AGENT SKILL

packing-inventory

Build or update the inventory of what is being carried — items, quantities, weights, carriage restrictions, and how replaceable each thing is at the destination. Use when the user is listing what they have packed or intend to pack, dictating the contents of a bag, or asking what is in their luggage. Writes inventory.yaml, which the solver and the leave-behind ranking both read.

License 许可
MIT
最近更新
2026-11-10
Stars
★ 0
Forks
0

01 前言 Front Matter

name
packing-inventory
description
Build or update the inventory of what is being carried — items, quantities, weights, carriage restrictions, and how replaceable each thing is at the destination. Use when the user is listing what they have packed or intend to pack, dictating the contents of a bag, or asking what is in their luggage. Writes inventory.yaml, which the solver and the leave-behind ranking both read.
allowed-tools
Read, Write, Edit, Bash(uv run *), Bash(ls *), Grep, Glob

03 SKILL.md 正文

Packing Inventory

$TPA below is this plugin's new-workspace skill directory — the folder holding its SKILL.md, scripts/, schema/, typst/ and template/. Find it by name in your tool's skills folder (some tools prefix it, e.g. travel-packing-assistant__new-workspace) and set TPA once per session.

The inventory is the input everything else depends on. A plan built on a half-inventory is a plan for a bag that does not exist.

It does not have to be complete to be useful — /travel-packing-assistant:weigh reconciles the itemised total against the actual scale reading and reports the gap, so a partial inventory is honest rather than wrong. What it must not be is silently partial.

Capture modes

Offer whichever fits what the user is doing right now. Do not force a mode.

Dictation. The user talks through a bag; you write items as they go. Fastest for a bag already packed. Batch back a confirmation every ten items or so rather than after each one.

Photograph. The user lays the contents out and photographs them. Open the image, list what you can identify, and ask about anything ambiguous. Good for clothing, poor for weights — everything from a photo is weight_source: estimated.

Category sweep. For packing not yet done: walk the categories in $TPA/data/packing-categories.md and ask what is going in each. Slower, but it is the only mode that catches the thing they have not thought of yet.

Import. A previous trip's inventory.yaml is the best starting point a repeat traveller has. If the user names an earlier trip workspace, read its inventory, carry over the items and their measured weights, and ask only what changed.

Weights

Ask how much of the inventory can be weighed. In descending order of usefulness:

  1. Measured — a kitchen scale, in grams. Ten minutes of weighing the heavy end of the list (shoes, chargers, books, toiletries, outerwear) removes most of the uncertainty, because weight distribution in luggage is strongly top-heavy: a small number of items usually account for the majority of the mass.
  2. Vendor / catalogue — a published spec weight for electronics and hardware. Usually accurate, usually excludes the cable and the case, so weigh those separately.
  3. Estimated — from $TPA/data/typical-weights.md. Fine for socks, not for a laptop.

Record which, per item, in weight_source. That field is what lets reconcile.py tell the difference between "the inventory is missing items" and "the estimates are drifting".

Weigh containers separately: a bag's tare_kg and each packing cube's tare_kg belong in bags.yaml, not as inventory items. Double-counting an empty suitcase is the classic way to produce a plan that is 4 kg pessimistic.

The fields people skip and later need

  • carriage — anything that cannot travel in an arbitrary bag. Power banks and spare lithium batteries are cabin-only in essentially every jurisdiction; liquids over 100 ml cannot go through the cabin security check; sharps, tools and some aerosols are checked-only or barred. See $TPA/data/carriage-rules.md. Set carriage_reason every time; the solver treats these as hard constraints and the reason is what the traveller reads when they wonder why the plan put a heavy thing in the small bag.
  • tier — essential / useful / nice-to-have / expendable. Assign it as you capture, not later. Ranking forty items under time pressure at a check-in desk does not work; ranking them at home does.
  • replaceable_at_destination — can it be bought at the other end, and roughly for how much. This is the field that makes the leave-behind list sensible: a 900 g bottle of shampoo that costs $4 to replace is the correct thing to abandon, and no weight-only ranking can know that.
  • value — replacement cost. Used with weight to rank; also the honest answer to whether something belongs in a checked bag at all.

Procedure

  1. Read inventory.yaml and bags.yaml. If bags are not defined yet, define them first — an item cannot be assigned to a bag that does not exist.
  2. Capture in the mode the user chose. Assign stable ids (I001…) and never renumber; the plan, the PDFs and the conversation all reference them.
  3. Fill carriage, tier and replaceable_at_destination as you go.
  4. Validate: uv run $TPA/scripts/validate.py . --json.
  5. Report: item count, itemised total in kg, the split by weight_source, and the ten heaviest items. The heavy-ten list is the single most useful thing to show a traveller — it is where every kilogram of headroom is going to come from.

Pinning

If the user says something must stay in a particular bag — documents in the cabin bag, a gift that must not be crushed — set assigned_bag and pinned: true. The solver will place it first and refuse to move it, and will report an error rather than quietly relocating it if the pin makes the pack infeasible.

04 同仓库更多 Skills