Research competitors and generate comparison reports

Competitor Analysis turns a defined company and a user-approved competitor set into cited research and a local comparison report. It covers category discovery, company identification, feature and pricing comparisons, optional screenshots, and synthesis. The workflow researches the user's own product to the same evidence standard as competitors, preserving observation dates, source scope, and uncertainty.

Work begins with the company URL, category, seed competitors, target count, and desired depth. Category include and exclude phrases help screen candidates discovered through alternatives, category, and comparison-title queries. Brand names must resolve to first-party sites; a guessed domain or first search hit is insufficient identity evidence. A lexical category gate can return PASS, REJECT, or UNKNOWN, but its results do not establish commercial competitiveness. Ambiguous selections require an authorized selection rule or user review.

Quick research collects marketing and pricing. Deeper modes add discussion, social material, news, technical benchmarks, and comparison synthesis. An aggregate provider-call and cost budget covers discovery, self-research, gating, screenshots, checking, and retries. When the budget ends, gaps remain visible rather than being filled with invented claims.

Use this skill when a comparison needs per-company evidence and inspectable local output. Restrict research to approved public HTTP or HTTPS targets. Browserbase calls can incur charges and disclose queries, URLs, and page content, so provider, budget, and target authorization come before remote calls. Installing browse is a separate authorized action.

The workflow does not authorize contacting competitors, submitting forms, bypassing access controls, or fetching private targets. Lexical URL checks need network-level private-address and redirect protections. Screenshots require approved browser scope, document appearance at capture time, and do not establish capabilities.

Each company receives a flat frontmatter Markdown file. The parser handles single-line strings rather than general YAML, and research retains company names, public websites, and references. Selected partial findings are reconciled before matrix.json is synthesized. Supported presence and absence use booleans; missing evidence uses null or omission. Unmentioned features remain unknown.

Claims about prices, compliance, licenses, service levels, and benchmarks receive fact-checking. Vendor assertions, search snippets, independent findings, synthetic examples, and measured results remain distinguishable. Benchmark comparisons retain versions, tasks, datasets, metrics, dates, and sponsors; unlike setups do not establish superiority. Optional battle cards include fair strengths and uncertainty.

Compilation produces index.html, matrix.html, mentions.html, results.csv, and individual company pages. Inspect every company page, the matrix, and CSV coverage. Preserve errors, original partials, and incomplete lanes. A spot-check remains partial coverage.

Local helpers use Node built-ins; network helpers additionally require an approved browse executable. The compilation command is:

node "{SKILL_DIR}/scripts/compile_report.mjs" "{RUN_DIR}" --user-company "Company"

Use a fresh private run directory, reject unsafe path components, and preserve prior artifacts before reusing a location. Keep credentials in approved secret storage. Inspect custom HTML templates before opening output, and remove external font links when an offline or private report is required. These helpers do not provide a security sandbox, complete discovery, or source verification. The underlying workflow is MIT-licensed.