Brandedflow icp

install

source · Clone the upstream repo

git clone https://github.com/JenCW/brandedflow

manifest: business-os/sales/research/icp/skill.md

source content

Skill: ICP

Purpose

Derive a deterministic ideal customer profile (ICP) JSON from existing client research for custom home builders targeting $5,000,000+ projects, custom homes, in Southern California. No live APIs; same

client_id

and same

research.json

always produce the same

icp.json

. Segment rules (

build_value_min

project_type

location

) are fixed constants in the runner/refine layer; eval continues to validate only the original required fields.

Task

Input (runner):

{ client_id }

— optional

projectRoot

(defaults to repo root).

Reads (only):

clients/{client_id}/research/research.json

Writes (only):

clients/{client_id}/icp/icp.json

Output schema (deterministic)

Eval still gates the original required fields only. The written JSON also always carries fixed segment metadata: numeric

build_value_min

, string

project_type

custom homes

, and string

location

Southern California

(enforced in runner/refine, not duplicated in eval rules).

Field	Type	Rules
`client_id`	string	Non-empty, path-safe (no `/` , `\\` , `..` ).
`target_company_profile`	string	Non-empty trimmed narrative: enforced SoCal $5M+ custom-home line plus research `company` + `services` context.
`decision_makers`	string[]	At least one non-empty string; fixed roles for this segment: Owner, Principal Builder, Project Director.
`pain_points`	string[]	At least one non-empty string; high-end custom-builder focus—lead quality, long cycles, brand mismatch.
`buying_triggers`	string[]	At least one non-empty string; permit/jurisdiction activity, pipeline gaps, growth-phase pressure.
`disqualifiers`	string[]	At least one non-empty string; tract/production builders and low-budget remodelers outside the $5M custom profile.

Execution flow

Runner — Validate
```
client_id
```
, load
```
research.json
```
(fail closed if missing or invalid JSON), build ICP payload, eval → if fail, refine once (refine may use internal
```
_research_snapshot
```
) → re-eval → strip internal keys → write only
```
icp.json
```
.
Eval — Structural validation; pass/fail only (orchestration envelope passthrough same as research).
Refine — On fail, fill missing fields from research snapshot or deterministic defaults; must pass eval after refine.

Rules

No external APIs.
No randomness; no
```
Math.random()
```
; stable ordering and string building only.
Fail closed on bad input, missing research, failed validation after refine, path escape, or I/O errors.