Claude skills for solo consultants: what they are, and the first three to build

A saved prompt has to be found before it can be used. Discovery debrief, meeting summary, weekly client update, scope draft — each one lives somewhere, gets pasted, gets three lines customized, and gets pasted again the following week. The prompt is fine. The retrieval is the tax, and it is charged every time.

This is what Claude skills replace.

A skill is a folder holding a Markdown file named SKILL.md. It tells Claude how to behave when a particular situation comes up, and Claude loads it when the wording matches what the skill says it handles. Describe the situation — "I just had a discovery call with a prospect" — and the instructions are already in play. No paste, no customize, no hunting. Three skills cover most of a consulting week.

This article covers what a skill actually is, why it beats a saved prompt for high-frequency consulting work, the three we'd build first from a standing start, and — the part most write-ups skip — how to install one on the surface you actually use.

What a Claude skill is (and what it isn't)

The simplest definition: a skill is a Markdown file with a frontmatter block describing when to activate, followed by the instructions Claude should follow when active. It lives in ~/.claude/skills/ (Claude Desktop, Claude Code) and is auto-loaded by Claude on every conversation. When a trigger phrase in the frontmatter matches what you've typed, the skill's instructions take over.

Not a prompt you paste in. Not a Custom GPT, although the concept is adjacent. Not a Claude Project (which is closer to a long-running conversation with shared context). A skill is closer to an installed plugin — Claude knows it exists, knows when to use it, and pulls it in automatically.

A minimal skill file looks like this:

---
name: discovery-debrief
description: Turns raw discovery-call notes into a structured one-page
  debrief. Triggers when the user mentions "just had a discovery call",
  "discovery notes", or describes a call they need to write up.
---

# Discovery Debrief

You are helping a solo consultant produce a structured debrief from
raw call notes. When the user pastes notes, return exactly six sections:

1. Who & what they do
2. The real problem
3. Budget signal
4. Decision process
5. Red flags
6. Recommended next step

Voice: sharp and direct, peer-to-peer. No corporate-speak. Don't
soften red flags.

That's a complete, working skill. Frontmatter tells Claude when to activate. The body tells Claude what to do. The whole thing is under 30 lines.

Skills can also be more elaborate — routing between multiple prompts depending on context, asking clarifying questions before running, enforcing specific output structures. The Proposal-Closer Pack ships a single skill that routes between eight different prompts in the same workflow (discovery → ICP check → scope → pricing → draft → objections → follow-up → debrief). Same file shape; more elaborate routing.

Why solo consultants benefit specifically

Most "AI productivity" advice is written for someone whose work is varied — a marketer doing 40 different things, a founder context-switching every hour. Solo consultants aren't that. The actual shape of consulting work, especially after the first 18 months, is a small number of high-frequency workflows, repeated with new clients. Discovery → proposal → kickoff → weekly update → wrap → case study. The variety lives inside each step (different clients, different situations) but the shape of each step repeats.

That's the workload skills were built for. Three reasons:

Repetition compounds. A skill you run 40 times a year — every discovery debrief, every weekly update — gets better the more you use it because you tune it once and the tune sticks. A saved prompt you paste in every time also gets better, but only if you remember to update the saved version. Most consultants don't. Their saved prompts drift; their skills don't.

Voice consistency. Skills bake your voice rules into the instructions. "Don't soften red flags." "Don't pad." Every output produced by the skill carries the same voice, which means your client-facing artifacts (briefs, status updates, debriefs) start to read like they came from the same person. They did — Claude is just the production engine.

Hands-off triggering. The biggest skill win is the auto-activation. When you've just wrapped a call and your context is fresh, you don't want to hunt for the right prompt. You want to type one sentence — "just had a discovery call with someone at a healthcare SaaS" — and have the skill load itself. That's the difference between using Claude as a tool and Claude operating as part of your workflow.

The first three to build

Starting from scratch — no skills installed, no prompts saved — these are the three we'd build first, in this order.

Skill #1: Discovery debrief

The highest-frequency post-call task for any consultant doing inbound work. You have a 30-minute discovery call on Tuesday; you need a structured read by Wednesday so you can decide whether to write a proposal. Without a skill, you stare at messy notes Thursday morning and try to reconstruct the call from memory.

The trigger phrases should include: "just had a discovery call", "discovery notes", "just got off a call with a prospect". The output structure should be six sections — who/what, real problem, budget signal, decision process, red flags, next step. The voice rules should explicitly forbid softening red flags, which is the failure mode most discovery debriefs share when generated by AI.

If you'd rather not build it from scratch, we ship this one as a free standalone download — the Markdown file, the Claude skill in skill/discovery-debrief/SKILL.md, and an Acme Widgets worked example. Drop it in ~/.claude/skills/ and it auto-activates the next time you wrap a call. Treat it as the reference implementation; modify the voice rules to your own voice if it sounds too much like ours.

Skill #2: Meeting debrief (not the same as discovery)

The discovery debrief is about whether to take a deal. The meeting debrief is what you do during an engagement — after a Tuesday check-in, a stakeholder interview, an internal strategy session. The shape is different: less "is this worth proposing for" and more "what was decided, what's owed by whom, what's at risk."

Trigger phrases: "just finished a client call", "meeting transcript to actions", "need a debrief from this". Output structure: decisions made, action items with owners, open questions, risks raised. The voice rules should forbid the model from inventing action items that weren't in the transcript — the failure mode here is fabrication, not softening.

This skill pays back the most when you have it triggered straight from a transcript. Paste a 30-minute meeting transcript; get back a 1-page artifact you can email to the client an hour later. That email is also the thing that makes you look organized in the engagement; clients notice when the follow-up arrives same-day with structure.

The article "The client-meeting workflow that turns transcripts into next actions" walks the prompt structure if you want a starting point.

Skill #3: Weekly client update

The Friday 5 p.m. status email is the most slipped-on weekly artifact in solo consulting. You're tired, the week is closing, you write a status update from memory and ship it with three typos. A skill turns this into a 2-minute task.

Trigger phrases: "weekly update for [client]", "Friday status email". The output structure should be five sections — what shipped this week, what's in progress, what's blocked, what's coming next week, one open question for the client. The voice rules should force the skill to flag anything that's drifting (a deliverable that's been "in progress" for three weeks running), not just list it.

Inputs you feed it: your week's notes, your task tracker, your gut read on what's slipping. Output: a five-section email the client opens once on Friday and refers back to once on Monday. A skill and a procedure are not the same thing, and the difference matters here: the skill is how the artifact gets produced, while the procedure is what fires it at the right cadence. A skill nobody remembers to invoke on Friday afternoon has not solved the Friday afternoon problem.

How to install (Claude Desktop and Claude Code)

There are two installation paths, and which one applies depends on where the work happens. They are genuinely different mechanisms, not two names for the same thing.

In the Claude app — no terminal. Go to Customize → Skills, click +, choose Create skill, then Upload a skill, and hand it a .zip. The zip has to contain the skill folder, not the folder's contents, and the folder name has to match the name in the frontmatter. Enable code execution in Settings first, or the upload option will not do anything. This is also how skills reach Cowork — Cowork loads whatever is enabled on your account, so one upload covers both.

In Claude Code — a folder on disk. ~/.claude/skills/<skill-name>/SKILL.md makes a skill available in every project. .claude/skills/<skill-name>/SKILL.md inside a project folder scopes it to that project, which is the better choice for anything client-specific. Claude Code watches those directories and picks up changes without a restart.

Three things worth knowing before you spend an evening on this:

• Skills do not sync between the two. A skill uploaded in the app is not on your filesystem, and a skill in ~/.claude/skills/ is invisible to the app and to Cowork. Install it wherever you actually work — or in both, deliberately.
• A skill built for Claude Code can fail to upload. Claude Code accepts frontmatter fields the portable format does not, and an upload rejects unexpected keys outright rather than ignoring them. For anything you intend to move between surfaces, keep the frontmatter to name and description.
• Downloading a repository from GitHub does not give you an installable skill. The archive wraps everything in a top-level folder and usually holds many skills. Getting one out means pulling the single skill folder and zipping that on its own.

Once a skill is in place, Claude uses it when your wording matches what the skill's description says it handles — which is why that field is worth more care than the instructions underneath it. If Claude is not picking it up, the description is the thing to change.

Mechanics verified 2026-08-29 against Anthropic's documentation — the Agent Skills specification, Claude Code skills, and using skills in Claude. Anthropic's own pages currently disagree on which plans include skill uploads, so check yours rather than trusting a number here. Installation details change with releases; this note is dated so you can tell how stale it is.

Pitfalls (the ones that bite)

Writing skills that are too generic. A skill that says "help the user draft a client email" loads on every email-related conversation and over-broadens what Claude does. Specific triggers + specific output structures beat broad ones every time. "Weekly Friday status email, 5-section format" loads only when relevant and produces a consistent artifact.

Over-engineering the first version. A skill doesn't need to handle every edge case. Build the 80% case, run it for two weeks, and tune the version-two based on actual failures. The temptation is to spec every possible scenario upfront; the result is a skill so detailed it never gets shipped.

Forgetting that the skill's voice is your client-facing voice. A skill that produces overly-formal output ships overly-formal artifacts to your client. If your real-life voice is sharper than the default Claude voice, encode that in the skill. "Use em-dashes. Don't open with 'I hope this finds you well.' Don't moralize."

Skipping the trigger phrases. Some consultants build skill files without any trigger language in the description and then wonder why Claude doesn't auto-activate them. The trigger phrases are the most important part of the frontmatter — they're what makes the skill self-loading instead of a saved prompt you still have to remember.

Building only one. A single skill is a curiosity. Three skills is a workflow. Once you have a discovery debrief, a meeting debrief, and a weekly update skill all installed, the way you use Claude during a week changes shape — most of your repeated work runs through skills, and the conversations you have with Claude become about the work that isn't repeatable.

Where to go deeper

A skill carries instructions. What it does not carry is the definition of what finished work looks like — and that is the half that decides whether the output is sendable. A skill that says "write the client update" still needs someone to have decided what a good one contains, in what order, under what word cap, for that particular client. The Claude Workspace Kit is that half written down: four written standards for finished deliverables, six ready-to-run engagement briefs, and context files — ICP, offers, rate floor, voice — in a folder Claude reads without being told. It ships filled in for a fictional practice, including a client folder holding a week of messy notes, so the shape is visible before you build your own.

If you want a worked example of a skill that routes between multiple prompts in the same workflow — closer to the elaborate end of the spectrum — the Proposal-Closer Prompt Pack ships a single skill (skill/proposal-closer/SKILL.md) that routes between eight prompts depending on which stage of a proposal cycle the consultant is in. Same file shape as the minimal skill above; more elaborate logic. It's the reference implementation for what a chained-workflow skill looks like.

The smallest version of all of this is one skill, installed today, on the workflow you run most often this week. Pick the one that maps to your calendar between now and Friday. Build it tonight. Use it on Tuesday's call.