List Similarity Check

Step 1 — Get the ICP to compare

Accept whichever form is available:

• icp-definition.md — if one exists in the current directory, read it
• Freeform description — e.g. “HR leaders at mid-market SaaS companies in the US”
• Current conversation context — if icp-capture or prospect-builder just ran, the ICP is already in context

Extract these dimensions for matching:

• Vertical / industry
• Geography
• Company size
• Persona / target title
• Targeting method (open market, account list, lookalike, keyword, tech stack)
• Special signals (hiring, funding, headcount change, etc.)

Step 2 — Find the history source

Try each source in order, use the first that’s available:

Option A: Google Sheet URL

If the user provides a Google Sheet URL, or one is saved in .landbase/config:

• If a Google Sheets MCP tool is available, read the sheet directly
• If not, ask the user to export it as CSV and provide the file path

Expected columns (flexible — the skill adapts to whatever’s present):

Option B: Local CSV or Excel file

If the user provides a file path, or a prospect-history.csv / prospect-history.xlsx exists in:

• Current directory
• ~/.landbase/prospect-history.csv

Read it and match against whatever columns are present. Adapt to the schema — don’t require exact column names.

Option C: prospects/ directory

Scan for past prospect-builder runs by finding brief.md files:

find ./prospects -name "brief.md" | sort -r

Read each brief.md for the original ask and ICP criteria. This is the zero-configuration path — works automatically if the user has used prospect-builder before.

Option D: No history found

If none of the above yields results:

“No prospect history found. Once you’ve run a few builds with prospect-builder, this skill will automatically find similar past work in your prospects/ directory. You can also maintain a prospect-history.csv for builds done outside this tool.”

Offer to create a prospect-history.csv template they can start filling in.

Step 3 — Score similarity

For each history entry, score against the current ICP:

HIGH — same or closely overlapping vertical + persona + geography

Likely reusable with minor edits. Surface immediately.

MEDIUM — shares 2–3 key dimensions

Useful as a starting template. Worth reviewing.

LOW — shares 1–2 dimensions

Reference only. Probably not worth reusing.

Rank by similarity. Surface top 5–10 matches.

Step 4 — Present results

Show as a table:

If no matches score MEDIUM or higher:

“No strong matches found. This looks like a new build.”

Step 5 — Act on results

If HIGH match found: Ask the user:

“A similar list was built on {date}. Want to start from that run, or build fresh?”

• Start from prior run — point them to the output path; suggest running prospect-builder with the prior query as a starting point
• Build fresh — proceed, but note what’s different about this run vs. the prior one

If MEDIUM match found: Show the prior query or brief and ask:

“This is similar but not identical. Want to use it as a reference?”

If no match: Proceed to the new build. Optionally log this run to the history file (Step 6).

Step 6 — Log the new run (optional)

After a new prospect build completes, offer to add it to the history:

Append to prospect-history.csv:
name, date, description, vertical, geography, company_size, persona, signals, output_path

If using the prospects/ directory, this happens automatically via prospect-builder’s brief.md.

History file template

If no history exists and the user wants to start one, write prospect-history.csv:

name,date,description,vertical,geography,company_size,persona,signals,output_path

Save to ./prospect-history.csv or ~/.landbase/prospect-history.csv per user preference.

History sources

Red Flags — STOP

• About to start a new build without checking history → run this skill first
• History file has no description or vertical column → adapt to whatever columns exist, don’t fail
• HIGH match found but user wants to build fresh anyway → note the duplication, proceed without blocking