The portability problem
Your best workflows live in one person’s head
Your best consultants write brilliant proposals. They nail discovery call documentation. They know which pain points to surface and how to frame the pricing. Clients love the output.
Then they go on leave. Someone else picks up the pitch and the quality drops — not because the second person is less capable, but because the knowledge was never written down. It lived in one head, shaped by months of experience nobody had time to document.
This is the portability problem, and it is one of the most reliable ways a consultancy stalls as it tries to scale. Claude Skills are how you fix it.
What they are
Reusable instruction files that travel with you
Skills are reusable instruction files. They tell Claude how to execute a specific task: what tools to use, what standards to follow, what format to deliver.
Once created, a skill is available everywhere — in a web chat, in Claude Code, via API. Any team member can trigger it. Any engagement can use it. The output is consistent whether it is your best consultant or the newest hire running it.
Before Skills, the workaround was copy-pasting instructions into every new chat, or building dedicated Claude Projects with custom instructions baked in. Both have the same flaw: the knowledge is trapped in a specific project or chat. You cannot combine workflows across engagements, and you cannot share them across your team without everyone replicating the setup by hand.
Skills extract the workflow out of the project and make it portable.
Where it fits
Skills, Projects and MCP are building blocks, not rivals
People conflate Skills with Projects or MCP tools. They are not the same thing, and understanding how they fit together matters for building something that actually scales.
- Claude is the reasoning layer — the intelligence making decisions.
- Skills are the instruction layer — reusable, portable workflows for how specific tasks should be done.
- MCP tools are the capability layer — connections to external systems like HubSpot, Notion, or your CRM.
- Projects are workspaces — where you combine skills, MCP tools, and client-specific context for ongoing engagements.
None of these competes with the others. The KWA framework maps directly onto them: Knowledge sits in your Projects and background context; Workflow is encoded in your Skills; Agents operate using MCP tools to take real action in production systems. You mix and match based on what a specific engagement needs.
Three flavours
Know which skills to enable and which to avoid
- Official skills (built by Anthropic) — enabled through Settings under Capabilities. The most relevant for professional services is the document creation skill. Do not enable everything at once: loading too many skills at the same time degrades consistency. Enable only the ones your team actually uses.
- Custom (organisation) skills — these are yours. Your workflow instructions, your templates, your delivery standards. You create a skill file and upload it. This is how you standardise Claude to your way of working, not a generic AI template.
- Community skills — created by other users and available online. Treat these with caution. You cannot verify what instructions they contain or how they will behave inside your workflows.
One critical note: once you enable Skills, add a line to your account-level custom instructions telling Claude to always consider relevant skills. Without this, skill triggering is inconsistent.
Build these first
Three skills worth encoding now
1. Branded proposal generation. If your firm runs workshops, you have a standard proposal template — pricing tiers, brand structure, delivery summary, terms. Build a branded proposals skill instead of pasting the template into every chat. Upload your template, instruct Claude on when to trigger it, what client context matters, and which sections are required every time. Once it is live, any team member can upload discovery call notes and generate a branded proposal: pricing applies, format is consistent, no version drift. This is the Amplify step of the 5 Steps for AI Leadership in action — a win sitting with one person becomes available across the whole team.
2. Discovery call documentation. Discovery quality is uneven at most consultancies, with outcomes stuck in one consultant’s head, rough notes, or a half-completed CRM record. Build a discovery documentation skill that captures your workflow: which sections to cover, what format the summary follows, what gets emailed to the client versus logged in HubSpot. If you use a HubSpot MCP server, include only the specific tools the skill should call — not the entire list. After every call, paste the transcript and the skill generates the summary, drafts the follow-up, and logs the activity. Every consultant follows the same process.
3. YouTube script generation with ICP alignment. This one comes from our own workflow. Every video has to hit the same pain points, follow the same structure, open with the same kind of hook, and close with the same CTA. That knowledge previously lived in a single Claude Project tied to our ICP definition. As a portable skill, any team member — writer, producer, guest contributor — can generate a script that follows the right structure without rebuilding context. The ICP definition, the hook framework, and the CTA logic travel with the skill. The same principle applies to any content you produce at volume: client briefings, bid documents, sector reports.
Before you build
A three-question test, and the warnings that come with it
Before investing time in a skill, ask three questions:
- Do you repeat the same instructions across different chats three or more times?
- Would you need to train a real new hire on this process when onboarding them?
- Does quality or format consistency matter every time this task is done?
If two or three are true, build the skill and share it. If the task is a one-off or the format does not matter, keep it in a project or a simple chat.
A few warnings worth heeding:
- Usage limits matter on Pro plans. Only build skills for your most repeated workflows. Do not burn context tokens on tasks that happen once a quarter.
- Skills are static files. If you reference MCP tool functions, those tools may change or deprecate over time. Always include fallback instructions for when a tool call fails.
- Keep MCP tool lists short. Loading too many tools into a single skill degrades consistency. If you need more than five or six tools, split the skill.
- OpenAI has announced Skills for ChatGPT. The portability pattern is not specific to Claude — but the quality of what you build is specific to your firm. That is the competitive advantage, not the platform.
Why it matters
The villain is trapped knowledge, not missing tools
The villain here is not a lack of AI tools. Most consultancies already have access to capable AI. The problem is workflow knowledge trapped in isolated projects and individual people, unable to compound across the team or replicate consistently from one engagement to the next.
Generic ChatGPT prompts, disconnected Claude Projects, instructions rebuilt every time someone new joins an engagement — none of these scale, and none survive a senior consultant going on leave.
Skills fix the portability problem. Your best workflow logic becomes a reusable asset: available to every team member, consistent across every engagement, deployable without rebuilding context from scratch. This is not about replacing how your consultants think. It is about encoding what they know so the whole team can work at the same level.