Market Intelligence

Market Intelligence Skills

A pair of Claude Code skills for evidence-first market research that produces consolidated, high-signal deliverables — without AI slop, hallucinated competitors, or invented pricing figures.

                 EXPENSIVE, ONCE PER MARKET          CHEAP, REPEATABLE
            ┌────────────────────────────────┐  ┌──────────────────────────────┐
            │         market-corpus          │  │         market-lens          │
            │                                │  │                              │
  user ───▶ │  GATE 1   Roster (approved)    │  │  GATE 3  Descriptive map     │
  question  │  GATE 2   Dossiers + evidence  │─▶│  GATE 4  Synthesis (blocked  │
            │           (live-sourced,       │  │          until 3 approved)   │
            │           confidence-labeled)  │  │                              │
            └────────────────────────────────┘  └──────────────────────────────┘
                  ask a SECOND question of the same market ──▶ lens run only (~10% cost)

Why this exists

Multi-agent "deep research" fails in predictable ways: rosters invented from stale training data, pricing figures with no source, recommendations written before the descriptive picture exists, and synthesis that reads like homogenized AI mush. These skills encode the fixes, learned from a real 200-agent competitive-research run:

• No claim without a live source. Every fact carries a URL, a confidence label (verified / single-source / inferred), and a compiled date. "I know this market" is treated as a hypothesis, not a fact.
• Human-approved gates. Roster → Corpus → Descriptive map → Synthesis. Synthesis is structurally blocked until the descriptive map is approved, because premature "whitespace" conclusions are the most expensive failure.
• Corpus / lens split. Evidence collection is the expensive part; analysis is a cheap re-read. Once a corpus exists, each new question (pricing, ICP, GTM, hiring signals) is a fast lens run — not a re-research.
• One editorial author per lens. Extraction may fan out across agents; exactly one agent writes each analysis document. Merged editorial swarms are how slop happens.
• Visible staleness. Every artifact header carries Compiled: <date>; lenses refuse to silently analyze evidence older than 30 days.
• The Intelligence Ledger. An ASCII status block shown at every check-in so you always know what is being captured and what question each phase will answer.

The two skills

Skill	What it does	Output
market-corpus	Defines inclusion criteria with you, discovers candidates via live search, gates the roster on your approval, then fans out collector agents (batches of 4) to build per-company dossiers with cited evidence	roster.json, per-company lens-*.md, dossier.md, evidence.json
market-lens	Runs analysis packs over an existing corpus. Ships with two packs: COMMERCIAL (pricing, packaging, engagement models, gating) and WHO-THEY-SELL-TO (ICP clusters, GTM motions, named-customer overlap)	analysis/dimensions/<lens>.md, then a synthesis answering your actual question with a per-claim source trail and an explicit Confidence & Gaps section

Lens packs are extensible — market-lens/lens-packs.md documents how to add packs (HIRING-SIGNALS, OPERATING-MODEL, NARRATIVE are sketched as candidates).

Install

Option A — Claude Code plugin marketplace (recommended)

/plugin marketplace add travcjohnson/market-intelligence-skills
/plugin install market-intelligence@market-intelligence-skills

Option B — manual copy (personal skills)

git clone https://github.com/travcjohnson/market-intelligence-skills.git
cp -r market-intelligence-skills/plugins/market-intelligence/skills/market-corpus ~/.claude/skills/
cp -r market-intelligence-skills/plugins/market-intelligence/skills/market-lens ~/.claude/skills/

Works in any agent harness that supports the Agent Skills format — copy the two skill directories to wherever your harness loads skills from.

Tooling recommendations

The skills work with whatever live-search tools your session has, but they are written to use:

Need	Recommended tool
Company/web discovery	Exa MCP — web_search_exa, company_research_exa, people_search_exa
Page-level facts (pricing pages, job boards, press)	WebFetch (built into Claude Code)
Funding cross-checks	Crunchbase/PR coverage via Exa; SEC EDGAR full-text search

No API keys are required by the skills themselves; they instruct the agent to use whichever of these is connected and to refuse memory-based claims when none are.

Usage

Start a market question and the corpus skill triggers:

"Research the market of boutique DevOps consultancies in North America — who the players are, how they price, who they sell to."

You'll approve the roster (Gate 1), then the corpus (Gate 2), then the descriptive map (Gate 3) before any conclusions are written (Gate 4). To ask a follow-up question of an existing corpus:

"Using the devops-na corpus, where is the pricing whitespace for a fixed-fee entrant?"

That's a lens-only run — minutes, not hours.

Provenance

Distilled from a real June 2026 research program: 23 firms, ~200 agents across roster discovery, siloed company audits, dimension analysis, and ICP mapping. The failure modes these skills guard against (fabricated rosters, premature synthesis, editorial-swarm slop, burst-rate-limit deaths) were all observed in that run, and the guardrails were pressure-tested against baseline agent behavior before release.

License

MIT