/collect

description: "Find, catalog, deduplicate, and optionally inventory bounded sets of artifacts, examples, resources, memes, tools, entities, media, or other collectible things."
argument-hint: "\"<things to collect>\" [--type <kind>] [--scale tiny|small|medium|large|huge] [--limit <N>] [--media archive|thumbnail|reference] [--inventory preview|none|corpus|records] [--ingest-sources] [--dry-run] [--wiki <name>] [--local] [--include-archived]"
allowed-tools: Read, Write, Edit, Glob, Grep, Bash(ls:*), Bash(wc:*), Bash(date:*), Bash(mkdir:*), Bash(curl:*), Bash(shasum:*), Bash(file:*), Bash(stat:*), WebSearch, WebFetch, Agent

Your task

Find and catalog a bounded set of "things" for the target wiki. This command is for collectible objects and examples where the deliverable is a curated catalog: memes, images, videos, quotes, tools, projects, products, examples, people, venues, source candidates, and similar discoverable artifacts.

Treat collect outputs as a provenance-rich discovery layer for the LLM. They are where the agent records "I found these things, with this context and confidence, but I have not yet decided what they mean." Later workflows may promote selected context pages into `raw/`, synthesize stable knowledge into `wiki/`, create durable tracking records in `inventory/`, or index large media/data collections in `datasets/`.

Collect is not the same as `/wiki:research` or `/wiki:ingest-collection`:

• Use `/wiki:research` when the user wants an answer, explanation, thesis, or

synthesized wiki articles.

• Use `/wiki:ingest-collection` when the input is a structured upstream corpus

such as a Git repo, MediaWiki dump/API, message archive, or Wayback CDX set.

• Use `/wiki:inventory` directly when the user already knows the records to

track and no discovery/catalog pass is needed.

• Use `/wiki:collect` when the user says "find", "collect", "catalog",

"curate", "gather", "make a list of", or "inventory all the examples of X."

Always be clear about boundedness. "All" means "all discovered within the stated search strategy, quality threshold, and limit", not the entire internet.

Resolve the wiki

Follow the standard resolution flow:

1. Read `$HOME/.config/llm-wiki/config.json`. If it has `hub_path`, expand leading `~` only and prefer that path; use `resolved_path` only as a fallback cache when the expanded `hub_path` is unavailable and `resolved_path` is initialized. If config has only `resolved_path`, use it. If the configured path can be statted but reading `wikis.json` or listing `topics/` fails with `Operation not permitted`, stop and ask the user to grant Full Disk Access/iCloud Drive access to the launcher; do not fall back to `~/wiki` or `resolved_path`. Do not write machine-specific `resolved_path` into shared configs. 2. If no config exists, read `$HOME/wiki/_index.md`. If it exists, use `$HOME/wiki` as HUB. If nothing is found, ask where to create the wiki. 3. Wiki location, first match: `--local` -> `.wiki/`; `--wiki <name>` -> `HUB/wikis.json` lookup with portable path resolution (`<HUB>`, `~`, absolute, or HUB-relative), falling back to `HUB/topics/<name>` if stale; current directory has `.wiki/` -> use it; otherwise use HUB. 4. Read `<wiki>/_index.md` to verify. If missing, stop with "No wiki found. Run `/wiki init` first."

Archive rule: collect operates on active topic wikis by default. If a named target is archived, stop and ask the user to restore it or rerun with `--include-archived`. When explicitly included, write only inside that archived topic and label the result as archived.

Read `skills/wiki-manager/references/inventory.md` and `skills/wiki-manager/references/research-infrastructure.md` before writing. Use inventory as tracking state, not evidence.

Parse arguments

• **Query**: the thing to collect, e.g. `"bitcoin memes"`, `"open-source LLM

eval tools"`, `"companies using Nostr"`, or `"examples of SeedQR backup UX"`.

• **--type <kind>**: optional collection kind. Examples: `meme`, `image`,

`video`, `tool`, `project`, `person`, `company`, `quote`, `paper`, `source`, `example`. If omitted, infer from the query.

• **--scale <scale>**: optional scale override. See Scale below.
• **--limit <N>**: maximum catalog rows. Defaults by scale. If the user asks for

"all", still apply the scale limit unless they explicitly set a higher one.

• **--media archive|thumbnail|reference**:
• `archive` default for media-bearing collections: download originals or

rich media when a direct public media URL is available and the collection remains bounded.

• `thumbnail`: cache small previews instead of originals when full originals

are too heavy for the requested collection.

• `reference`: opt out of binary downloads and store URLs plus metadata only.
• **--inventory preview|none|corpus|records**:
• `preview` default: suggest inventory shape but do not create records unless

the user's wording explicitly asked to inventory the findings.

• `none`: only write the collection output.
• `corpus`: create one `inventory/corpora/` record pointing at the collection

output.

• `records`: create one inventory record per collected item when the count is

small enough to remain useful.

• **--ingest-sources**: ingest strong supporting pages as raw sources after the

catalog is built. Do not copy binary media into raw; raw is source text.

• **--dry-run**: show the search strategy, candidate schema, scale, media

policy, and inventory fit without writing files.

• **--include-archived**: explicitly allow writes inside an archived target.

Scale

Define scale by operational cost, not only row count. Use the maximum pressure from candidate count, boundedness, payload weight, provenance burden, volatility, and inventory usefulness.

| Scale | Rows | Source Shape | Default Behavior | |-------|------|--------------|------------------| | `tiny` | 1-3 | Known items | Skip collect; use ingest, que