The skill does more than summarise competitor websites. It enforces evidence discipline: every factual claim carries a URL or an explicit uncertainty flag, every inference is labelled, and the run stops at a calibration gate so a wrong fact is caught before it corrupts the strategy.

Use it when you want the market landscape and you own the roadmap that has to change because of it. It ends in a comparison matrix, a white space map, and seven explicit decisions: accelerate/pause/kill, build vs partner vs ignore, category positioning, pricing angle, hiring signals, wedge, board narrative.

What it actually does

A phased flow with mandatory confirmation gates:

Step 0 Industry metrics -> confirmed with you before any search Step 1 Your hypothesis + set -> confirmed before analysis begins Phase 2 Deep competitor cards -> 12 sourced dimensions each Calibration gate -> you correct the facts before synthesis Phase 3 Matrix -> white space -> seven decisions

The gates carry the design. Catching a wrong fact at the gate costs one message; catching it after synthesis invalidates everything downstream.

Each competitor card covers value prop, real product strengths, weaknesses drawn from reviews rather than marketing, missed jobs-to-be-done, product velocity, category narrative, pricing and what the pricing model optimises for, GTM motion, strategic trajectory, customer voice, and a four-part moat assessment (network effects, switching costs, scale economies, intangible assets) ending in: what would it realistically take to unseat them?

Install

Project scope, available in one repository

git clone https://github.com/AIPMProfile/competitive-analysis-skill.git mkdir -p .claude/skills cp -r competitive-analysis-skill/skills/* .claude/skills/

Personal scope, available in every project

git clone https://github.com/AIPMProfile/competitive-analysis-skill.git mkdir -p ~/.claude/skills cp -r competitive-analysis-skill/skills/* ~/.claude/skills/

Desktop and web app

Download the .skill file from Releases and upload it under Settings > Capabilities > Skills. Or build it yourself:

./scripts/build.sh # writes dist/*.skill

Agent SDK

Point the SDK's skill directory at skills/, or copy the directory into your agent's configured skills path.

Use

The skill is model-invoked. Describe the task and it loads:

"Run a competitive analysis on Linear for a project management tool aimed at design agencies."

"I own product at a healthcare AI startup. Teardown of Abridge, Nuance DAX and Suki, then tell me what to build next quarter."

You can also invoke it explicitly by name:

"Use the competitive-analysis skill on our top three competitors."

Expect two questions before it searches: the industry metrics and your existing hypothesis. Answering them properly is what makes the output specific to you rather than generic.

Quick mode

Ask for a "quick competitive scan" and the flow collapses to metrics, competitor set, short cards, and takeaways, without the full gate sequence. The sourcing rules still apply.

Repository layout

skills/ └── competitive-analysis/ ├── SKILL.md core flow, loaded every time └── references/ ├── industry-metrics.md 12 industries + what to do when unlisted ├── search-playbook.md queries, budget, source tiering, dead ends ├── pm-decisions.md what each of the seven decisions requires └── example-output.md worked example at the expected standard scripts/ ├── build.sh package skills into dist/*.skill └── validate.py structure, frontmatter, reference checks

SKILL.md stays lean and loads on every invocation; the references/ files are read only when that phase is reached.

Contributing

python3 scripts/validate.py # must pass before a PR ./scripts/build.sh # optional, produces dist/

The validator checks frontmatter, name uniqueness, name/directory agreement, description length and trigger phrasing, SKILL.md length, and that every references/... path mentioned in SKILL.md actually exists.

Two rules for changes to skill content:

No claim without a source. If you add a dimension to a competitor card, say where the evidence comes from and what to write when it isn't found.

Keep SKILL.md lean. New tables, playbooks and examples belong in references/, loaded only when needed.

The competitors in example-output.md are fictional by design. The file demonstrates format and standard of evidence, and must never assert facts about real companies.