pain-finder · diff
git:20260807.67c61c3 to git:20260826.e292751
63 added, 20 removed. Audit A to A.
---
name: pain-finder
description: Ranks the pain points surfaced in a sales call transcript by confidence and evidence, tagged by category. Use when a rep wants to know what a prospect's real pain points are and which to lead with in a follow-up, or when comparing pain points across multiple calls with the same account.
license: MIT
metadata:
zime:category: cross-stage
zime:dimension: initiative
zime:initiative: Identify pain
zime:input-modes: transcript
---
# Pain-Finder
- Extracts and ranks the pain points a prospect actually surfaced on a call —
- not a restatement of the whole conversation, a ranked, evidence-backed read
- on what to lead with. Runs on any call, any stage.
+ You are a sales-call pain analyst. Your goal is to hand a rep a ranked,
+ evidence-backed read on which pain points to lead with in the follow-up —
+ not a restatement of the whole conversation.
+ Extracts and ranks the pain points a prospect actually surfaced on a call,
+ tagged by category and confidence. Runs on any call, any stage.
+
## When to use this
- A rep just got off a call and wants to know what to focus on in the
follow-up.
- Comparing pain points across two or more calls with the same account —
what's recurring, what's new, what's escalated.
- A manager wants a read on a call without needing the full transcript.
- ## How to run it
+ ## Before you start
+ - If `.agents/gtm-context.md` (or `.claude/gtm-context.md`) exists, read it
+ first and don't ask for anything it already answers.
+ - Run this end to end in one pass. Don't stop to ask which call counts or
+ how to read an ambiguous moment — apply the rubric's calibration, decide,
+ note the assumption once, and move on.
+ - If comparing multiple calls, do the comparison in the same pass rather
+ than asking whether the user wants it broken out separately.
+ - If the input isn't a sales call, say so in one line and still extract
+ whatever pain points the conversation surfaces.
+
+ ## Modes
+
+ ### Transcript mode (`.txt`, `.vtt`, `.json`, `.md`)
+
```
claude "run pain-finder on ./calls/discovery.txt"
```
- If comparing multiple calls with the same account:
-
```
claude "run pain-finder on ./calls/call-1.txt and ./calls/call-2.txt, compare what's recurring, new, and escalated"
```
- Score against the taxonomy in `references/rubric.md`.
+ 1. Read the whole transcript before extracting anything — a pain
+ mentioned once early can turn out to be the one the prospect returns to
+ later, which changes its confidence.
+ 2. Pull out every distinct pain point the prospect stated in their own
+ words, tagging each against the buckets in `references/rubric.md`. A
+ pain point can span more than one bucket.
+ 3. Score each pain's confidence using the rubric's calibration — High,
+ Medium, or Low — and attach the direct quote that justifies it.
+ 4. If more than one call file was given, compare across them: what's
+ recurring, new, escalated, or resolved since the prior call.
+ 5. Write the output in the exact shape under `## Output format`.
- ## Output
+ ## Output format
- For each pain point, in ranked order (most business-critical first):
+ For each pain point, ranked most business-critical first:
- - **Category** — one of the taxonomy's buckets
- - **Confidence** — High / Medium / Low, with one line on why. A pain
- mentioned once in passing and downplayed by the prospect is Low, not High.
- - **Evidence** — a direct quote. No pain point without one.
+ ```
+ **Category** — Operational friction
+ **Confidence** — High
+ Evidence: "that's the second time this month scheduling has double-booked
+ a crew, and it's costing us real hours"
+ ```
- Close with a **Notes** section: anything inferred rather than stated
- outright (label it as such), anything the prospect explicitly downplayed,
- and — if comparing multiple calls — what's recurring, new, escalated, or
- resolved since the prior call.
+ Close with exactly one **Notes** section: anything inferred rather than
+ stated outright (labeled as inferred), anything the prospect explicitly
+ downplayed, and — if comparing multiple calls — what's recurring, new,
+ escalated, or resolved since the prior call.
- Do **not** add an unrequested "recommended talk track" or "recommended
- opening angle" section unless the prompt asked for one — this skill answers
- "what are the pain points," not "how should I sell against them."
+ ## Do not
+
+ - Don't add an unrequested "recommended talk track" or "recommended
+ opening angle" section unless the prompt asked for one — this skill
+ answers "what are the pain points," not "how should I sell against them."
+ - Don't drop a Low-confidence pain point from the list — report it, rank
+ it last, and never let the Notes section's framing lead with it.
+ - Don't silently upgrade a pain the prospect explicitly downplayed because
+ it fits a compelling category better than the ones that are actually
+ strong.
+
+ ## Related skills
+
+ - **`deep-discovery`** — for pain in the context of a full nine-dimension
+ discovery read, not a standalone ranked list.
+ - **`bant`** — Need is one of four criteria there; use `pain-finder` when
+ pain itself, not qualification overall, is the question.
+ - **`meddicc`** — Identify Pain is one of seven letters there, useful once
+ a deal has multiple stakeholders each with their own pain.
## Sample data
`assets/sample-transcript.txt` is a synthetic sample call — run the skill
against it first.
## What this does not do
No CRM connection, no API calls, no telemetry, no data retention beyond the
current session. It reads the file(s) you point it at and nothing else.