reshelf-collector

name

reshelf-collector

description

Resurface the REST of the user's reshelf collection — everything catalogued in their reshelf macOS app EXCEPT the Yard Sale — led by The Collector: the shelved-and-forgotten middle that was never promoted to Top Shelf and often never cloned. Third of the reshelf trio: `reshelf` reads the source of cloned repos, `reshelf-catalog` indexes the whole shelf with a top-shelf bias, and THIS one digs through the rest so kept-around saves don't rot unseen. Trigger it whenever the user asks about the middle of their collection: "what's in my collector?", "the rest of my collection", "what did I shelve and forget?", "resurface my shelf", "oldest stuff I saved", "what's been sitting uncloned the longest?", "anything I collected for X besides my top picks?", "spring-clean my collector", "what should I promote or clone from the rest?". It maps the keeper collection (Yard Sale always excluded) with clone status and shelf age, leads with the forgotten middle, recommends what deserves promoting/cloning, and can pull a not-cloned repo's README from GitHub on demand. Hand off to `reshelf` to read cloned source, or to `reshelf-catalog` for the whole-shelf index / top-shelf gap. Do NOT use it for: developing the reshelf app itself (its Swift/UI code), Yard Sale contents (on the way out, by design), or discovering brand-new projects on the open web.

reshelf-collector — the rest of the shelf, before it's forgotten

The user runs reshelf, a macOS app that catalogs open-source repos and clones the keepers to disk. Each catalogued project carries a status: topShelf (best picks), collector (kept around — the default), yardSale (on the way out). Two sibling skills already cover the edges: reshelf reasons over cloned source in ~/reshelf/repos/<Category>/<repo>, and reshelf-catalog indexes the whole shelf with a top-shelf bias.

This skill owns the rest: the keeper collection minus the Yard Sale, led by The Collector — things the user deliberately kept but never promoted and often never cloned. Its job is resurfacing: nothing kept should rot unseen.

Where the data lives (no app, no MCP needed)

~/reshelf/catalog.store              live SwiftData store (SQLite)  ← primary
~/reshelf/backups/catalog-*.json     timestamped JSON exports        ← fallback
~/reshelf/repos/<Category>/<repo>    the actual clones (for clone-status join)

Each record carries: name, category, GitHub URL, descriptions, stars, license, tags, fit score, status, and added date — enough to survey, age, and recommend without cloning or hitting the network.

How to work through a request

1. Map the collection (always start here)

bash ~/.claude/skills/reshelf-collector/scripts/collection-map.sh                     # summary + the forgotten middle
bash ~/.claude/skills/reshelf-collector/scripts/collection-map.sh --status collector  # just The Collector
bash ~/.claude/skills/reshelf-collector/scripts/collection-map.sh --category "AI / Agent"
bash ~/.claude/skills/reshelf-collector/scripts/collection-map.sh --uncloned          # kept, but not on disk
bash ~/.claude/skills/reshelf-collector/scripts/collection-map.sh --oldest            # longest-shelved first
bash ~/.claude/skills/reshelf-collector/scripts/collection-map.sh --search <term>     # name/desc/tags/category/url
bash ~/.claude/skills/reshelf-collector/scripts/collection-map.sh --tsv               # raw rows to filter yourself

Yard Sale rows never appear — that shelf is on its way out, by design. Top-shelf rows do appear (so a topic survey misses nothing) but Collector rows are listed first; when a top-shelf item comes up, note it's already among the user's best picks.

The summary shows totals, cloned/total by status and category, and the headline list: Collector picks not cloned, longest-shelved first — the forgotten middle.

Notes that keep results correct:

• The script prefers the live store and silently falls back to the newest JSON backup; it prints which source it used.
• Clone status is joined on the GitHub repo slug (URL basename) — rely on the script's [cloned]/[ shelf] flag, don't infer from folder names yourself.
• The added column is the month the repo was shelved — the resurfacing signal.

2. Pin down intent: resurface, survey, or triage

• Resurface ("what did I forget?", "oldest stuff I saved") — lead with the forgotten middle: uncloned Collector rows, longest-shelved first. Group by category and remind the user why each was probably saved (description + tags).
• Survey ("what's in the rest of my collection for X?") — list the relevant slice, Collector first, clone flags and ages visible.
• Triage ("spring-clean my collector", "what should I promote or drop?") — for each candidate suggest one of: promote (it earned Top Shelf), clone (worth studying — do it in the app), leave (fine where it is), or yard-sale it (say so honestly). The user acts in the app; you only recommend.

3. Go deeper only when asked

• Read a not-cloned repo's README from GitHub without cloning (gh repo view <owner>/<repo> if available, else the raw README URL). Summarize what it is and whether it still deserves its shelf space.
• Suggest cloning when the user wants to actually study or borrow code — that's the reshelf app's job (right-click → Clone Repository). Once cloned, switch to the reshelf skill to read the real source.

4. Recommend, with age and clone state explicit

From the rest of your shelf, for <goal>: <repo>   [cloned ✓ | not cloned] · shelved <YYYY-MM>
  Why: <relevance, stars, license, fit — from the catalog>
  GitHub: <url>
  Next: <promote / clone in the app / open with the reshelf skill>

Runner-up: <repo> — <when you'd pick it instead>   [clone state · age]

Be honest when the middle shelf is thin for a topic. "Your collector has nothing for this; closest is X on your top shelf" beats a forced pick.

Relationship to the sibling skills (don't overlap)

• reshelf = the source: reading and borrowing code from cloned repos.
• reshelf-catalog = the index: the whole shelf, the top-shelf not-cloned gap.
• This skill = the rest: The Collector middle, aged and resurfaced, Yard Sale excluded. When a survey lands on a cloned repo the user wants to study, hand off to reshelf; for whole-shelf questions or top-shelf gaps, hand off to reshelf-catalog.

Guardrails

• Read-only. Don't write to catalog.store, the backups, or the clones. Shelf changes (promote, clone, yard-sale) happen in the app — you recommend, the user acts.
• Don't run or install anything from the catalog just to inspect it.
• Respect licenses before suggesting code be borrowed (NOASSERTION means verify on the repo).
• Don't touch the reshelf app's own code — that's a development task, not this skill.