prospect-evaluation - SKILL.md Agent Skill

name: Prospect Evaluation description: Evaluates US-based businesses against Biffrey Braxton / EarnedOut's acquisition Buy Box. Use when the user asks to screen, score, evaluate, or write a deal memo for a prospective business acquisition, LBO target, small-cap M&A prospect, roll-up add-on, or asks to "run this through the buy box" or "score this prospect". Produces a Buy Box screening header, 26-field scorecard with a 0–100 lead score (up to 110 for roll-up add-ons), and an exhaustive supporting deal memo in both Markdown and HTML. Handles both single-business deep dives and batch screening of candidate lists. Supports roll-up add-on evaluation for Applied Development (sign language/CART) and Fambro Waste Management (commercial waste). Also screens SBIC (Small Business Investment Company) GP / management-company acquisitions.

Prospect Evaluation Skill

Evaluates US-based businesses as acquisition targets for Biffrey Braxton (EarnedOut). Produces a deterministic Buy Box screen, a 0–100 lead score, and a full deal memo.

Path note. Every references/… and templates/… path in this skill is relative to this skill's own directory — the folder that contains this skill.md. Load those files from beside this file. Do not resolve them against the caller's working directory, and never fall back to a copy under ~/.claude/ or any other location. The rubric in references/buy-box-and-scoring.md is the single source of truth — read it from disk and apply it verbatim; never score from memory or approximation.

Skill copies & sync policy. This skill exists in THREE places on this machine. If they disagree, the PRIMARY wins.

PRIMARY — make all edits here: /Users/biffreybraxton/.claude/skills/prospect-evaluation/ (user-level installed copy; the one Claude Code executes in most sessions).

Git/GitHub mirror: /Users/biffreybraxton/Library/CloudStorage/GoogleDrive-bbraxton@applied-dev.com/My Drive/Investments/Prospect Evaluation/Prospect-Evaluation-Skill/.claude/skills/prospect-evaluation/ (repo github.com/biffrey/Prospect-Evaluation-Skill). Never edit directly. To publish to GitHub: rsync -a --delete from the PRIMARY into this folder, then commit — commits must only ever contain content synced out of the PRIMARY.

Project mirror: /Users/biffreybraxton/published-listing-search/.claude/skills/prospect-evaluation/ (used by sessions rooted in published-listing-search, e.g. overnight-search). Never edit directly; refresh from the PRIMARY.

Additional clones of the git repo exist at /Users/biffreybraxton/Prospect-Evaluation-Skill/ and /Users/biffreybraxton/Code/Prospect-Evaluation-Skill/ — treat them as pull-only: never edit or commit from them (uncommitted work there has been lost to drift before). The same policy applies to the sibling valuation-estimate skill. Note the project mirror (location 3) is additionally tracked in the earnedout-workspace GitHub repo — commits there must also contain only PRIMARY-synced content.

Core outputs (always, in this order)

Buy Box Screening Header — six ✅/❌ checks, each with a cited data point.
26-field Scorecard — the exact output fields listed in references/research-playbook.md.
Lead Score 0–100 (or 0–110 for roll-up add-ons) — computed using the exact rubric in references/buy-box-and-scoring.md, with per-line math shown.
Supporting Deal Memo — the 11 sections (Executive Summary → Appendix) as defined in templates/single-report.md.

Outputs are written to the user's current working directory as both <company-slug>-report.md (the working doc) and <company-slug>-report.html (a shareable single page). Never skip either format.

Workflow

Step 1 — Mode + context detection

Decide single vs batch:

Single when the user gives one company name, one folder of documents, one URL, or one CIM.
Batch when the user gives a list, a spreadsheet, a folder containing multiple sub-folders, or says "screen these / score this list / which of these should I pursue".

Also detect roll-up add-on context:

If the target is in translation or interpretation services (NAICS 541930) — sign language interpreting, CART/captioning, VRI, deaf services, or spoken/foreign-language translation & interpreting → it's a potential add-on for Applied Development. There is no size floor for Applied Development add-ons — evaluate every such business regardless of size. Include the +10 bonus scoring line.
If the target is in commercial waste, construction waste, dumpster rental, C&D hauling, or debris recycling → it's a potential add-on for Fambro Waste Management. Apply the relaxed add-on thresholds (5 FTE / $500K EBITDA / 5 yrs) and the +10 bonus line.
For all other target industries → evaluate as standalone with standard thresholds and max score of 100. SBIC targets are standalone — see rule 13 below.

Also detect rescore mode: if the caller states "this is a rescore" / "rescore mode" (e.g. the process-pending-rescores runner re-evaluating a flagged prospect against its updated working folder), this is always single mode. Carry the rescore flag through to Step 9, where it changes the publish behavior (no disposition prompt; existing Disposition preserved). See Step 9 and references/publish-to-pipeline.md.

Ask the user for clarification only if the mode or roll-up context is genuinely ambiguous.

Step 2 — Gather provided materials

Use Read to load every document the user references (PDFs, CIMs, P&Ls, tax returns, emails, listing screenshots).
Note any document that appears to need OCR (scanned images without text) and tell the user.
Extract structured facts first (revenue by year, EBITDA, employees, location, asking price, founding year, principal names).

Step 3 — Industry + geography gate

Load references/industries-and-geography.md. Before doing any deep research:

Confirm the business sits inside a target industry (or is a disallowed subset — e.g., nuclear pharmacy retail, precious metals jewelry retail).
Apply the exclusion list (restaurants, retail, distressed, weapons manufacturers).
Apply geography rules, including the law-firm jurisdiction rule (AZ, PR, UT, MD, DC, VA only).
If the target fails the gate, say so loudly up top but still fill in the scorecard and score so the user has a full record of why it was rejected.

Step 4 — Buy Box screening

Load references/buy-box-and-scoring.md. Evaluate the six hard criteria:

≥10 full-time employees
EBITDA $1M or more (no upper ceiling)
10+ years in business
3+ years of YoY revenue and profit growth
Asking price ≤ 4x EBITDA
DSCR > 1.4 (informational only — not part of the 0–100 score)

For each, emit ✅ or ❌ with a one-line evidence citation. If a criterion can't be evaluated (data missing), mark it ⚠️ "insufficient data" — never guess a pass.

Step 5 — Data gathering (fill the 26 fields)

Load references/research-playbook.md for the source-priority order per field. For any field not in the provided docs, aggressively research the web using WebSearch and WebFetch:

Company website, About / Team / History pages
LinkedIn company page and principal profiles
State Secretary of State business registries (for founding date, officers, registered agent)
County assessor records (for owned real estate)
BizBuySell / BizQuest / Axial / broker listing pages (for asking price)
News, press releases, podcast interviews, awards
Industry-specific registries (FAA repair-station search for aerospace, NRC licensing for nuclear pharmacy, state bar for law firms, NPI/CMS for medical)
Court records (PACER / state dockets) for law firms
Wayback Machine for history and earliest capture

Estimation rules:

Mark every estimated value with (est.) inline in the scorecard.
Every estimated value MUST have a corresponding line in the Appendix showing the source, method, and math.
Never fabricate financials. If a field cannot be sourced or reasonably estimated, leave it blank.
LTV and CAC are left blank unless explicitly provided (per spec).

Step 6 — Score (0–100 or 0–110 for add-ons)

Apply the rubric from references/buy-box-and-scoring.md. Show the per-line math in the "Lead Score Breakdown" section — not just the total. The base nine rubric items sum to 100. If the target is a roll-up add-on, also evaluate line 10 ("Roll-up / add-on strategic fit") for up to 10 bonus points (max 110). Show the denominator clearly: "XX / 100" for standalone, "XX / 110" for add-ons.

Step 7 — Write the report

Single mode: instantiate templates/single-report.md AND templates/single-report.html with the collected data. Write both files to the current working directory. File names: <company-slug>-report.md and <company-slug>-report.html.
Batch mode: instantiate templates/batch-screen.md AND templates/batch-screen.html, sorted by score descending, with bucketed lists (Top/Review/Pass/Excluded). Write both files to the current working directory. File names: <batch-slug>-batch-screen.md and <batch-slug>-batch-screen.html. After writing, offer to generate full single-business reports for the top candidates.

Step 8 — Self-check before finishing

Verify all of:

Every Buy Box ✅/❌/⚠️ has a cited data point in the evidence column.
All 26 scorecard fields are filled OR explicitly blank per the rules (LTV/CAC blank is acceptable).
The score math sums correctly and matches the header number.
The appendix lists every URL actually fetched (not a generic "research sources" list).
Every (est.) value has an appendix entry showing the method.
Both .md and .html outputs exist (single mode AND batch mode).
For law-firm targets, the state of domicile is named and the legality rationale (AZ/PR rule, UT sandbox, or Brandon Thornton jurisdictions MD/DC/VA) is stated.

Step 9 — Publish to the EarnedOut dashboard (single mode only, EarnedOut-specific)

After Step 8 self-checks pass and only when running in single mode, check whether the EarnedOut publish pipeline is set up on this machine. The full procedure is in references/publish-to-pipeline.md. Summary:

Skip this step entirely if any of these is false: the folder /Users/biffreybraxton/published-listing-search/output/reports/ exists, the user did not opt out, the Airtable MCP tools (mcp__airtable__*) are available, and this is single mode (not batch).
Otherwise, load references/publish-to-pipeline.md and follow steps A → F (the row is created BEFORE the copy so the report can embed a deep link to its own Airtable record):
- A: Ask the user to pick a Disposition from the Airtable single-select list (Active, Contacted, Maybe Later, Revisit for Roll-up, Passed, Dead Link). Wait for their answer — this is a deliberate gate. Rescore-mode exception: if this run is a rescore, do not prompt; reuse the row's existing Disposition unchanged.
- B: Search the Master Deal Pipeline Airtable for an existing row matching the company name. Create a new row OR update the matching one (don't make duplicates). Capture the record ID.
- C: Embed the Airtable deep link (https://airtable.com/{base}/{table}/{recordId}) into the report HTML — fill the {{Airtable Record ID}} placeholder in the airtable-deep-link block.
- D: Copy <slug>-report.html into ~/published-listing-search/output/reports/name-<slug>[-<state>]/. The .md and the cwd .html stay where they are as the working source.
- E: When updating an existing row, leave existing Disposition alone unless the user explicitly says otherwise; append to Notes instead of replacing; and set/refresh the Working Folder Path field to the absolute path of the prospect's working folder when known.
- F: Confirm to the user — print the Airtable record ID, the path of the published HTML, and the live dashboard URL.

The launchd file-watcher will sync the new HTML to smbsteward.com within ~30 seconds. The new/updated row appears on the dashboard on the next page load.

Step 10 — Offer to chain into Valuation Estimate (single mode only)

After the single-prospect report is finished (and published, when Step 9 applies), offer to run the valuation-estimate skill on the same prospect: it computes an EBITDA multiple score and an estimated valuation from the report you just wrote plus the supporting documents in the working folder, and writes <company-slug>-value-estimate.html alongside the deal memo. If the user accepts, invoke valuation-estimate pointed at this prospect's working folder — it reuses everything already gathered. Skip this offer in batch mode (offer it instead for any top candidates the user asks to deep-dive).

Behavioral rules (always)

Never fabricate financials. Missing > guessed.
Tag estimates with (est.) and justify them in the Appendix.
Leave LTV and CAC blank unless the user provided them.
Valuation multiple tiered scoring (≤4.0x = 15 pts, 4.0–5.5x = 7 pts, >5.5x = 0 pts). See scoring reference for details.
EBITDA tier is non-cumulative — take the highest applicable tier only.
DSCR > 1.4 is informational only — it appears in the header but is NOT part of the 0–100 score.
Law firms: explicitly state the state of domicile and the legality rationale. Reject law firms outside AZ, PR, UT, MD, DC, VA.
Precious metals and nuclear pharmacy: confirm the target sits in the allowed subset (refiners/recyclers; radiopharma/oncology compounding). Otherwise reject.
Weapons manufacturers: reject. Government / DOD / cybersecurity / aerospace-support / emergency-management firms are fine.
Fail-the-gate targets still get a full scorecard and report — the user wants a clear record of why a prospect was rejected, not just a one-word "no".
Roll-up add-on detection is automatic. If the target's industry matches translation/interpretation services (NAICS 541930 — sign language, CART, VRI, deaf services, or spoken/foreign-language translation) → Applied Development add-on, evaluated with no size floor (any size is worth evaluating). If it matches commercial/construction waste → Fambro Waste Management add-on, evaluated with the relaxed add-on thresholds (5 FTE / $500K EBITDA / 5 yrs). Both include the +10 bonus scoring line and use the add-on interpretation bands (max 110). The report must name the platform company explicitly.
Platform companies:
- Applied Development — communications access, sign language interpretation, CART.
- Fambro Waste Management — construction and commercial trash management.
SBIC targets. An SBIC acquisition is the purchase of the GP / management entity holding an SBA SBIC license. The sole gating criterion is that the license is active and in good standing; the standard EBITDA / FTE / growth / valuation criteria are reported but non-gating, and the 0–100 score is informational only. Any change of control of a licensed SBIC requires SBA approval (closing condition). See references/buy-box-and-scoring.md → "SBIC (Small Business Investment Company) targets".
Rescore mode. When the caller signals a rescore (the process-pending-rescores runner re-evaluating a flagged prospect against its updated working folder), behave exactly as a normal single-mode evaluation except at Step 9: do not prompt for a Disposition — preserve the existing one — and refresh the Working Folder Path. Everything else (full scorecard, fresh score, appended Notes) is unchanged.

Reference files (load only when needed)

These live in this skill's own references/ directory, beside this skill.md (see the Path note above).

references/buy-box-and-scoring.md — Buy Box criteria, full financial/operational criteria, deal structure prefs, and the exact 0–100 rubric.
references/industries-and-geography.md — Target industries with keyword lists, exclusions, geography priorities, law-firm jurisdiction rules.
references/research-playbook.md — Field-by-field source playbook for the 26 scorecard items.
references/publish-to-pipeline.md — EarnedOut publish flow (single mode): copy to output/reports/, prompt for Disposition (skipped in rescore mode), create/update Airtable row, set Working Folder Path. Loaded only at Step 9.

Templates (instantiate on output)

templates/single-report.md — Full Markdown deal memo.
templates/single-report.html — Styled HTML single page.
templates/batch-screen.md — Batch ranked screen (Markdown).
templates/batch-screen.html — Batch ranked screen (styled HTML; mirrors single-report.html visual language).