Planning Tracker Builder

---
name: planning-tracker-builder
sheetId: "0.07"
description: >-
  Build or extend Zynkr's planning Main Tracker (the cycle's system of record)
  from a NORMALIZED item list — the planning-session-synth handoff or a pasted
  table. Fresh cycle: copy the template tracker (never edit it) → 「<year>
  <cycle> Planning Main Tracker」 → cycle-named tabs → fill the SOR tab with L1
  header rows + N.NN numbering, 重要×緊急 → Priority formula, the 專案項目小記
  pivot, 狀態 colours. Existing tracker: fill the empty SOR tab session-synth
  left, append rows under the right L1 block, or add a new tab — never
  overwrite the SOR tab in place, never renumber. Prints
  a LINT REPORT (P0 cap > 6 / > 25%, owner > 3 P0, items 掛 All, missing
  owners/dates, L2 off-taxonomy, duplicate 項目, P3 not 放棄) and the C1 cash
  pre-mortem before writing; lint warns, never blocks.
  Trigger on /planning-tracker-builder or "建 tracker", "把項目清單做成 Main
  Tracker", "建立 YE Planning Main Tracker", "把正規化清單寫進 tracker", "把這些專案
  項目排進 tracker", "build the planning tracker", "extend the tracker with these
  items", "lint the project list". Distinct from planning-session-synth
  (upstream — transcript + whiteboard → normalized list), planning-tracker-sync
  (downstream weekly READ; never builds), planning-suite-reconciler (docs/OKR
  sheet reconcile TO the tracker) and project-status-update (course-project
  weekly; owns bare 「週報」).
category: strategy
project: planning-tracker-builder
platform: claude
status: Done
author: Peter Tu
input: "A normalized item list (session-synth handoff or pasted rows: L1/L2 · 項目 · 重要 · 緊急 · 負責人 · dates · 備註) + owner/priority decisions · cycle (H1/H2/YE) · optional existing tracker ID"
process: "Resolve cycle + sources → ingest, normalise, number N.NN → lint report + C1 cash pre-mortem → row plan → on explicit go: copy template / fill or extend existing → SOR tab (RAW) + Priority formula, 專案項目小記 pivot, 狀態 colours → read back → report wrote / did-not"
output: "A cycle-named Main Tracker Sheet (or new rows/tab in an existing one) — SOR tab · pivot · colours — plus the printed lint report, C1 section, row plan and wrote/did-not summary"
synergy:
  - "planning-session-synth"
  - "planning-tracker-sync"
  - "planning-suite-reconciler"
house-style: bound

---

# Planning Tracker Builder

```bash
npx skills add https://github.com/peter-tu-zynkr/zynkr-skill-builder --skill planning-tracker-builder
```

After a planning session the room has a list of projects; the company needs
the tracker that the weekly meeting, the OKR sheet and every plan Doc reconcile
to. This skill turns a normalized item list into that tracker — a copy of the
proven template with cycle-named tabs, L1 header rows, `N.NN` numbering, the
重要×緊急 → Priority formula, the 專案項目小記 pivot and 狀態 colours — or fills
the empty SOR tab of the file planning-session-synth created, or extends an
existing tracker without disturbing what is already numbered. It is the one
skill in the family that lints the list and prints the C1 cash pre-mortem
before a single cell is written; the lint warns, the founder decides, and
nothing is written until they say go.

---

## How this differs from its neighbours

- **planning-session-synth** (upstream) — digests the transcript and whiteboard
  photos into `回顧總結` / `② 白板原文` / `③ 去重與歸類決策` / `④ MECE 檢查` and a
  normalized item list. It writes the SOR tab itself only when the user says so;
  the default handoff is to this skill, which owns numbering, priority, pivot,
  colours and the lint. When session-synth has already created this cycle's
  tracker (`create_spreadsheet` with all seven tabs, SOR + pivot left empty),
  this skill **fills** those two empty tabs in place (mode = `fill`, Step 6) —
  it never copies the template into a second file beside it.
- **planning-tracker-sync** (downstream, weekly) — READS the SOR tab this skill
  built and produces the Team Weekly block, nudges and snapshots. It never
  builds or renumbers, and it writes **nothing** into the Main Tracker: its
  only write is `tracker-latest` + `tracker-snapshots` in the OKR & KPI
  Tracker (spec row 0.09). Nothing in this file is reserved for it.
- **planning-suite-reconciler** (downstream, docs) — reconciles the plan Docs and
  the OKR & KPI Tracker TO the finalized tracker. This skill never touches a
  Doc; the reconciler never touches the tracker.
- **project-status-update** — the course-project weekly on its own tracker
  Sheet; owns bare 「週報」/「weekly report」. Unrelated Sheet, unrelated trigger.

## Fixed facts (read the references first)

- **Knowledge pack** `./references/planning-knowledge-pack.md`: taxonomy L1/L2
  (§2), priority rule + status vocabulary + do-now cap (§3), C1–C4 frame (§4),
  tracker tab layout (§6), MECE precedents (§7), versioning (§8), never-list (§9).
- **Sources** `./references/planning-sources.md`: the Google account, the
  template = the current-cycle Main Tracker (§A, with tab → gid map), the hub
  folder (§A). Every ID this skill uses comes from that file or the user — none
  is hard-coded here.
- **Layout contract** `./references/tracker-layout.md`: exact column map A–M,
  the Priority formula string, the pivot grid + formulas, colour defaults, the
  lint rule table (T1–T8), the C1 keyword list, and the report shapes.
- **Tools** (`google-workspace` MCP): `copy_drive_file` · `get_spreadsheet_info`
  · `read_sheet_values` (`include_formulas=true`) · `create_sheet` (new tab, or
  duplicate a tab under a new name; pass `insert_sheet_index` so the tab lands
  in the pack §6 order — else list the reorder as a leftover) ·
  `modify_sheet_values` (write / `clear_values=True`; `value_input_option`
  `RAW` for **every data column**, `USER_ENTERED` for **every formula write** —
  column G Priority AND the whole `專案項目小記` grid: COUNTIF / COUNTIFS / SUM /
  ratios) · `resize_sheet_dimensions` (`insert_rows_at`, freeze — **always with
  `sheet_name`**: the MCP defaults to the FIRST tab, i.e. README; after every
  row insert, reset the new rows' format with `format_sheet_range` bold=False +
  background cleared so header formatting does not bleed) · `format_sheet_range`
  · `manage_conditional_formatting`. There is **no rename-sheet or delete-sheet
  call**: cycle-named tabs are *duplicates* of the template tabs; the old tabs
  are emptied and listed as a manual leftover.
- **Numbering (matches the live tracker, layout §1):** the pack §2 L1 number
  lives in the `主類別` TEXT (`6.0 Tech & Platform`); the `#` prefix of an L1
  block is **positional** — sequential by presence (`1.0`, `2.0` … in the order
  the blocks appear; the July tracker's `6.0 Tech & Platform` block carries `#`
  `5.0` and items `5.01`–`5.04` because no Dev-Ops block exists). Items are
  `<prefix>.01`, `<prefix>.02` … (block prefix + two-digit serial), ordered by
  L2 prefix then input order. Fresh / fill: all eight blocks are written, so
  positional prefix = pack §2 number (`1.0`–`8.0`). Extend: locate blocks by
  `主類別` text (fallback `#` matching `^\d\.0$`), continue serials from
  `max(existing serial under that block's # prefix) + 1`, and an L1 with no
  block yet gets a **new header block appended** whose `#` prefix = last
  existing prefix + 1 and whose `主類別` text carries the pack §2 number. This
  rule is printed in every row plan. `#` is **text** (the live tracker stores
  `'1.10'`, `'2.0'`): written `RAW`, never `USER_ENTERED`, or `1.10` becomes
  1.1 and `1.0` becomes 1.

## Hard rules

1. **Print before writing.** The lint report, the C1 section and the row plan
   are printed in the chat and the user says go — no Sheet call happens
   before that. Lint **warns, never blocks**; the founder overrides by saying
   so and the override is written into `備註` or the README.
2. **Never overwrite the SOR tab in place, never renumber.** Existing tracker ⇒
   `fill` only when the SOR tab is empty (no header — the session-synth
   handoff) **and** `專案項目小記` is empty or missing; else append rows under
   the right L1 block (serial continues from the last used under that block's
   `#` prefix; a new L1 = a new block appended, prefix = last + 1) or add a new
   tab. Existing `N.NN`s and existing `#` prefixes are never rewritten. A
   `專案項目小記` that already holds a grid belongs to a live cycle: never
   rewrite it — add a new grid two rows below it or a new tab — and never
   touch the README counts of another cycle.
3. **The template is read-only.** Fresh build = `copy_drive_file` of the
   template tracker (sources §A) and every write goes to the copy. If the
   copy step fails, stop. Never copy the template when the user named an
   existing tracker (fill/extend) — one cycle, one file.
4. **No invented owners, dates or numbers.** An owner not decided is `未定`
   (a T4 finding), a date not given is the literal `YYYY-MM-DD`, a runway
   number not supplied is `（待補）`. Owners come from the user's decisions or
   the tracker's 負責人 column — never from memory.
5. **Sheet only.** No Doc edits (planning-suite-reconciler), no mail
   (planning-session-synth drafts the recap), no calendar. Every run ends by
   printing what it wrote (file ID/URL, tabs, row ranges) and what it did NOT
   do (leftovers, skipped rows, overrides).
6. **Every call names its tab.** Every `range_name` carries the `'<tab>'!`
   prefix and every `resize_sheet_dimensions` passes `sheet_name=<tab>` — an
   un-prefixed call lands on the FIRST tab (README in the pack §6 order), and
   a row insert there followed by a prefixed write is exactly the in-place
   SOR overwrite rule 2 forbids. Every data column is written `RAW`; **every
   formula write** — the column-G Priority call AND the whole `專案項目小記`
   grid (COUNTIF / COUNTIFS / SUM / ratio cells) — is `USER_ENTERED`, or the
   formulas land as literal text.

---

## Workflow

### Step 0 — Resolve cycle + sources

Read `./references/planning-sources.md`. Take `cycle` (`H1` / `H2` / `YE`) and
the planning `year` from the user; take any ID overrides (template tracker,
hub folder, existing tracker to extend). Defaults: template = the sources §A
Main Tracker; parent folder = the sources §A hub folder; Google account = the
one named in the sources file. Print one line:
`Operating on cycle <year> <cycle> · mode = fresh|fill|extend · template <ID> · parent <ID>`.
Mode: no tracker named ⇒ `fresh`. The user names an existing tracker (or says
「加進現有 tracker」, or hands over the session-synth file) ⇒ `get_spreadsheet_info`
+ `read_sheet_values("'<cycle> 專案項目'!A1:M3")` + `read_sheet_values
("'專案項目小記'!A1:K3")`: SOR tab missing or empty (no header row) **and**
`專案項目小記` missing or empty ⇒ `fill` (the synth handoff — write into that
file, Step 6); SOR header row present ⇒ `extend`; SOR tab empty but
`專案項目小記` already holds a grid ⇒ the file carries a live cycle — `extend`
via the new-tab branch (Step 6.4: fill the empty SOR tab, second pivot grid
below the first, README counts untouched). Never `fresh` on a named file —
that would split the cycle across two Sheets.

### Step 1 — Ingest the normalized item list

Accept (a) the planning-session-synth handoff (a chat block, or its `③ 去重與
歸類決策` + `④ MECE 檢查` tabs via `read_sheet_values`) or (b) a pasted table.
Map every row onto the item schema in `./references/tracker-layout.md` §1
(`L1 · L2 · 項目（正規化）· 重要 · 緊急 · 負責人 · 協助者 · 開始 · 結束 · 狀態 · 備註`).
`重要`/`緊急` accept `Y/N`, `✓/✗` etc. and are normalised to the exact pack §3
strings; unrated stays blank (Priority stays blank) and is noted. Owner and
priority **decisions** come from the user here — record them, never fill gaps
yourself. Rows marked 排除 in ③ (layout labels, arrows) are dropped and listed
under 「未寫入」 in the final report.

### Step 2 — Normalise, classify, number

- **L1/L2** — match on the numeric prefix (`2.2`), not the wording (the July
  tracker shortened some L2 labels). Unknown L2 ⇒ keep the user's text, raise
  T6 with the nearest pack §2 row as a suggestion. Apply pack §7 rulings as
  precedent when two rows look alike; every merge/split you propose is a lint
  line, not a silent edit.
- **Order** — group by L1 (pack §2 `1.0` → `8.0`, always all eight headers
  even when a block is empty — an empty 5.0/6.0/8.0 is echoed as a coverage
  gap, pack §2), then by L2 prefix, then input order.
- **Number** (Fixed facts + layout §1) — fresh / fill: all eight blocks are
  written in pack §2 order, so `#` prefix = pack §2 number; items `1.01`,
  `1.02` … per block. Extend: the block prefix is whatever the live tracker
  carries for that `主類別` (positional — read it, never derive it from the
  pack), serials continue from `max(existing serial under that prefix) + 1`;
  an L1 with no block yet ⇒ new header block appended after the last one,
  prefix = last existing prefix + 1, `主類別` text = the pack §2 label with its
  own number (e.g. `#` `7.0` · `主類別` `8.0 Finance & Admin`).
- **Priority** — the formula in `./references/tracker-layout.md` §2 (the
  template's own): blank when either input is blank, else P0/P1/P2/P3 per
  pack §3. Values are also acceptable when the user asks for a formula-free
  tab; say which you used.
- **Defaults (only where the user gave nothing)** — `狀態` = `未開始`; P3 ⇒
  `狀態` = `放棄` and 負責人/協助者 = `N/A` **when none given** — an owner the
  user assigned to a parked P3 is kept as-is and echoed in the lint header
  (「P3 保留 owner：<#s>」, revisit at the mid-cycle gate, pack §3); dates =
  literal `YYYY-MM-DD` when not given.

### Step 3 — Lint report + C1 cash pre-mortem (print, do not write)

Run the list through the T1–T8 rules in `./references/tracker-layout.md` §4 and
print the report in the §6 shape — header (cycle · item count · items per L1 ·
P0/P1/P2/P3 counts and %), then one line per finding with the item numbers,
then the **C1 cash pre-mortem**: every P0 whose 項目/備註 matches the §5 spend
keywords (hire · ads · venue · tooling · contractor · bonus/分潤 …), listed with
負責人 and 開始, followed by the question 「以上 P0 進場後，C1 底線（pack §4：runway
≥ 4 個月）是否仍成立？runway 目前 = <user number or（待補）>」. State which cap
rule the room chose (≤ 6 owned P0 vs. the July 25%). Wait for the user: fix
rows, override with a reason, or accept as-is.

### Step 4 — Row plan + confirmation

Print the row plan (shape in `./references/tracker-layout.md` §6): target
file (new copy name / existing ID) · tab names · for each L1 block the header
row and the item rows `first # – last #` · the **numbering rule line**
(「`#` 前綴依位置排序（現有 <prefix> = <主類別> …）；主類別文字帶 pack §2 編號；
新 L1 → 前綴 <last+1>」 — in extend mode printed with the prefix → 主類別 map
read from the tracker) · pivot owner rows · colour rules · tabs to
duplicate/empty · leftovers you already foresee. Then ask for an explicit go.
Nothing below runs without it.

### Step 5 — Fresh build (mode = fresh)

1. `copy_drive_file(file_id=<template>, new_name="<year> <cycle> Planning
   Main Tracker", parent_folder_id=<hub or user folder>)` → new ID; print it.
   `get_spreadsheet_info` on the copy; confirm the pack §6 tabs exist.
2. Cycle-named tabs: `create_sheet(spreadsheet_id, source_sheet_name="<template
   SOR tab, sources §A>", sheet_name="<cycle> 專案項目", insert_sheet_index=
   <index of the template SOR tab + 1>)`, same for the template retro tab →
   `<cycle> 回顧總結` (index of the template retro tab + 1) — a duplicate keeps
   widths, header format, formulas, conditional rules, and the index keeps
   the pack §6 order (README · 回顧總結 · 專案項目 · 專案項目小記 · ② · ③ · ④); if
   the order still differs, list the reorder as a leftover. **Name collision:** when the target name
   already exists in the file (cycle = `H2` built from an `H2 …` template,
   or `H1`), Sheets rejects the duplicate — use the pack §8 form
   `<cycle> 專案項目 — YYYY-MM 現行版` (and `<cycle> 回顧總結 — YYYY-MM 現行版`)
   instead, and use that exact name wherever `<cycle> 專案項目` appears below
   (pivot references, colour ranges, README, report). Empty the old tabs
   (`modify_sheet_values(range_name="'<old>'!A2:M1000", clear_values=True)`)
   and list them as leftovers (「請在 UI 刪除或改名為 (Archive) …」). Also empty
   `② 白板原文` · `③ 去重與歸類決策` · `④ MECE 檢查` below their header row 1
   (`'<tab>'!A2:Z1000`), and both retro tabs — the duplicated `<cycle> 回顧總結`
   AND the old template retro tab — in two ranges: `A1:F6` (the previous
   cycle's 5 重點結論) and `A9:F1000` (the table body below the header at row
   8; keep row 8) — session-synth fills those; a fresh copy carries nothing
   of the previous cycle.
3. SOR tab (all ranges prefixed `'<cycle> 專案項目'!`): **grid pre-flight** — read
   each tab's row/column count from `get_spreadsheet_info` (the July template is
   `65×13` SOR · `10×11` 小記 · `55×6` retro); every clear/write range below is
   bounded by that grid (write `A2:M<rows>`, not `A2:M1000`), and when the plan
   has more rows than the grid, `resize_sheet_dimensions(sheet_name=…,
   insert_rows=<needed>)` FIRST — Sheets rejects `values.update` beyond the grid
   with "exceeds grid limits". Then clear `A2:M<rows>` and
   **two writes** — (a) header row (layout §1) + every planned row for columns
   A–M with G left blank, one `modify_sheet_values` call,
   `value_input_option="RAW"` (so `1.0`, `1.10`, `YYYY-MM-DD` stay literal
   text — Hard rule 6); (b) `G2:G<last>` with the layout §2 formula string
   per item row (header rows blank), `value_input_option="USER_ENTERED"` so it
   evaluates. Freeze row 1: `resize_sheet_dimensions(sheet_name="<cycle> 專案
   項目", frozen_row_count=1)`.
4. `專案項目小記`: rewrite the pivot per layout §3 with every `'<template SOR>'`
   reference re-pointed to `'<cycle> 專案項目'`; owner rows = the distinct 負責人
   values (including `All` when present, so the smell stays visible). The
   whole grid is labels + formulas — one `modify_sheet_values` call with
   `value_input_option="USER_ENTERED"` (COUNTIF / COUNTIFS / SUM / ratio cells
   must evaluate; Hard rule 6). Then **clear every row below the grid**
   (`modify_sheet_values(range_name="'專案項目小記'!A<grid end + 1>:Z<tab rows>",
   clear_values=True)` — `+1`, not `+2`: a smaller owner list would otherwise
   leave one stale owner row alive) — the copy may carry stray rows from the
   previous cycle and they must not travel into the new file. Nothing else writes
   here: planning-tracker-sync's snapshots go to the OKR & KPI Tracker
   (`tracker-latest` / `tracker-snapshots`), never to this tab.
5. Colours: keep the duplicated tab's 狀態 rules if present; otherwise add the
   four `CUSTOM_FORMULA` rules in layout §3 via `manage_conditional_formatting`
   on `'<cycle> 專案項目'!A2:M1000`. L1 header rows bold + light fill via
   `format_sheet_range` on `'<cycle> 專案項目'!A<r>:M<r>`.
6. README: if it still holds the previous cycle's content, replace the
   key·value rows (source · method · counts · confidence `（待補）` · coverage
   gaps from Step 2 · tab guide); if session-synth already wrote it, refresh
   only the counts line.

### Step 6 — Fill or extend an existing tracker (mode = fill | extend)

1. `get_spreadsheet_info` on the tracker the user named, then
   `read_sheet_values("'<SOR tab>'!A1:M1000")` (`<SOR tab>` = `<cycle> 專案項目`,
   or the tab the user points at) and `read_sheet_values("'專案項目小記'!A1:K40")`.
   Classify:
   - **SOR tab missing / row 1 empty, AND `專案項目小記` missing or empty** ⇒
     **fill** (the session-synth handoff: `create_spreadsheet` left
     `<cycle> 專案項目` + `專案項目小記` empty) → 6.2;
   - **SOR tab missing / row 1 empty, but `專案項目小記` holds a grid** ⇒ a live
     cycle owns that grid — **extend, new-tab branch** → 6.4 (write into the
     empty SOR tab, second pivot grid below the existing one, README counts
     untouched);
   - **row 1 = the layout §1 header, exact strings A–M** ⇒ **extend** → 6.3 /
     6.4; locate every L1 block **by `主類別` text** (the pack §2 label with
     its own number, e.g. `6.0 Tech & Platform`), fallback `#` matching
     `^\d\.0$` (a hand-edited tracker that stores `#` numeric reads back
     `1`–`8` — accept only when `主類別` carries the label); record each
     block's actual `#` prefix (positional — the July tracker's `6.0 Tech &
     Platform` block reads `#` `5.0`) and its max serial; a new L1 continues
     from the last block's prefix + 1;
   - **row 1 present but ≠ layout §1** ⇒ **stop** and say why — never guess a
     column mapping, never clear a tab that holds anything.
2. **Fill** (nothing is overwritten — both tabs are empty): if `<cycle> 專案
   項目` does not exist, `create_sheet(sheet_name="<cycle> 專案項目")` (blank
   tab, no template copy, `insert_sheet_index` per the pack §6 order; same for
   a missing `專案項目小記`). Then exactly Step 5.3 (RAW data write +
   USER_ENTERED G formulas + freeze, all ranges prefixed) → Step 5.4 pivot grid
   written `USER_ENTERED` into the empty `專案項目小記` (nothing below it to
   clear) → Step 5.5 colours (a `create_spreadsheet` tab carries no
   rules, so add the four layout §3 rules + the L1 header fill). README ·
   `<cycle> 回顧總結` · `②/③/④` are session-synth's — read the README and
   refresh only the counts line (Step 5.6, second branch); touch nothing else.
   The old-cycle tracker (sources §A) is not opened.
3. **Append under L1** (extend default): per L1 block with new items,
   `resize_sheet_dimensions(sheet_name="<SOR tab>", insert_rows=<n>,
   insert_rows_at=<row of the next L1 header>)` (or after the last block —
   `sheet_name` is mandatory, Hard rule 6), read the inserted range back and
   confirm it is blank, reset its format (`format_sheet_range(range_name=
   "'<SOR tab>'!A<r1>:M<r2>", bold=False, background_color="#FFFFFF")` — an
   insert inherits the row above, and directly under an L1 header that is
   the bold grey fill), then write the new rows into it — data columns `RAW`
   with serials continued under the block's **existing `#` prefix**, then the
   G formulas `USER_ENTERED` — every `range_name` prefixed `'<SOR tab>'!`. An
   L1 with no block yet: insert `1 + n` rows after the last block, write its
   header row (`#` = last existing prefix + 1, `主類別` = the pack §2 label,
   bold + fill per layout §3) and its items `<prefix>.01` …. Existing rows are
   never rewritten or moved.
4. **New tab** (user says so, the L2 set/columns differ, or the 6.1 「SOR
   empty + live 小記」 case): if the target tab already exists empty, use it;
   else `create_sheet(source_sheet_name=<SOR tab>, sheet_name="<cycle> 專案項目",
   insert_sheet_index=<index of the old SOR tab + 1>)` — if that name is
   taken, `<cycle> 專案項目 — YYYY-MM 現行版` (pack §8) — clear
   `'<new tab>'!A2:M1000`, fill as in Step 5.3; add a second pivot block two
   rows below the first (`USER_ENTERED`, layout §3), pointing at the new tab;
   the old tab and the first grid stay untouched as the archive (pack §8), and
   the README counts of that other cycle are not touched.
5. Pivot (extend): add owner rows for new 負責人; never delete an owner row.

### Step 7 — Verify + report

Read back the SOR tab (`include_formulas=true`): row count = planned, first/
last `#` per L1 read back as the literal strings (`1.0`, `1.10` — not `1`,
`1.1`), G resolves to P0–P3, `專案項目小記` counts evaluate (numbers, not
literal `=COUNTIF…` text, no `#REF!`); in extend mode also
re-read README `A1:B3` and confirm it is unchanged (no stray insert). Then print
the final report (layout §6): **WROTE** (file name + ID + URL, tabs, row
ranges, pivot rows, colour rules) · **OVERRIDES** (T# + reason) · **DID NOT**
(排除 rows, tabs to rename/delete manually, README `（待補）` lines) · **NEXT**
(`/planning-tracker-sync <ID>` for the first Team Weekly block;
`/planning-suite-reconciler` once the tracker is declared final).

---

## Outputs

- **Fresh:** one new Sheet 「<year> <cycle> Planning Main Tracker」 in the hub
  (or named) folder — SOR tab `<cycle> 專案項目` (header + L1 blocks + `N.NN`
  items + Priority formula), `專案項目小記` re-pointed, 狀態 colours, cycle-named
  retro tab, emptied `②/③/④` tabs for session-synth, README skeleton; the
  template's old tabs emptied and listed as leftovers.
- **Fill:** the session-synth file's empty `<cycle> 專案項目` + `專案項目小記`
  filled in place (header + L1 blocks + items + formula, pivot, colours);
  README counts refreshed; every other tab untouched; no second file.
- **Extend:** new rows under the right L1 blocks (serials continued) or a new
  `<cycle> 專案項目` tab; pivot rows added for new owners.
- **Always in the chat:** the lint report (T1–T8 + coverage echo), the C1 cash
  pre-mortem, the row plan, and the wrote / did-not-write summary with IDs.

## Reference files

- `./references/planning-knowledge-pack.md` — the family's shared pack (byte-
  identical across the eight `planning-*` skills; do not edit here).
- `./references/planning-sources.md` — live IDs: template tracker, hub folder,
  tab → gid map, Google account (byte-identical across the family).
- `./references/tracker-layout.md` — this skill's contract: item schema +
  column map, Priority formula, pivot grid + formulas, colour defaults, lint
  rules T1–T8, C1 spend keywords, and the exact lint / row-plan / final-report
  shapes.

## Limitations

- Tab rename/delete is not exposed by the `google-workspace` MCP: cycle-named
  tabs are duplicates; the emptied old tabs are a manual leftover.
- Lint is heuristic: T6 matches L2 by numeric prefix, T7 catches exact/near-
  exact 項目 text, C1 is keyword-based — a P0 that spends money without saying
  so is not listed; the founder's confirmation is the gate.
- Pivot owner rows are whatever the list carries; the roster is read from the
  負責人 column, never assumed (pack §9).
- Extend mode trusts the existing header; a hand-modified one stops the run.
  Fill mode only fires on a genuinely empty SOR tab — a tab with a partial
  header or stray cells is treated as hand-modified and stops too.
- Column `#` is written as text (RAW). A tracker whose `#` was later
  re-typed numeric by hand still extends (the L1 label in `主類別` is the
  fallback key), but its `1.10`-style serials may already read as `1.1` — the
  read-back names them; the skill does not repair them.
- One tracker per run; retro table, whiteboard tabs and recap mail belong to
  planning-session-synth; Doc addenda to planning-suite-reconciler.
- **Fill mode extends the spec row** (SKB-007 row 0.07 names fresh + extend
  only): it exists so the file planning-session-synth creates is completed
  in place instead of duplicated; it is guarded — SOR tab AND `專案項目小記`
  both empty — and otherwise falls back to extend.

## 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.