator) and read them without mistaking zero click-through for a problem.
`prompt-set-design`	Add tracked prompts for a domain: what a good prompt set looks like, the tagging trap, and why nothing here can tell you the credit cost.

Organic search diagnostics#

Skill	Purpose
`traffic-drop-triage`	Work out why organic traffic moved: localise it to a query or page, date the step, and rule out indexing, outages and site changes before blaming rankings.
`did-our-change-work`	Correlate a site change with what happened after, and why an empty annotation history is not evidence of anything.
`indexing-triage`	Investigate indexing problems: which tool answers which half, why a null verdict is not a no, and the 404-without-redirect pattern.
`cannibalization-cleanup`	Find pages competing for the same query, and why the top of the severity ranking is usually brand noise.
`page-families`	Read a page and its translations as one family, and tie a traffic change to an hreflang or canonical edit.
`site-audit-review`	Price, run and interpret sitemap site audits to fix technical issues and improve internal link equity.
`backlinks-review`	Judge whether monitored backlinks still carry value: what “active” hides, lost versus unreachable, and comparing authority with competitors.

Money#

Skill	Purpose
`profitability-deep-dive`	Report ecommerce profit honestly: check cost coverage before quoting any margin, because an unconfigured cost reads as zero.
`customer-value`	Lifetime value, cohorts and repeat purchase, and why the window has to be long.
`paid-waste-review`	Find wasted ad spend without inventing profit, including when conversion value is not actually money.
`seo-to-revenue`	Argue organic work in money rather than clicks, after checking the channel data to attribute revenue exists.

Operations and deliverables#

Skill	Purpose
`platform-onboarding`	Connect a data platform to a domain and import its history: the quote-then-confirm chain and where money is committed.
`uptime-setup`	Set up monitors, alerting, heartbeats, maintenance windows and a status page without half-configured defaults.
`availability-review`	Answer “was the site up, what broke, and is anything down now”, including the SLA report’s calendar-month rules.
`recurring-reports`	Produce a weekly SEO or monthly ecommerce report that is comparable to the last one: fixed windows, fixed tools, caveats carried forward.

What a skill changes in practice#

Two examples of what the playbooks catch that the tool schemas cannot:

The provider breakdown hides silent engines. The dashboard’s by-platform breakdown only lists engines that cited you. ai-visibility-audit sends the agent to get_prompt_detail, which shows for each engine whether it ran and whether it cited you, so “24% visibility” becomes “cited by Perplexity, ignored by ChatGPT”. See missing engine data.
Unconnected is not zero. using-amicited makes the agent check whether a platform is connected before reporting that it has no data, and link you to Data Sources when it is not.

Name the skill in your request when you know the question: “Use the traffic-drop-triage skill to explain last week’s drop on example.com.” The agent reads the playbook first and follows its order. The recipes page has ready-made prompts.

Recipes Copy-paste prompts that use these skills. Toolsets and tools The find, inspect, run loop the skills assume.