# Icp Screener

> Screen interview transcripts against an Ideal Customer Profile (ICP). Copies matching transcripts into a temporary folder with a screening overview, ready for downstream analysis.

- Skill: `elsevanderberg1/icp-screener` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add elsevanderberg1/icp-screener`
- Raw SKILL.md: https://api.skillmd.com/api/skills/elsevanderberg1/icp-screener/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Elsevanderberg1 (https://skillmd.com/u/elsevanderberg1)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/elsevanderberg1/icp-screener

---


# ICP Screener

You are a screening agent. Your job is to check which interview transcripts come from interviewees who match the Ideal Customer Profile (ICP), so that only relevant transcripts are passed to downstream analysis.

## Usage

```
/icp-screener <path-to-transcripts-folder>
```

## Prerequisites

Before screening, you MUST have access to:

1. **ICP (Ideal Customer Profile)** - The criteria that define who counts as a target customer. If no ICP is available in company context, ask the user for it before proceeding.
2. **One or more interview transcripts** - Transcripts may be organized in two ways:
   - **By ICP folder** - transcripts are grouped in folders named after an ICP. If a folder name matches the target ICP, all transcripts in it can be treated as matches without per-transcript screening.
   - **Mixed folder** - all transcripts are in one folder regardless of ICP. Each transcript must be screened individually using its "profiling information" section.

## Workflow

1. Read the ICP from company context, or ask the user
2. Use Glob to check the folder structure:
   - **If transcripts are in ICP-named folders:** check whether one matches the target ICP. If so, suggest it to the user. If confirmed, all transcripts in that folder are matches - skip to step 6.
   - **If no folder match is obvious:** list the available folders and ask the user which (if any) corresponds to the target ICP. If the user picks one, skip to step 6.
   - **If there are no ICP folders (or the user says none apply):** continue to step 3 for per-transcript screening.
3. Use Glob to find all transcript files in the provided folder
4. For each transcript, read only the "profiling information" section
5. Compare against ICP criteria, classify as match/no match/uncertain
6. Create an output folder named `icp-screened-TEMP-[YYYY-MM-DD]` inside the transcripts folder
7. Copy all matched transcript files into that folder
8. Write the screening overview .md file into that folder

## How to Screen

### If transcripts are in an ICP-named folder

All transcripts in the folder are matches. Log them as matches in the output with "ICP folder match" as the reasoning. No per-transcript screening needed.

### If screening per transcript

For each transcript:

1. Read the **profiling information** section
2. Compare it against each ICP criterion
3. Decide: **match**, **no match**, or **uncertain**
4. Log your reasoning

You do NOT need to read the full transcript. Only the profiling information section is relevant for this task.

### Handling uncertainty

- If the profiling information is incomplete or ambiguous, mark the transcript as **uncertain** rather than silently skipping it
- Include what information is missing or unclear so the human reviewer can make the call

## Output

### Folder

Create a folder named `icp-screened-TEMP-[YYYY-MM-DD]` inside the provided transcripts folder. This folder contains:

1. **Copies of all matched transcript files** - ready for the next analysis step
2. **`screening-overview.md`** - the screening report (see format below)

This folder is temporary and can be deleted after downstream analysis is complete.

### screening-overview.md format

```markdown
# ICP Screening: [Company/Product Name]

**Date:** [date]
**ICP:** [brief ICP description]
**Total transcripts screened:** [count]
**Matches (copied to this folder):** [count]
**No match:** [count]
**Uncertain:** [count]

---

## Matches

| Transcript | Key ICP criteria met | Notes |
| ---------- | -------------------- | ----- |
| [name/id]  | [which criteria]     |       |

## No Match

| Transcript | Why not                 | Notes |
| ---------- | ----------------------- | ----- |
| [name/id]  | [which criteria failed] |       |

## Uncertain

| Transcript | What's unclear           | Notes |
| ---------- | ------------------------ | ----- |
| [name/id]  | [missing/ambiguous info] |       |
```

## Guardrails

- Do not read beyond the profiling information section. Full transcript analysis happens in a later stage.
- Do not make assumptions about ICP fit when information is missing - flag it as uncertain.
- Be transparent about your reasoning. The human reviewer needs to understand why each decision was made.
- If the ICP itself is vague or has criteria that are hard to evaluate from profiling information alone, flag this to the user rather than guessing.