deep-recon-pimp

Active router for ALL research and recon requests — classifies and routes to the correct research sub-skill before any response. Use when doing deep research, competitor analysis, audience analysis, content research, or person research.

Model	Source
sonnet	pack: recon

Full Reference

If the request involves research in ANY way — fact-finding, competitor analysis, audience research, content research, person research, market research, verification, or intelligence gathering — you MUST route through this skill FIRST.

This is not optional. This is not negotiable. You cannot skip this.

Deep Recon Pimp

The orchestration layer for all research expertise. Not documentation — an active router. Every research request flows through this routing table before any response.

Mandatory Announcement — FIRST OUTPUT before anything else:

┏━ 🔍 deep-recon-pimp ━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ [one-line description of what request/routing]  ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛

No exceptions. Box frame first, then route.

Quick Context

The recon pack is armadillo’s intelligence gathering ecosystem — 5 sub-skills covering web research, competitor analysis, audience analysis, content research, and person research. Takes a vague question and produces a structured, cited, verified research brief.

Routing Table

Classify the request. Invoke the matching skill. No response before invocation.

Request Pattern	Skill
Fact-finding, verification, multi-source research, “research X”	web-research
Competitive landscape, market positioning, pricing analysis, feature gaps	competitor-analysis
Target audience, buyer personas, JTBD, psychographics, demographics	audience-analysis
Topic clusters, content gaps, keyword research, search intent	content-research
Public figure, bio, social presence, publications, speaker research	person-research

Cross-Cutting Rules

All research sub-skills enforce these constraints:

Constraint	Requirement
Minimum sources	3 for factual claims, 5 for statistics
Confidence levels	high / medium / speculative — flagged inline
Date-aware queries	Current year in all searches — never trust training-data dates
Citation format	Every claim links to source URL
Contradiction detection	When sources disagree, present both with evidence
Firecrawl fallback	All sub-skills may use firecrawl for deep extraction when WebFetch fails or for site-wide crawls

Resilience

All sub-skills follow these resilience rules:

Rule	Detail
Exponential backoff	1s → 2s → 4s → 8s on fetch failures
Max retries	3 per URL before marking as inaccessible
Timeout	30s per fetch (WebFetch or firecrawl)
Rate limiting	Minimum 1s between external API calls; back off on 429
Budget tracking	Log fetches consumed vs budget remaining

Firecrawl Budgets

Per-skill firecrawl page limits (in addition to WebFetch budgets):

Skill	Firecrawl Pages	Use Case
web-research	5 per session	Fallback when WebFetch fails
competitor-analysis	10 per competitor	Site crawl + messaging extraction
audience-analysis	5 per review platform	Review mining (G2, Amazon, Trustpilot)
content-research	10 per audit	Competitor article extraction
person-research	0	Uses WebSearch + WebFetch only

State Detection

Before routing, check project state to inform the research:

• brand.json → if exists, incorporate brand context into research
• stack.json → if exists, tailor technical research to the stack
• .claude/docs/plans/ → check for existing research briefs to build on
• business.json → if exists, use as context for competitor/audience research

State	Recommendation
brand.json exists	Pre-load brand positioning for competitor analysis
business.json exists	Use NAP data for local competitor research
Existing research brief found	Build on it, don’t start from scratch
No prior research	Start with web-research for broad context

Chaining Patterns

User Says	Chain
”Research my competitors”	competitor-analysis
”Who’s my target audience?”	audience-analysis
”Full competitive landscape”	competitor-analysis → audience-analysis
”Content strategy research”	content-research → competitor-analysis (content audit)
“Research this person for outreach”	person-research
”I need to understand this market”	web-research → competitor-analysis → audience-analysis
”Verify these claims”	web-research (verification mode)

Priority Order (when multiple skills apply)

1. web-research — broad fact-finding, verification
2. competitor-analysis — market positioning, competitive landscape
3. audience-analysis — who to target, buyer understanding
4. content-research — what to create, topic authority
5. person-research — individual intelligence gathering

What This Skill Does NOT Route

• SEO audits → seo-pimp handles
• Brand asset work → brand-pimp handles
• Content creation → content-creation pack handles
• Client reports → client-ops pack handles

Verification Gate

Every research output MUST include a verification summary table. No brief ships without this.

Column	Required
Source URL	Full URL
Tier	1 (primary) / 2 (official secondary) / 3 (journalism) / 4 (blog/forum)
Credibility Score	0-8 per source-verification.md rubric
Confidence Level	HIGH / MEDIUM / SPECULATIVE

Minimum thresholds:

• 3 sources for factual claims, 5 for statistics
• If any claim has only 1 source with credibility < 5, flag as SPECULATIVE
• All source URLs must be included in the final output — no citation without link

Research Brief Schema

Standard output shape for all research types. Downstream consumers (content-researcher, prospecting, client-reports) rely on this format:

Field	Type	Required
title	string	yes
type	web / competitor / audience / content / person	yes
date	ISO date	yes
sources[]	array of {url, title, tier, credibility_score, date_published}	yes
findings[]	array of {claim, confidence, supporting_sources[]}	yes
confidence_summary	{high: n, medium: n, speculative: n}	yes
contradictions[]	array of {claim, sources_disagree[], resolution}	if any

Downstream consumers:

• content-researcher — uses research brief for content writing pipeline
• prospecting — uses research findings for prospect intelligence
• client-reports — incorporates research data into client deliverables

Hard Rules

• Never respond about research before invoking the target skill
• No summarizing, planning to invoke, or explaining what you’re about to do
• If unclear, ask ONE clarifying question, then route
• The skill’s content has the verified methodology — always defer to it
• “Research this” without specifics → start with web-research
• ALL research outputs must follow the brief template from the sub-skill’s output-format.md
• Privacy: person-research is PUBLIC information only — flag and refuse if request targets private individuals