Stats
Actions
Tags
From you-read-i-read
Use when adding a paper, blog, or report by URL, tracking a research group or topic, or sweeping tracked sources for new candidates. Triggers include "/add <url>", "/track-group", "/track-topic", "/update", "add this paper", "follow this group", "watch this topic", "any new papers since last week", "summarize and save this".
How this skill is triggered — by the user, by Claude, or both
Slash command
/you-read-i-read:paper-curateThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Ingestion skill. Flows: `/add`, `/track-group`, `/track-topic`, `/update` (sweep). State: `~/.you-read-i-read/data/`; never write to the plugin tree.
Ingestion skill. Flows: /add, /track-group, /track-topic, /update (sweep). State: ~/.you-read-i-read/data/; never write to the plugin tree.
/add and [r]ead-now. /update uses abstracts — never fetch PDFs in a sweep.last_checked stays put; retried next sweep.manage_data.py signal-log --event update_accepted|update_rejected.Pick the highest tier available; tell the user which ran.
| Source | Tier 1 | Tier 2 | Tier 3 |
|---|---|---|---|
| arXiv | WebFetch Atom XML | — | — |
| Semantic Scholar | semanticscholar lib | WebFetch S2 API | — |
| Hacker News | WebFetch Algolia API | — | — |
Apify apidojo/tweet-scraper* | superpowers-chrome:browsing | Nitter WebFetch | |
Apify curious_coder/linkedin-post-search* | superpowers-chrome:browsing | skip + message |
*Requires APIFY_TOKEN + budget.
/add <url>manage_data.py paper-set-raw <id> --ext ...
(txt for grounding, plus pdf when arXiv). Skipped if
summary.store_raw_content is false.manage_data.py paper-add --url ... --title ... --authors ...echo "$summary" | manage_data.py paper-set-summary <id>signal-log --event add_via_url./track-group <name>Resolve to S2 author IDs. Confirm. manage_data.py group-add. state-set-last-checked --kind groups --id <id>.
/track-topic <topic>Distill to per-source queries. Confirm. manage_data.py topic-add. Init watermark.
/updatePer source: (1) get watermark; (2) discover via highest tier; (3) dedupe against papers/index.json; (4) rank from signals.jsonl tail (~100) + preference-prompt.md — score accept_likelihood ∈ [0,1], blend recency + trending weights; (5) triage:
[k/N] <id> rel=0.82 like=0.74 rec=0.93 trend=0.85
<Title> · <authors> · via: <topic>
One-liner: … Why: <rationale>
[a]dd [s]kip [r]ead-now [?]info [q]uit →
Log on a/s. ? fetches context, no persist. r → /add then /read. Advance watermark after full triage; q advances finished sources only.
The paper-summarizer subagent always produces a deep 7-section summary AND persists raw content. Dispatch only when that's the desired output:
/add: dispatch one summarizer per URL./update [r]ead-now: dispatch one summarizer for the chosen candidate, then drop into /read./update sweeps themselves: do NOT dispatch the summarizer for one-liners. Source adapters return abstracts; derive the per-card one-liner in-context from the abstract. The /update sweep is two-tier-quick by definition.When dispatching for parallel /add (rare in v1), cap at 8 concurrent.
/update (token-bombs sweep).preference-state.yaml not signals.jsonl (digest lacks tag/exemplar data).npx claudepluginhub zhangzhuosjtu/youreadiread --plugin you-read-i-readCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.