Start here
Claude stopped being a chat box
A year ago, Claude answered questions. You asked something, it replied, you pasted the output into a document or an email, and that was that. That is no longer the ceiling. Claude can now be handed work, not just questions — and the mechanism that makes the shift is skills and plugins.
What they give you is not speed. It is leverage. The reason most firms have not made this work yet is not the technology. It is the AI hype problem: generic tools, vendor noise, and demo-day showcases that fall apart on a real brief. Skills and plugins are the antidote. They are how you make AI work the way your firm actually works, rather than the way a vendor deck suggests it might.
Skills and plugins are how you make AI work the way your firm actually works, not the way a vendor demo suggests it might.
The onboarding pack
What a skill actually is
A skill is a reusable set of instructions that tells Claude how to do a specific task, every time, in exactly the way you want it done. Your process, your tone, your formatting, your guardrails — codified once and applied consistently.
Think of onboarding a new junior consultant. You would not hand them a blank document each morning and say “work out how we do proposals.” You would give them a template, an SOP, and some examples of what good looks like. A skill is that onboarding pack. The new joiner is Claude. The pack is a text file.
This is the core of the KWA framework — Knowledge, Workflow, Agents. Before you can automate anything reliably, you need your process codified at the Knowledge and Workflow layers. Skills are how you do that with Claude.
The anatomy
Three sections, one non-negotiable file
Every skill lives in a folder, and one file inside it is non-negotiable: SKILL.md, a plain markdown file with three sections.
- Metadata. The name and description. The description is the trigger — it is what Claude reads to decide whether the skill is relevant to the task in front of it. If yes, it loads the rest of the file; if no, it moves on. Claude does not load the entire skill just to check relevance.
- Workflow. The bulk of the file. It walks Claude through the task step by step. In a proposal skill: extract client name and firm size, identify the problem statement, structure the business impact, write the proposed approach, add deliverables and timelines, confirm pricing, include next steps. Three steps or fifteen — the point is that it is your process, written down, applied consistently.
- Constraints. The guardrails. British English only. No em dashes. Always include next steps. Maximum two pages. Whatever your firm’s standards are, this is where they live.
Two optional additions are worth knowing. A references folder holds static assets Claude pulls from during the task — brand guides, proposal templates, past deliverables, tone-of-voice documents. That is what separates generic AI output from output that looks like it came from your firm. A scripts folder holds Python scripts for tasks plain language cannot handle efficiently: calling an external API, processing a CSV, generating a formatted Word document.
Wired in
What connectors do
Connectors let Claude authenticate with the tools your firm already uses — Gmail, Google Calendar, Notion, HubSpot, and many others. Once connected, Claude can read from and write to those tools as part of a workflow. That is the move from talking about doing things to actually doing them inside your systems.
For a consultancy, the four connectors to set up first are CRM, calendar, email, and a project management tool. That covers roughly 80% of the workflows firms automate in practice.
From notes to proposal
Building a skill, step by step
A common problem in consulting: you take a discovery call, scribble notes, then spend three days assembling a proposal. By the time it lands, the prospect has spoken to someone else. A well-built skill takes those raw call notes and produces a structured proposal — problem statement, business impact, proposed approach, deliverables, timelines, pricing, next steps — in minutes rather than hours.
Build it reliably by following a six-step framework — Scope, Plan, Scaffold, Write, Test, Iterate — before touching any configuration.
- Scope. Forces you to be honest about what you actually do manually. Tell Claude “write me a proposal” without this and you get something generic. Scoping is what makes the output yours.
- Plan. Asks what Claude needs to complete the task. Reference files? Brand guidelines? A connector to your CRM? Identify the inputs before you build.
- Scaffold. Creates the folder structure and the
SKILL.mdfile. - Write. Fills in the workflow and constraints.
- Test. Runs sample inputs to verify the skill triggers correctly and produces the expected output.
- Iterate. Where most people stop too early. The difference between a skill that is fine and one that is genuinely reliable is usually two or three rounds. Use it on a few real proposals and you notice things — the executive summary is too long, the pricing section needs a different format, the approach should pull from your case studies. You ask Claude to update the skill, and it does.
The skill does not replace your judgement. You still read the proposal, adjust pricing for this client, decide whether the approach is right. What it replaces is the three hours of formatting, structuring, and brand-checking that nobody should be doing at 11pm.
The maths is straightforward. If a skill saves one partner three hours per proposal, and the firm produces ten proposals a month, that is 30 hours recovered at £150 per hour — £4,500 per month, per partner, from one skill.
The bundle
What a plugin is, and why it matters
A skill does one job. Real consultancy work is a chain: discovery leads to proposal, proposal leads to onboarding, onboarding leads to weekly status updates. Running each link manually — even with individual skills — means you are still doing one thing at a time. A plugin is a bundle: multiple skills, multiple connectors, packaged into what is effectively a job role.
Individual workstreams — discovery, analysis, stakeholder interviews, benchmarking — each have a clear scope and output. That is a skill. An engagement manager does not run every workstream themselves; they know which one runs first, which feeds into which, and when to pull in the right resource. A plugin is the engagement manager.
The anatomy mirrors a skill. One required file at the top — plugin.json — holds the metadata. Then optional folders: skills for repeatable tasks, slash commands for shortcuts, sub-agents for heavier background work, hooks for automatic triggers, and connectors so tools are wired in from the moment someone installs it.
There are five practical reasons to bundle rather than keep loose skills:
- Distribution. Install once and your whole team gets the same toolkit. Loose skills mean someone always has an old version.
- Name spacing. Run three client plugins, each with its own proposal skill, and they do not collide.
- Versioning. Update centrally and everyone gets the update. No chasing teammates to refresh files.
- Productionisation. A plugin is something you can hand to a client at the end of an engagement — a leave-behind. Skills are personal tooling; plugins are assets.
- Trust. Plugins from Anthropic’s official marketplace carry a verified badge, which matters when you recommend tooling to clients who care about governance.
Something you can ship
A consulting ops plugin in practice
Here is a full client-lifecycle plugin — six skills, one bundle:
- Discovery. Turns raw call notes into a structured summary.
- Proposal. Turns discovery notes into a full proposal with pricing and next steps.
- Onboarding. Produces a welcome email, project charter, and kick-off agenda for a new engagement.
- Summary. Publishes meeting notes into a client-facing follow-up with decisions and actions.
- Status. Generates weekly progress-against-plan updates.
- Case Study. Turns a completed project into a case study.
All six connected to Gmail, HubSpot, Notion, and Google Calendar. Once built and tested, you export the plugin and share it with your team, so everyone operates to the same standard. You can hand it to a junior consultant on day one, give it to a client as a leave-behind, or package it as a paid asset for a specific sector — accounting firms, law firms, marketing agencies — using the same skeleton with different references and tone.
A skill is a tool you use. A plugin is something you can ship.
Where to start
Foundations before tools
Most firms have more automatable workflows than they realise, and fewer AI-ready processes than they need. The 5 Steps for AI Leadership framework frames the gap: firms rushing into tools and automation have usually skipped Align (a written AI position the team can act on) and Activate (the team actually using AI in daily work). Without those foundations, even well-built skills do not stick.
The right starting point is a clear-eyed look at where your senior time actually goes, which of those tasks follow a repeatable process, and which processes are documented well enough to automate.
The goal is not AI for its own sake. It is to codify how your firm works so the leverage compounds — a skill you use, a plugin you ship, and a standard your whole team can operate to without you in the room.