Listops

ListOps

Investment-grade company list building for Claude Code, powered by the Acqwired DRA platform. Turn a one-line request — "commercial landscape maintenance companies within 90 miles of these addresses, owner-operated, >$5M, no PE" — into a screened, enriched, thesis-ranked export.

Discovery runs through the DRA submit_company_list API, never through Claude web search. Live sources in v1 are Google Maps (POI) and Exa (neural); Yelp, Foursquare, and Tavily are accepted but not yet implemented (they return zero hits with a not_implemented per-source error — don't plan to depend on them). Claude skills supply the intelligence: which sources, keywords, locations, radii, schemas, filters, and thesis.

Pipeline

PLAN → LIST → INTAKE SCREEN → [no requirements? → export offer]
     → (ENRICH tier → FILTER)×N → THESIS → QA → GAP-FILL → EXPORT

Stage	What happens
Plan	MECE source payload (keywords × locations × radius ≤ 90mi), schema pack with explicit pass/fail rules, thesis text, reserved QA probes, credit budget
List	submit_company_list → poll → companies.jsonl with per-source provenance
Intake screen	Free local screening: dedupe, PE/franchise blocklist (~60 curated platforms shipped), category sanity, website-gap flags
Enrich + Filter (tiered)	Schemas ordered by kill-rate per credit; tier 1 → filter → tier 2 on survivors → … Each tier credit-gated; confidence <50 = data gap, not rejection
Thesis	completeCompanyProfile + thesis text on survivors → rank by thesis_fit_score
QA	Independent judge: spot-checks, mandatory ownership re-verification, coverage probes via the list API
Gap-fill + Export	reconstruct-first field completion → final.csv with provenance + confidence

Every run is tracked in lists/<slug>/session-tracking-<id>.json (atomic writes) — any session or command resumes exactly where the last one stopped.

Setup

This plugin is a thin bootstrap. Installing it gives you two commands — /listops:connect and /listops:update. The skills, pipeline commands, and the QA agent are not shipped in the marketplace; they're downloaded from the DRA API on connect, gated by a valid (org-validated) key, and written under ~/.claude/. The proprietary logic never sits in a public repo.

1. Install the plugin (marketplace README one level up). The bundled .mcp.json wires the dra-research MCP server.
2. Connect. Get your assigned key from the Acqwired dashboard (Settings → API key), then run /listops:connect dra_.... This validates the key, stores it in user-scope ~/.claude/settings.json, and downloads + installs the ListOps skill pack. An invalid key is rejected server-side (401) and nothing installs.
3. Restart Claude Code once — the keyed MCP server and the downloaded skills/commands/agent all load at startup. Then /listops:status confirms.
4. /listops:update re-syncs the skills to the latest version anytime.

Requires Python 3.8+ on PATH (the bootstrap script is stdlib-only). (CI / power users: export DRA_API_KEY=dra_... in the shell — it takes precedence over settings.json — then connect.py update fetches the pack.)

Commands

connect and update ship with the plugin; the rest are installed on connect (they appear as /listops:* after the one-time setup + restart).

Command	Stage
/listops:connect <dra_ key>	One-time setup: authorize + download the skill pack
/listops:update	Re-sync the installed skills to the latest version
/listops:build <description + requirements>	Full pipeline
/listops:plan	Source payload + schema pack + budget
/listops:list	LIST + INTAKE via the DRA API
/listops:enrich	Tiered enrich→filter loop
/listops:filter	Standalone re-filter (no new spend)
/listops:thesis	Thesis scoring + ranking
/listops:qa	Independent QA
/listops:export	Gap-fill + final deliverable
/listops:status	Session status, funnel, gates, credits

Plain language works too ("find founder-owned HVAC companies near Tampa") — the list-building skill triggers on intent.

What's delivered on connect

The marketplace package contains only the bootstrap (.mcp.json, the connect/ update commands, and connect.py). Everything below is served by the DRA API to valid keys and installed under ~/.claude/ — it is not in this repo:

• Skills: list-building (the planner — candidate acquisition planning, enrichment waterfall & down-selection planning, QA planning) and list-operations (the operator — runs the list/research APIs, combines sources + dedupes, creates schemas, executes spend responsibly, performs scoring/sorting/down-selection)
• Agent: qa-judge (independent verification)
• Scripts: session.py (atomic session tracking), intake_screen.py (blocklist + dedupe screening), dedupe.py, dra_client.py (resumable batches with credit pre-flight)
• Assets: pe-platforms.txt (curated PE/franchise blocklist — append as QA confirms new ones)
• The list_companies API spec for the backend team: docs/list-method-design.md at the marketplace repo root.