competitive-research

Competitive Research Skill for Claude Code

A weekly competitive analysis workflow that auto-discovers your product from its codebase, researches competitors on the web, and produces a ranked, evidence-backed list of features you're missing — scored with PM frameworks (Jobs-to-be-Done, Kano, RICE) and deduped against prior weeks.

Zero-config: drop the skill in, run it from your repo, get a report. No filling out forms.

---

What you get

Every run produces a report.md with:

1. Executive summary — the 3–5 most important findings
2. Product dossier — what the skill inferred about your product (so you can correct it if wrong)
3. Landscape snapshot — how the competitive set has shifted since last run
4. Feature matrix — a ✅/❌/🟡 table across you and 8–15 competitors
5. Top 10 feature recommendations — each with RICE scoring, evidence quotes from real users, and specific files/modules in your codebase where the work would land
6. "Don't build" list — features competitors have that you should deliberately skip
7. Non-feature observations — pricing, positioning, and messaging shifts
8. Sources — every claim traceable to a URL

State persists across runs in ./competitive-research/, so each week dedupes against the last and surfaces what's genuinely new.

---

Prerequisites

• Claude Code installed and working — see the official installation guide if you don't have it yet
• Python 3.7+ for the dedupe script (almost certainly already installed)
• gh CLI (optional but recommended) for richer git/issue/PR signals — install instructions
• tokei or cloc (optional) for repo size metrics

---

Installation

This is a Claude Code plugin. Install it via the marketplace flow or directly from a local clone.

Option A — Install from this GitHub repo (recommended)

Inside Claude Code:

/plugin marketplace add dharrisHub/Competitive-Research-Skill
/plugin install competitive-research@Competitive-Research-Skill

The first command registers this repo as a marketplace; the second installs the plugin from it. Claude Code resolves the plugin name (competitive-research) from .claude-plugin/plugin.json.

Option B — Local development install

If you cloned this repo and want to test or hack on it locally:

git clone https://github.com/dharrisHub/Competitive-Research-Skill.git
cd Competitive-Research-Skill

Then in Claude Code:

/plugin marketplace add ./
/plugin install competitive-research@Competitive-Research-Skill

Edits to skills/competitive-research/SKILL.md and the references take effect on the next skill invocation — no reinstall needed.

Verify the plugin loaded

/plugin

You should see competitive-research in the installed plugins list.

What's in the plugin

Four skills, each invokable inside any project repo:

Invocation	Purpose
natural language ("run the competitive research") or /competitive-research:competitive-research	The full 8-phase weekly analysis. The main event.
/competitive-research:preview	1–2 minute sanity check — runs only Phase 0 (dossier) + Phase 2 (competitor list), then asks before doing the full deep dive.
/competitive-research:setup	Interactive walkthrough of the optional override questions (target user, known competitors, strategic constraints, monorepo scope). Writes overrides.yaml.
/competitive-research:track	Mark past recommendations as shipped / in-progress / rejected / wontfix. Future runs filter accordingly so you don't see the same suggestion next week.

The three slash-only skills (preview, setup, track) only fire on explicit invocation — they won't auto-trigger from natural language. The main analysis skill triggers broadly so you can ask for it conversationally.

---

Usage

First run

1. cd into your product's repo (or a monorepo subdirectory if scope C).

2. Open Claude Code.

3. Say something like:

   Run the competitive research

   Or any of these — the skill is set to trigger broadly:

   • "What features are competitors shipping that we're not?"
   • "Do a gap analysis on my product"
   • "What should I build next quarter?"
   • "Run the weekly competitive review"

The first run takes the longest (typically 10–20 minutes of work) because it's discovering everything from scratch — your product identity, your features, your competitors. Subsequent runs are faster because state persists.

Tip — preview before committing to a full run. If you want to sanity-check the inferred product identity and competitor list before spending 15 minutes, run /competitive-research:preview first. It does only Phase 0 (dossier) and Phase 2 (competitor list), prints them, and asks before continuing. The full analysis afterwards reuses the dossier rather than redoing it.

Weekly cadence