How to build a Claude Skill for competitor analysis

What a Claude Skill is, and why it beats re-pasting a prompt

Anthropic describes a Skill as a folder of instructions, scripts, and resources that Claude loads when a task calls for them. The one required file is SKILL.md. It starts with a short YAML header holding two fields, a name and a description, and the rest of the file is ordinary Markdown instructions. Nothing about it requires code. For competitor analysis, the instructions are the eight headings and the sourcing rules you would otherwise paste at the top of every chat.

The difference from a saved prompt comes down to how Claude loads it. At the start of a conversation, Claude sees only the name and description of each Skill you have turned on. When your request matches a description, Claude reads the full SKILL.md and follows it. Anthropic calls this progressive disclosure. In practice it means you type "analyze Acme Analytics" and the whole structure arrives without you pasting anything, and every run starts from the same version of the instructions. A pasted prompt drifts: you shorten it one day, forget a rule the next, and a month later your five competitor reports no longer line up.

Skills also work in more than one place. In the Claude apps you upload one as a zip file. In Claude Code, the same SKILL.md sits in a folder on your disk and you can call it with a slash command. Developers can upload one through the Skills API. These copies do not sync, so a Skill added in the Claude app has to be added again in Claude Code or the API.

The guide to competitor analysis with Claude covers the one-off prompt and Research mode. This guide covers turning that prompt into a Skill and living with it afterward.

What the Skill gets right, and where it hits a ceiling

The Skill fixes the two problems that make chat-based competitor research hard to reuse. Structure drift shrinks, because the headings and their order are written down once and Claude reads the same text every time. Sourcing improves, because the rule about putting the URL in the same sentence as the claim is always present instead of remembered on good days.

Structure drift shrinks but does not vanish. Claude still decides how long each section runs and whether a section uses a table or bullets, and two runs on the same competitor a week apart will differ in wording. The Skill gets you the eight headings in the right order with the right rules. Lining up five reports side by side still takes a light pass from you.

The Skill does not verify facts. It tells Claude to write "no public source found" instead of guessing, which makes gaps visible, but a source URL next to a wrong number is still a wrong number. Everything you plan to act on still gets checked against the competitor's live pages.

The Skill also does not watch competitors over time. It produces a snapshot on the day you run it. Keeping that snapshot current is a separate decision, covered further down.

The method, step by step

1. Write the SKILL.md

Copy the text above into a plain-text editor and save it as SKILL.md inside a folder named competitor-analysis. The folder name should match the name field. A few rules keep the header valid. The name must use lowercase letters, numbers, and hyphens, up to 64 characters, and it cannot contain the words "anthropic" or "claude". The description must say what the Skill does and when Claude should use it, in the third person, because Claude matches your request against it. Anthropic's help center lists a 200-character limit for the description in the Claude apps and the developer docs allow up to 1,024, so a description under 200 characters works everywhere. Do not add other header fields unless you know the surface accepts them. Claude Code accepts extra fields, while the Claude apps and the API reject a file that contains fields outside the shared spec.

Edit the body freely. If your team cares about integrations or security certifications, add a line under Product & Features asking for them. Keep the body short. Anthropic's authoring guidance says Claude already knows what a SWOT is, so spend your words on your rules and not on the concepts.

2. Add it in Claude

Zip the folder so that competitor-analysis is the root of the archive and not a subfolder. In the Claude apps, open your settings, go to Customize, then Skills, and upload the zip. Skills depend on code execution being turned on, which lives under Capabilities in the same settings area on individual plans. On Team and Enterprise plans an owner has to enable Skills for the organization first. The help center currently lists Skills as available on every plan including Free, while Anthropic's developer docs describe custom uploads as a paid-plan feature, so check Anthropic's help center for the current plan list and the current settings path if you do not see the option.

If you use Claude Code, skip the zip. Put the folder at ~/.claude/skills/competitor-analysis/ and it applies in every project, or at .claude/skills/competitor-analysis/ inside one repository. You can then type /competitor-analysis followed by the competitor name.

Test it before relying on it. Open a fresh chat and type a request that should trigger the Skill, such as "run a competitor analysis on Acme Analytics". If Claude answers without the eight headings, tell it "use my competitor-analysis skill" for that run, then sharpen the description so the next request triggers it on its own.

3. Keep your product context in a Project

A Skill holds procedure, and it works everywhere across Claude. A Project holds background knowledge that loads in every chat inside that Project. Put your product description, your positioning, your current pricing, and the list of competitors you track in a Project's knowledge and instructions. Then run the Skill inside that Project. Claude reads the competitor against your product without you re-pasting who you are, and the Skill supplies the report format.

Do not put your product details in the Skill itself. A Skill is meant to be reusable procedure, and you may want to hand it to a colleague who works on a different product.

4. Run one competitor per chat

Start a new chat for each competitor, inside the Project. Chats in a Project share the knowledge base but not each other's history, so running two competitors in one chat invites Claude to blur their pricing or features together. Make sure web search is on, and use Research mode if your plan includes it, because the Skill tells Claude what to write and web search is how it finds current facts. Name the competitor and its URL in the first message, and say what you sell in one line if the Project does not already say it.

5. Verify what you will act on

Read the report against the checklist below. Start with the Pricing & Sales section and anything dated in the last few months, because those change fastest and they are usually the parts you act on. Open the competitor's pricing page in another tab and compare plan by plan. Open the source behind any number you plan to repeat to your team. Resolve every "no public source found" line by finding the fact elsewhere or by accepting it as unknown.

A worked example: Acme Analytics

Acme Analytics is a fictional company, used here to show what a good section looks like. Nothing below describes a real product.

You open a new chat inside your Project and type: "Run a competitor analysis on Acme Analytics, acme-analytics.example." Claude matches the request to the Skill's description, loads the SKILL.md, and works through the eight headings. A well-formed Pricing & Sales section reads like this (illustration only):

Acme Analytics lists three plans on acme-analytics.example/pricing. Starter is priced per seat per month with annual billing only, and the page notes a three-seat minimum (acme-analytics.example/pricing). Growth adds unlimited dashboards and is labeled "starts at" on the same page, which points to usage-based pricing above a base fee (acme-analytics.example/pricing). Enterprise has no public price and sends visitors to a sales form (acme-analytics.example/contact). Whether a free trial exists: no public source found. Acme sells self-serve on Starter and Growth and through sales on Enterprise (acme-analytics.example/pricing).

The parts to notice: every sentence carries its source, the "starts at" wording is quoted rather than turned into a number, the seat minimum is called out, and the missing trial fact is marked rather than filled in. At the bottom of the full report you would find the research date and a "Likely to go stale first" list, which for a company like this would name the Growth "starts at" pricing and any feature marked as announced.

Before you trust the output: 9 checks

1. Prices match the competitor's live pricing page, plan by plan, including billing period, seat minimums, and usage limits.
2. Every factual claim has a source URL in its own sentence, and you have opened the sources behind anything you will act on.
3. Every "no public source found" line is resolved or consciously accepted as unknown.
4. Messaging quotes match the competitor's site word for word.
5. Features are split into live and announced, and nothing announced has crept into the live list.
6. Recent moves each carry a date you can check.
7. Company numbers (funding, team size, customers) each have a dated source.
8. Sentiment quotes link to reviews you can open, and not to summaries of reviews.
9. The research date and the "Likely to go stale first" list are present, so you can tell how old the report is and what to re-check first.

If a report fails check 1 or 2, fix the Skill before you fix the report. A missing rule in SKILL.md will fail the same way on the next competitor.

Keeping the analysis current

Everything the Skill produces is a snapshot. Competitors keep moving after you file it, and the worst way to learn about a rival's price change is from a customer. You have three honest options.

Do it by hand. Set a recurring calendar reminder to re-run the Skill on each competitor and compare the new report with the old one. The fixed headings make the comparison easier than it would be with free-form output. This costs nothing and works as long as you keep doing it.

Build it in Claude. Claude Cowork, Anthropic's tool for delegating work to Claude, has scheduled tasks on paid plans. You describe a task once, pick a cadence (hourly, daily, weekly, weekdays, or on demand), and it runs remotely even when your computer is closed. Anthropic's help center says scheduled tasks have access to your Skills, so a task that says "run my competitor-analysis skill on these five companies and list what changed since last week" is a reasonable thing to set up. You still own the setup: which competitors, how often, what counts as a change, and how you receive the result, since runs collect in Cowork's Scheduled view unless you build a delivery step.

Buy it. A monitoring product does the re-checking on a schedule and sends you what changed. You trade flexibility for not maintaining anything. That option is where our own product comes in.

Where Competely fits, and where it does not

Competely runs the method above as a product. You enter your product and your competitors, and it analyzes each one across the same eight dimensions the Skill uses, more than 100 data points in total, with a source link on every one. An analysis takes about 30 minutes, and because the structure never varies, the reports line up side by side without cleanup. Monitoring is included on every paid plan. On a schedule you set, every two weeks or monthly on any plan and weekly on Pro and Scale, it re-reads each competitor's public web pages, flags what changed, and emails you a short brief. You can export to CSV or PDF and share a report by link with as many viewers as you like.

The limits are just as concrete. Competely watches public web pages only. There is no social listening, no press tracking, no win-loss interviews, and no CRM integration. A launch announced on LinkedIn will not reach your brief until it reaches the competitor's website. There is no API and no native Slack app, though Slack works through a channel email address. There is no free trial of the full product, but you can create a free account with no credit card and run a real analysis of your product against two competitors before you pay. Depth, exports, and monitoring stay on the paid plans. Full sample analyses are also open to anyone with no account, such as Figma vs Adobe XD. Plans are self-serve, from $39/mo, billed monthly or yearly, and you can cancel anytime.

If the Skill plus a calendar reminder or a Cowork task covers you, you do not need Competely. What it sells is never doing the re-checking yourself. The competitor monitoring page describes what the briefs contain.