readaware-read

name	ReadAware Read
description	共读核心。用户扔来一段书里的原文（可能附问题，也可能只是一段话），就定位它在书中的卷/章并按读者的偏好精读分析。当用户贴出原文片段、问"这段怎么理解"、想精读/细读某段时触发。The close-reading loop that locates a thrown passage and analyzes it per the reader's profile.
allowed-tools	Bash, Read

readaware:read — co-reading, passage by passage

The user will throw you a passage — that's usually a fragment of the book they're reading. Sometimes with a question attached, sometimes just a passage. Your job: first locate exactly where it sits in the book, then read it through per this reader's profile.

Load two things up front

1. The reader profile ~/.claude/readaware/profile.md — decides the tone, depth, format, and lenses you use with them. This profile overrides any default style in this file. If it's missing, prompt the user to run readaware:onboarding first (or read in a neutral, restrained, non-verbose way for now — don't improvise a style).
2. The active book: read active_book from ~/.claude/readaware/state.json. If that file is missing, or the library holds only one book, use the one in the library (the sole directory under ~/.claude/readaware/books/). Multiple books with none specified — ask the user which one.

Remember the book dir BOOK=~/.claude/readaware/books/<active_book>, holding text.txt, manifest.json, and possibly card.md (read it if present — it tells you the book's motifs/characters/structure).

Step 1: locate the passage (don't hand-grep and count line numbers)

Take a fairly distinctive short phrase from the user's fragment (a dozen-odd characters is enough) as the argument:

python3 "${CLAUDE_PLUGIN_ROOT}/scripts/locate.py" "$BOOK" "a distinctive short phrase"

• Insensitive to whitespace/punctuation (matches on whitespace-stripped text first, then falls back to a punctuation-loose match, then to a cross-paragraph match if the pasted text straddles a paragraph break). A distinctive short phrase still works best.
• Defaults to 2 paragraphs of context on each side; use -B N -A N for more; -n N when there are multiple hits. Body hits are ranked ahead of front/back-matter hits.
• The script distinguishes body parts/chapters / front matter (author's note, TOC, etc.) / back matter, so it won't mistake an afterword for the epilogue. It reports the part/chapter using the book's own headings, whatever the layout/language.
• Reports "not found": retry with a more distinctive short phrase, a name, or an image; if still not found, say so honestly — it may not be in this book, may be a different translation, or the user may be paraphrasing. Don't fabricate a source.
• When you need a larger span of context, Read $BOOK/text.txt at the file line numbers the script reports.

Step 2: read it through

• Has a question → answer it head-on first. Textual analysis serves the question; don't dodge it.
• No question → give your own reading: what this passage is doing, why it sits here, its relation to the book's core motifs/structure (reference card.md), and what's worth stopping for at the character and rhetorical level.
• You may quote sentences from the original as evidence, but don't re-paraphrase whole chunks of text the user already gave you.
• If unsure, say so — don't fabricate textual details, sources, or scholarly consensus.

Style: entirely up to the profile

How to speak — tone, depth, length, layout, whether to lead with the conclusion, which lenses to enter from, what the landmines are — follow profile.md throughout. This skill only carries the method of "how to read a book accurately and thoroughly"; "how to speak to this person" is set by the profile. Don't conflate the two.