skill
buy-box-mapper
Type a buy box. Get back a ranked list of off-market companies that fit it. Six skills, three agents and one command for Claude Code.
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/corebits-io/buy-box-mapper itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 26, 2026Buy Box Mapper
Type a buy box. Get back a ranked list of off-market companies that fit it.
Six skills, three agents and one command that take an acquisition mandate and return a prioritised target list — sector, geography, revenue, estimated EBITDA, ownership, platform or add-on fit, and a dated reason to call this quarter — with the rule that produced every verdict written next to it.
No associate reading two hundred companies one at a time.
Built for PE funds, family offices, search funds, independent sponsors, holdcos and corporate development teams.
Runs on Claude Code and local files. No database, no API keys, no signups.
👉 Start here: SETUP.md
Fifteen minutes from zero to a ranked list. Every step in plain English.
Want to see the end product first? samples/sample-output.md is a complete finished run, invented end to end, so you know exactly what you are getting before you install anything.
Came from the LinkedIn post? — this is the system
One command
Fill in one file, then type:
/map-buy-box
The run stops and shows you its work before every write. Five steps, five approval gates, nothing happens behind your back.
Every row comes back with this
| Field | What it holds |
|---|---|
| company, domain, HQ | identity and geography |
| sector, segment | where it sits in the map |
| ownership_type | founder, family, management, ESOP, sponsor, corporate, or Not found |
| ownership_evidence | the filing or register that establishes it, with a tier and a date |
| sponsor_screen | the check that catches companies already bought but still marketing themselves as family owned |
| est_revenue, est_ebitda | a range, labeled an estimate, with the basis attached |
| financial_confidence | verified / high / medium / low / Not found |
| fit_role | platform or add-on, and for an add-on, the named platform it attaches to |
| signal | the reason to call this quarter, quoted verbatim, with a working URL and a date |
| signal_strength | high / medium / low / Not found |
| verdict | PRIORITY / WATCH / PASS |
| fit_reason | the rules that fired, so a verdict can be argued with in March |
Three verdicts, not a score out of 100. A number between 1 and 100 invites an argument about whether a 71 beats a 68 and tells nobody what to do on Monday.
PRIORITY — strong mandate fit, strong reason to engage now WATCH — fits the thesis, needs a better trigger PASS — wrong ownership, economics, geography, or strategic fit
The six skills
| Skill | What it does |
|---|---|
buy-box-intake | Checks the box is complete enough to map against, rewrites untestable must-haves into tests, scopes the run. Runs a principal-buyer gate that has no override |
buy-box-market-map | Maps the segment the box implies: fragmentation, who is already consolidating, the typical owner, the segment economics every later estimate depends on, and where these companies are actually findable. Ends with a verdict that is sometimes "do not source this segment" |
buy-box-discovery | Enumerates the population from licence rosters, registries, association directories, exhibitor lists and dealer locators — the sources bought lists never cover — with a source per row and a coverage log that names what the pull missed |
buy-box-ownership-and-size | Who actually owns it, the sponsor screen, a defensible size read on companies that publish nothing, and the platform or add-on call |
buy-box-signals | The four families — growth, hiring, financing, leadership — quoted verbatim, sourced, dated, and aged against a per-family recency window |
buy-box-ranking | Builds a rubric from your box, gets it approved before ranking a single row, ranks every row identically, then audits itself |
Install, then run /skills to see all six.
The three agents
| Agent | What it does |
|---|---|
buy-box-mapper | Box, map, population. Stops at the list of company names |
buy-box-analyst | Ownership, size, role, signals. Fills in what the ranking needs to know |
buy-box-ranker | Rubric, ranking, audit. Turns research into decisions |
Install, then run /agents to see all three.
Both lists are the same system. The agents run the skills.
The chain
buy-box-intake box checked, run scoped
↓
buy-box-market-map segment economics, sources, verdict
↓
buy-box-discovery population enumerated, deduped, sourced
↓
buy-box-ownership-and-size owner, sponsor screen, revenue, EBITDA, role
↓
buy-box-signals the dated reason to call this quarter
↓
buy-box-ranking PRIORITY / WATCH / PASS, then the audit
The rules that make the list trustworthy
Any system can return two hundred ranked rows. These are the rules that decide whether the ranking means anything.
- Unknown is not failed.
Not foundrevenue is not a failed size test. Ranking gaps as negatives systematically deletes the least-marketed companies from the list, which is to say the proprietary ones - Estimates are labeled, and the label travels with the number. A labeled estimate outside your size box makes a row WATCH, never PASS. Only a verified figure excludes on economics
- Sponsor-owned is always PASS. The screen runs on every row, including the ones that look obviously founder-owned, because companies keep the family-owned language for years after a sale
- A thin website is not a negative signal. The companies with no marketing budget are the ones no banker has called
- Fit and timing stay separate. A loud signal never promotes an out-of-mandate company. A missing signal never deletes an in-mandate one
- Signals are quoted verbatim or they do not exist. "We are exploring options for the next phase" is a quote. "The founder is looking to sell" is a fabrication, and it is the kind that ends up in an email to the founder
- The audit runs every time and gets reported whether or not it is clean
- Nothing is deleted. A PASS carries the rule that excluded it, so the same company is not rediscovered and re-researched in six months
What this deliberately does not do
No contact finding. No outreach drafting. No pipeline tracking. No sending.
The package stops at the ranked list, which is the part that costs a fortnight of associate time. Finding the owner, writing the approach and handling the replies are a different job with different rules, and doing them badly is worse than not doing them.
Repo contents
SETUP.md— the full setup guide, start herecommands/— the/map-buy-boxcommandskills/— the six skills Claude Code readsagents/— the three subagentsreferences/— the ranking rubric, estimation rules, signal evidence rules, and where the companies actually aretemplates/— the buy box template, a fully worked example buy box, and the 29-columntargets.csvsamples/— a complete finished run, so you can see the output before you install
A word on the research
Everything here runs on public sources: registers, licence rosters, filings, procurement awards, trade press, association directories, and what companies publish about themselves.
Nothing in this package scrapes behind a login, pretends to be somebody it is not, or guesses an email address. If you are a fee-based intermediary rather than a principal buyer, the intake skill will stop and tell you to talk to securities counsel before it researches anybody on your behalf. That gate is deliberate and has no override.
Free and MIT licensed. Built by Corebits.