adv-comp-research

name	adv-comp-research
description	Competitive intelligence research methodology for source code and public information comparison
keywords	["competitive","comparison","market","pricing","features","intelligence","research"]
metadata	{"priority":"medium","source":"agent-created"}

Competitive Intelligence Research Skill

Purpose

Reusable competitive intelligence methodology for ADV comp-scan workflows. Provides two-mode research strategy, auto-mode detection, comparison table format, and evidence requirements.

Two-Mode Strategy

Mode 1: Source Code Analysis

When the competitor is a GitHub repository (URL matches github.com/*):

1. Extract repo metadata — Exa for candidate repos; searchcode file-tree, file-fetch, and repo-analysis capabilities for README, package manifests, main language
2. Read feature surface — analyze source structure, API endpoints, key modules
3. Compare architecture — layer structure, dependency graph, tech stack

Mode 2: Public Information Scraping

When the competitor is a website, product page, or documentation:

1. Firecrawl primary — Firecrawl scrape for pricing, features, changelog pages
2. Exa fallback — Exa web search (standard or advanced) for recent news, reviews, comparisons
3. Structured extraction — identify feature list, pricing tiers, target audience

Auto-Mode Detection

URL Pattern	Mode	Override
github.com/*	source	--mode source
Everything else	public	--mode public

User can force mode via --mode source|public. When ambiguous, default to public mode.

Output Format

Comparison Table

Feature	This Project	Competitor	Notes
{feature}	{our_status}	{their_status}	{observation}

Structured Finding

{
  "category": "feature|pricing|performance|security|ux",
  "our_status": "present|missing|partial|unknown",
  "their_status": "present|missing|partial|unknown",
  "delta": "parity|advantage|gap|unknown",
  "source": "https://..."
}

Evidence Requirements

• Always cite source URLs — every finding must include a source field
• Flag unverifiable claims — mark as confidence: low when source is secondary
• Timestamp findings — note when data was collected (pages change)
• Distinguish observed vs inferred — observed = direct from source; inferred = agent deduction

Constraints

• Read-only guidance — this skill does not mutate ADV state
• No gate completion — the command owns scan orchestration
• No workflow sequencing — the command owns phase ordering and sub-agent dispatch
• Respect robots.txt — Firecrawl handles this; do not bypass
• No confidential data — do not access private repos or authenticated endpoints without explicit user consent