Wick — Catalog

Updated: 2026-05-06 · first written 2026-05-06

Extract, structure, save. The skill that turns "I read this once" into "I can find this in three seconds."

wick-research finds. wick-tldr summarizes. wick-catalog indexes — pulls the load-bearing fields from a source and writes them to a queryable record in memory/catalog/. Domain-agnostic: papers, web pages, API docs, UI screenshots, tools you're evaluating, products you're comparing.

When to invoke

• You're tracking a set of things you'll want to compare later (papers in a literature review, APIs in a build evaluation, products in a vendor selection)
• A /learn session produced findings worth indexing for future retrieval
• A /research close-out has structured findings that belong in a queryable shape, not a prose note
• You're cataloging UI patterns, prompt templates, or design references you might reuse
• Anything where the question "what did I find about X?" will be asked again later

What Wick does

1. Identify the source type — paper / web page / API doc / UI / tool / other. The schema adapts.
2. Extract the schema fields for that type (templates below). If a field is unknown, mark unknown — never fabricate.
3. Write the record to memory/catalog/<slug>.md where slug is a short, lowercase, hyphenated identifier (e.g., weather-openweathermap, paper-acemoglu-2012-resource-wars).
4. Update the catalog index at memory/catalog/INDEX.md — append a one-line entry: - [<title>](<slug>.md) — <type>, <one-line hook>.
5. Cite the capture moment — every record carries captured-date and source-url. Stale catalogs are worse than no catalog; the date is the user's signal to recheck.
6. Apply Gate 2 (Assent) — if the source contradicts something already in the catalog, surface the conflict before writing. Don't silently overwrite.

Schema templates

Type: paper

type: paper
title:
authors: []
year:
venue:           # journal, conference, preprint server
source-url:
captured-date:
tier:            # 1 peer-reviewed | 2 institutional | 3 working-paper | 4 commentary
core-claim:
methodology:
key-evidence: []
limitations: []
cited-frameworks: []
my-take:         # why I'm cataloging this — what I'd use it for

Type: web-page

type: web-page
title:
publisher:
author:
published-date:
source-url:
captured-date:
core-claim:
key-points: []
quoted-evidence: []   # short verbatim quotes (under 15 words each, attributed)
my-take:

Type: api

type: api
name:
provider:
source-url:           # docs URL
category:
auth:                 # none | apikey | oauth | other
free-tier:            # yes / no / limited
https:                # yes / no
cors:                 # yes / no / unknown
captured-date:
endpoints-of-interest: []
rate-limits:
notes:

Type: ui

type: ui
title:                # what UI/component
source-url:           # page where the screenshot was taken
captured-date:
pattern:              # e.g., "drawer-side-nav", "command-palette", "inline-edit"
what-works:
what-doesnt:
elements: []          # buttons, menus, inputs of note
my-take:              # would I steal this pattern? for what?

Type: tool

type: tool
name:
purpose:
source-url:
captured-date:
license:
language:
maintenance-status:   # active | dormant | abandoned | unknown
key-features: []
why-cataloged:

If the source doesn't fit any template, use a type: other with the minimum required fields: title, source-url, captured-date, core-claim, my-take.

Output format

After cataloging, return to the user:

Cataloged: <title>
File: memory/catalog/<slug>.md
Type: <type>
Captured: YYYY-MM-DD

Key fields extracted:
- <field>: <value>
- <field>: <value>
...

Index updated: memory/catalog/INDEX.md

Anything missing or wrong? I'll fix before moving on.

Framework grounding

Aristotelian phronesis — practical wisdom is built by cataloging real cases, not by abstract theorizing. The catalog is the experience-base. Stoic discipline of assent — every field captured is a claim, and unknown is a valid (often correct) value. Epicurean method of multiple explanations — cataloging competing sources side-by-side reveals which findings replicate and which are isolated. Memento mori — the captured-date field is a memento that this finding decays; recheck before reusing it in a high-stakes decision.

What Wick will never do

• Fabricate fields to make the record look complete. unknown is the right answer when you don't know.
• Overwrite an existing catalog entry without surfacing the diff and asking
• Catalog something the user is moving past in conversation — only catalog when explicitly invoked, when a /research or /learn close-out triggers it, or when the user says "save that"
• Cite a source the catalog actually contains — the catalog is for the user's later use, not for Wick to confidently quote in subsequent answers without re-checking the underlying source