consult-session-notes git:20260927.31de767A

Immutable. This exact content never changes and is served at /api/v1/blob/e722621ceb521a43.

---
name: consult-session-notes
sheetId: "2.39"
description: >-
  Turn a consulting-session transcript or messy notes — a discovery call or a
  shadowing observation — into the four-section session summary (Summary Update ·
  Progress · Blockers · What's Next) PLUS a 痛點 ledger (one row per pain actually
  voiced or observed), filed as a [Notes] Google Doc into the engagement's EXISTING
  numbered [N] Drive folder with a note on the CRM deal; the ledger is the raw
  material /consult-brd-writer consumes. Trigger on /consult-session-notes or when
  Peter says "整理這次客戶訪談筆記", "整理 shadowing 紀錄", "把逐字稿變成會議摘要",
  "整理客戶會議筆記", "structure this client session", "tidy the shadowing notes",
  "session summary for the client project" — fire eagerly whenever session material
  belongs to a CLIENT engagement. Distinct from project-note-specialist (the
  generic source skill — ONLY tidies text, files nothing, touches no CRM; for
  non-engagement notes), consult-project-specialist (creates a NEW deal + folder
  from a sales meeting; THIS skill updates an EXISTING engagement, never creates
  either), consult-transcriber (upstream — produces the transcript this skill
  structures), and consult-brd-writer (downstream — consumes the ledger).
category: sales-consultant
project: consult-session-notes
platform: claude
status: Done
author: Peter Tu
input: "A Fireflies meeting (id/URL/name — preferred, no paste), pasted text, a Google Doc link, or consult-transcriber's filed transcript, + the engagement (deal URL / company)."
process: "Collect input + resolve the engagement → structure the four-section session summary → extract the 痛點 ledger → create the [Notes] Doc in the [N] folder → CRM note on the deal → hand the ledger to consult-brd-writer"
output: "A filed [Notes] session-summary Doc (four sections + 痛點 ledger) in the engagement folder, linked on the CRM deal"
synergy:
  - "consult-transcriber"
  - "consult-project-specialist"
  - "consult-brd-writer"
  - "project-note-specialist"
house-style: bound

---

# Consult Session Notes

```bash
npx skills add https://github.com/peter-tu-zynkr/zynkr-skill-builder --skill consult-session-notes
```

Source: forked from `project-note-specialist` (3.08) — itself the standalone
promotion of the admin-meeting-note agent (sheetId 3.04), original prompt in [this
Google Doc](https://docs.google.com/document/d/1oW7enyJtGfUeRlnl63pxhjJ89dMZQzHlBEOeap7039c/edit).

A consulting session — a discovery 訪談, a shadowing observation, a working 會議 —
leaves two things worth keeping: what happened, and what hurts. This skill turns
the raw transcript or messy notes into both: the four-section session summary the
source skill defines (**Summary Update · Progress · Blockers · What's Next**), plus
a **痛點 ledger** — one table row per pain the client actually voiced or Peter
actually observed — filed together as one `[Notes]` Google Doc in the engagement's
existing numbered `[N]` Drive folder, with a note on the CRM deal. The ledger is
the consult-specific addition and the point of the fork: its rows are the raw
material /consult-brd-writer turns into `R-n` requirements, so every row carries
its source — who said it, or which transcript section.

## How this differs from its neighbours

- **project-note-specialist** — the generic source skill. ONLY tidies text into
  the four sections: files nothing, touches no CRM. For non-engagement notes.
- **consult-project-specialist** — creates a NEW deal + `[N]` folder from a
  sales meeting. THIS skill updates an EXISTING engagement, never creates either.
- **consult-transcriber** — upstream: produces the transcript this skill structures.
- **consult-brd-writer** — downstream: consumes the 痛點 ledger (and the rest
  of the discovery pile) to write the BRD/PRD.

## Fixed facts (don't re-derive these)

- **Google account** for all Gmail/Drive/Docs tools: `<your-google-workspace-account>`
- **Drive parent folder** (`[2.2] 業務與顧問部門：專案`, where numbered project folders live): `1hkXPX7OXPFOU0BcloPbJSFp8O0zArM8t`
- **CRM deal URL** for the report/backlink: `https://platform.zynkr.ai/deals/{deal_id}`

## Hard rules

1. **Never create a folder or a deal.** No `[N]` folder → STOP and point at
   /consult-intake (inbound lead) or /consult-project-specialist (meeting
   debrief). One numbered workspace per engagement is the 2.x invariant.
2. **Never invent a pain.** Ledger rows come only from pains actually voiced or
   observed in this session's material; a thin session yields a short ledger.
3. **Empty summary sections say 「本次無相關內容」** — the per-session adaptation
   of the source skill's "No updates this week" rule. Never pad.
4. **Client-facing email is ALWAYS a Gmail draft** — if Peter asks to send the
   summary out, `mcp__google-workspace__draft_gmail_message`. Never send.

---

## Workflow

### 1 · Collect the input and resolve the engagement

The session material arrives four ways. **Prefer Fireflies** — it needs no paste
and no upload, so reach for it before asking Peter for anything.

- **A Fireflies meeting** — Peter names it ("整理範例科技那場的筆記"), gives a
  meeting id, or an `app.fireflies.ai/view/<id>` URL. Resolve with
  `mcp__fireflies__fireflies_search(query="keyword:\"範例科技\" from:2026-09-01")`,
  confirm the row by title + date + attendees, then
  `mcp__fireflies__fireflies_get_transcript(transcriptId="<id>")` — it returns
  speaker-labelled, timestamped sentences (`[12:43 - 14:00] 高May: …`), which is
  exactly the shape §2 and the 痛點 ledger want. Cite a moment with
  `https://app.fireflies.ai/view/<id>?t=<seconds>`.
- **Pasted text** — use directly.
- **A Google Doc link** — read with
  `mcp__google-workspace__get_doc_content(user_google_email="<your-google-workspace-account>", document_id="<id>")`.
- **A handoff from consult-transcriber** — the transcript Doc it just filed
  (already in context); read it the same way.

⚠️ **Fireflies request budget.** The free plan allows **50 API requests per day**
and every tool call spends one. Resolve in a single `fireflies_search`, fetch the
transcript once, and never loop over meetings. If a search returns nothing, ask
Peter rather than paging through history.

Then resolve the engagement (the standard 2.x pattern):

- **Deal** — from a `…/deals/{id}` URL, or by company name. Prefer
  `mcp__zynkr__get_deal` / `mcp__zynkr__list_deals`
- **Folder** — the deal's `notes` carry a `專案資料夾：<url>` backlink (written
  by consult-intake / consult-project-specialist); extract the folder id. If
  missing, list the parent (`mcp__google-workspace__list_drive_items`,
  folder_id `1hkXPX7OXPFOU0BcloPbJSFp8O0zArM8t`) and match `[N] Company（…）`.
- **No folder at all** → STOP (hard rule 1).

Classify **SESSION_TYPE** from the material: interviewer-and-client Q&A → `訪談`;
narrated observation of someone working → `shadowing`; anything else → `會議`.
Infer; ask only if genuinely ambiguous.

### 2 · Structure the four-section summary

Preserve the source skill's format faithfully — same sections, order, labels:

```
{{COMPANY}} — {{SESSION_TYPE}} {{YYYY-MM-DD}}

1. Summary Update      — a concise overview of this session's key activities or themes
2. Progress            — specific actions completed, decisions made, facts confirmed
3. Blockers / challenges — issues or delays surfaced, with root causes or dependencies
4. What's Next         — planned actions, decisions needed, priorities agreed in-session
```

(In the real output each label sits on its own line, content beneath it — the worked example below shows the exact rendered shape.)

Rules carried over from the source, unchanged in spirit: reorganize scattered
ideas into clear bullets or short paragraphs; if input is unclear, infer context
logically but do NOT fabricate; professional but conversational tone, written for
internal stakeholders; a category with no data gets 「本次無相關內容」 (hard rule 3).

### 3 · Extract the 痛點 ledger

The fork's addition — a table under the four sections, one row per pain:

```
## 痛點 Ledger

| 痛點 | 現況流程（as-is 片段） | 影響（時間/錯誤/成本） | 頻率 | 來源（誰說的／哪段逐字稿） |
|------|----------------------|----------------------|------|--------------------------|
```

- Only pains actually voiced or observed (hard rule 2).
- **影響** — quantify only with numbers the client gave (minutes, error counts,
  NT$); otherwise describe qualitatively. Never estimate on their behalf.
- **來源** — a named person and/or transcript section reference, so
  consult-brd-writer can tag the `R-n` requirement it becomes.

### 4 · Create and file the `[Notes]` Doc

Title: `[Notes] {{COMPANY}} — {{SESSION_TYPE}} {{YYYY-MM-DD}}`. Content: the
step-2 summary + the step-3 ledger. Create via the reliable two-step (creating
a Doc directly in a folder via `create_drive_file` returns HTTP 400):

```
## 1. create the doc (lands in My Drive root)
mcp__google-workspace__create_doc(user_google_email="<your-google-workspace-account>",
  title="[Notes] {{COMPANY}} — {{SESSION_TYPE}} {{YYYY-MM-DD}}", content="<summary + ledger>")
## 2. move it into the engagement folder
mcp__google-workspace__update_drive_file(user_google_email="<your-google-workspace-account>",
  file_id="<doc id from step 1>", add_parents="<target folder id>")
```

Target folder: for `shadowing` sessions, look inside the `[N]` folder for a
`Shadowing — YYYY-MM-DD` subfolder (via `list_drive_items`) and file there when
it exists; otherwise — and for all other session types — the `[N]` folder
itself. Never create the subfolder.

### 5 · CRM note on the deal

Append to the deal's notes — `mcp__zynkr__update_deal` preferred; SQL fallback:

`mcp__zynkr__update_deal` REPLACES `notes` wholesale, so append in three steps:

1. `mcp__zynkr__get_deal(id="<deal_id>")` — read the current `notes`
2. build the new value: the existing notes, then a blank line, then the block below
3. `mcp__zynkr__update_deal(id="<deal_id>", notes="<combined>", confirm=true)`

Call it once without `confirm` to preview, then again with `confirm=true`. Never
send `notes` without the existing text in front of it — the field is overwritten,
not appended, and skipping the read loses every earlier backlink.

### 6 · Report and hand off

A compact artifact table, then the headline:

```
本次 session 已歸檔：宏宇精密 — 訪談 2026-08-03

| 產出 | 內容 |
|------|------|
| 文件 | [Notes] 宏宇精密 — 訪談 2026-08-03（<doc url>）|
| 資料夾 | [4] 宏宇精密（報價流程自動化）|
| CRM | <deal url> — notes 已附文件連結 |
| 痛點 | 2 筆（皆含來源標註）· ledger 可直接餵給 /consult-brd-writer |
```

## Worked example

Input fragment (a 宏宇精密 discovery call):

```
（訪談片段 1）王小明（業務主管）：報價都是我自己用 Excel 拉的，一張大概要
四十分鐘，有時候型號打錯客戶就直接退回來……上週就退了兩張。對了，下週我們
會把新的價格表定下來，到時候要再對一次。喔還有，倉庫出貨單還是手抄的，
月底對帳都要花一個下午。
```

The four sections:

```
宏宇精密 — 訪談 2026-08-03

1. Summary Update
報價與出貨兩條流程皆為手動作業，錯誤與月底對帳成本是主要痛點；新價格表下週定案。

2. Progress
- 完成報價流程訪談：確認現況為 Excel 手動拉單，單張約 40 分鐘；新價格表下週定案

3. Blockers / challenges
- 報價型號手動輸入易錯，上週兩張報價單遭客戶退回
- 出貨單仍為手抄，月底對帳需耗費一個下午

4. What's Next
- 新價格表定案後重新核對報價基準；評估兩條流程自動化的優先順序
```

The ledger (2 rows — exactly what the session supported, no more):

| 痛點 | 現況流程（as-is 片段） | 影響（時間/錯誤/成本） | 頻率 | 來源（誰說的／哪段逐字稿） |
|------|----------------------|----------------------|------|--------------------------|
| 報價單手動製作、型號易打錯 | 業務以 Excel 手動拉報價單 | 每張約 40 分鐘；上週 2 張因型號錯誤遭退回 | 每張報價 | 王小明（業務主管）· 訪談片段 1 |
| 出貨單手抄、月底對帳耗時 | 倉庫手寫出貨單，月底人工對帳 | 每月對帳約半個下午 | 每月 | 王小明轉述倉庫現況 · 訪談片段 1 |

## Why it's built this way

- **A full fork, not a delegation wrapper.** The source's entire mechanics are a
  ~20-line output format; wrapping would put half of every run (filing, ledger,
  CRM) behind another skill's contract for no saving. The format is copied once,
  kept faithful — labels unchanged, because Peter and downstream readers already
  parse that shape from every project-note-specialist output.
- **Ledger and summary share one Doc** so consult-brd-writer reads a single
  artifact per session, and the summary gives every ledger row its context.
- **Pains are never invented** because a fabricated ledger row becomes a
  fabricated `R-n` requirement a client is later asked to sign off.
- **The folder is resolved, never created** — the same 2.x invariant as every
  sibling: a second folder for the same client would fork the record.

## Inference defaults (Peter overrides by just saying so)

- **SESSION_TYPE** → inferred per step 1 (訪談 / shadowing / 會議).
- **Session date** → the date stated in the material; fallback today.
- **Doc language** → zh-TW body; the source skill's English section labels kept.
- **Empty sections** → 「本次無相關內容」.
- **Shadowing subfolder** → used when it exists, never created.
- **Deal stage** → untouched; this skill only appends a note.

## Provenance

Forked from `project-note-specialist` (3.08) @ b6bfb04c (2026-08-03) — consult
adaptation; diverges by design. The source is itself the standalone promotion of
the admin-meeting-note agent (sheetId 3.04).
Re-sync check: `git log --oneline b6bfb04c..origin/main -- skills/3-operations/project-note-specialist/`
→ review hits → port or waive with a dated line here.

## Reference files

None — the summary template and the 痛點 ledger table live inline in this file
(steps 2–3), so a standalone marketplace install is self-contained by design.

## Limitations

- One session per run; a multi-session digest = multiple runs, one Doc each.
- The ledger is only as good as the session — it never backfills pains from
  memory, other deals, or industry priors (hard rule 2).
- Requires an existing engagement; it will not bootstrap a deal or folder.
- It does not transcribe audio (consult-transcriber, upstream) and does not
  write requirements (consult-brd-writer, downstream).

## House style

Writing style is **not owned by this file**. The house voice lives in two Google Docs under
`[@] 寫作指南` (`12DBdFz3SK22ie9im_ThFMI7IBRXsTZsV`), read at runtime:

- 《[2.0] Zynkr 通用風格指南 House Voice》 `10bOIQwRm9Pxwgct4hlwCwK_B4Pipai1HqBPZKzyRHSE` —
  the universal core, plus the addendum for this surface
- 《[3.2] 禁用詞清單 Forbidden Words》 `1N5sHLP4qzmmhpCGsi6KElxi1z0MFe4QZ0Q_35T10Uyg`

Read both before producing client- or reader-facing text, and scan the draft against 《[3.2]》
before handing it over. If Drive is unreachable, say so in the output rather than proceeding
unchecked. Never re-implement either list inside this file.