From 0dd8808993f8d6fd2b33c4043510528ed17bec13 Mon Sep 17 00:00:00 2001
From: Zhongwei Li <lizhongwei.nkcs@gmail.com>
Date: Sun, 30 Nov 2025 08:53:34 +0800
Subject: [PATCH] Initial commit

---
 .claude-plugin/plugin.json |  12 ++
 README.md                  |   3 +
 plugin.lock.json           |  49 +++++
 skills/candlekeep.skill    | Bin 0 -> 4483 bytes
 skills/candlekeep/SKILL.md | 412 +++++++++++++++++++++++++++++++++++++
 5 files changed, 476 insertions(+)
 create mode 100644 .claude-plugin/plugin.json
 create mode 100644 README.md
 create mode 100644 plugin.lock.json
 create mode 100644 skills/candlekeep.skill
 create mode 100644 skills/candlekeep/SKILL.md

diff --git a/.claude-plugin/plugin.json b/.claude-plugin/plugin.json
new file mode 100644
index 0000000..e26c55e
--- /dev/null
+++ b/.claude-plugin/plugin.json
@@ -0,0 +1,12 @@
+{
+  "name": "candlekeep-cli",
+  "description": "Personal knowledge base CLI that gives AI agents direct access to your books and annotations",
+  "version": "0.1.0",
+  "author": {
+    "name": "Sahar Carmel",
+    "email": "sahar@example.com"
+  },
+  "skills": [
+    "./skills"
+  ]
+}
\ No newline at end of file
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..5e2ec78
--- /dev/null
+++ b/README.md
@@ -0,0 +1,3 @@
+# candlekeep-cli
+
+Personal knowledge base CLI that gives AI agents direct access to your books and annotations
diff --git a/plugin.lock.json b/plugin.lock.json
new file mode 100644
index 0000000..054c392
--- /dev/null
+++ b/plugin.lock.json
@@ -0,0 +1,49 @@
+{
+  "$schema": "internal://schemas/plugin.lock.v1.json",
+  "pluginId": "gh:SaharCarmel/CandleKeep:plugins/candlekeep-cli",
+  "normalized": {
+    "repo": null,
+    "ref": "refs/tags/v20251128.0",
+    "commit": "c710acc3dd5361811aface1a59166f25b3e5a50d",
+    "treeHash": "4f074f51c1f304eb0e02cbc79398f810f8d753141e3bffcf1210abad91f38436",
+    "generatedAt": "2025-11-28T10:12:43.342588Z",
+    "toolVersion": "publish_plugins.py@0.2.0"
+  },
+  "origin": {
+    "remote": "git@github.com:zhongweili/42plugin-data.git",
+    "branch": "master",
+    "commit": "aa1497ed0949fd50e99e70d6324a29c5b34f9390",
+    "repoRoot": "/Users/zhongweili/projects/openmind/42plugin-data"
+  },
+  "manifest": {
+    "name": "candlekeep-cli",
+    "description": "Personal knowledge base CLI that gives AI agents direct access to your books and annotations",
+    "version": "0.1.0"
+  },
+  "content": {
+    "files": [
+      {
+        "path": "README.md",
+        "sha256": "c333a756ac195d2566c7e7f739c1c8d7eaeb74599eebf07af6a9c6dc8cc1131c"
+      },
+      {
+        "path": ".claude-plugin/plugin.json",
+        "sha256": "bbdf45b2d2650e4cbb12e0e8185e612bca9ae1ebab11fa50c82ce09de2ae81de"
+      },
+      {
+        "path": "skills/candlekeep.skill",
+        "sha256": "11f51f93770d3fb204a395d882cadc429ad1d7fabdad8f8b091908ee52da04d5"
+      },
+      {
+        "path": "skills/candlekeep/SKILL.md",
+        "sha256": "661f1c990ad3688ad679f0970dfdbdb7e1fd2569d83a17feda47fe7e10ef16a6"
+      }
+    ],
+    "dirSha256": "4f074f51c1f304eb0e02cbc79398f810f8d753141e3bffcf1210abad91f38436"
+  },
+  "security": {
+    "scannedAt": null,
+    "scannerVersion": null,
+    "flags": []
+  }
+}
\ No newline at end of file
diff --git a/skills/candlekeep.skill b/skills/candlekeep.skill
new file mode 100644
index 0000000000000000000000000000000000000000..1a498fbaf9f6a5171e3f919504d56406096dace2
GIT binary patch
literal 4483
zcmV-}5q$1YO9KQH0000805-OCTZ@3D^7{|~03R*@022TJ0ApcpWNc+?Wo2+LQ%gxq
zOfGF?&05WJ8#$7`*Ha+qh%u>QQKWusH^bW#YRR5fx7{sE>hZzr2trjLS*>EBrcgzR
zbC~Fu!#==<V@~rV`#64*{W1Ym6@TnlspqhLXxl{~0c3ue`DFrO7zUYKsIiD-7N_b~
zsXU04iHamI6P=C4g({5BWGZemeV3|uro@djN|bY1ioDRv1jEJ2xsWrJl}5x#p`udA
zC{o6VQi~g{Z_Pmbf;o3{l?j=dJ5?mvO#HP{h88haSfr~o7B?zW)1(yBLN9o3vvA)E
zs0_9wq`4LHMpq@)$dkxm`cf4t7D)ziDk(&o+!V4{xmk9rv}~m5kNr|b^DK#QvRReH
zGK<`qc)wzi%;qIzO?9#83thCAR`+?DL`j*hM4|W)cB6%wVh0d}<6ab6MrE>87!=J_
zbW5A^Te8B|59e20&_)$xnH-u?P?AiOay1A-m@0T8PT5XB!cHP=jPu3`#&*dTCv1x#
zXbqlWFGj3%wQV8E=Xz>H?#1!e+-3(5eyhY>E=hTkcBW(z&0ztPm{MjXj-v@bP#6E3
z_2N_kbFWxjA8L?!sS4I+4z#dFXl0@AGBd#8a5`NmaWPL)ZFD|g1;O6lDQnMKWttSG
zwA$NL++3ykU~f;HPaC<wh;bm17A9>IM|!c)8QF)B^ZB<=78uxDiEU=4kMlrh<pR1C
zrYvNsW{JXQvofVx2;e@MWqlFrg#_q;3RRSeQ6$M*yn!KrJ*C4HKm@H~bV}IJKx={N
z^r=1)bnzW;(PG(~$4e?yPDa44><ZCbn>CFXT@?<6WCv@`NM|L@p~E(7F4RJMK+2^|
zQhAdqYm4`~xSgi@PMoPIG4NAyT__a<lgT9bh1Bc79|-*S`O`n#pa1&&>9^1S{2$Rd
ziDPUqG2-Iv^*(K(<45xUKe=%f7-SrWc|0X#&|eoZ8wcS1g6wxLa#(X(0Nf}SS`Y(n
zEauGGf`0O7AS|YmY1^9;Qg>1&aJRpyc<0Om@&y2}R+K|hw+Wswk|}#TXCDp-oUp%I
zlGCtJ)IElq@!I(HS$|eAGq0tm-iGTdmOAR~5a9Lu>K?|1zndIpoWcdj_tquwm)=h6
z91X+?YZV|QysSa#j{;H9!)?P<pX&W*sz69-gJpdLLS>|2v<#tErYQn8c^p~SYBy<=
z#@Hilo_kA49?15D(^wa?J8ewc7l3>+U5TW8x@P1tgBj^;PJz{9qBTHtQ#vn2)xYl#
zS{k!46rR%yQX|4tuK!Mf8n(`K*_PH9MU{<%C>AgCw3;PZSc~~q{NWFxTGD`4Cg#cZ
z%zJwqM~(`GfMmvddqF5BYpd^1kQ&}nOseH?k7@ooh;L}Z)w?%j%vb_#l;FDyRZ9Ao
zpdi`Xmf*sXTdlZ_5ylarQvz2TCOf9;BEj)xYIyR`?q3`z7Hg7RQs7u1E`3r7f|E4$
zA;K`UaL~3hpT0RCh--vSt7u{7OfEP<VJe3uS(S*~B~gGBS?<(}lN6<4hr6Vli^-Nz
zCYG9%jqqB_95<g(fZ;G<V|aH$Az{)Bf-6N7sLnsMP7_I7sP{>?jm+<)f!mT4(M8H8
z=YX=jd3p~46bT`4zOK?n!b1g!Fq~pO$+_}DWtyloHjk0S6Y&GGD+c#VIWr#eKUZar
z+)W%zmdNd04N|&b$0V10^X4sK%v1~Js=YDI^Z3E$JSlQ^xYn@!SRD2kRP<4tpN+*R
zxCWI213pa^ILxVzRd7<3b6t$ZrM^K<7pDUPc$s9u^(x0mq^jV8V&+&JJv$1J=nzZ9
z!SLWH9F4+*p%@Lvhr{u3xW3YXU+IjQ0|luws6RX>H!<L6;v^?#$qsJnP1vQykEzOY
zAautjo{gSuZesL}7#)v~j%gE4o1TH&knu}RkUA|)##)>c)l+}!U<`*u+9sl+Cs0Yl
zv>l_V2|Z?9mAqNvi3U)mK*erE(nCaXSUDDVAQskBkWeYVUHklYTIaV3vKsl!UlMg^
zlMWbI6_phh1ZABjhzE5lH2@l^tmQQi03ucrFL{GW{H<5kt9!Y~VJqnBks?PsQS1aZ
zD1EZy^x@ah88m+vU4noZQSv^u{926KgWZ7V?EweC6n=-9;Kt%7lz+y{d2U(2Wkpe(
zd<?&mIo;1M?9YTG1}$klyi`ktyo6C}qF%I@JRBqBWY$V<z#%rh6PZ5|8Spm7XM)%k
z_;w&T)>!-l8S<kra^;Bs2cG$n@qCfOeQ2Y1^8^KqB|=z9S&C#ad1x_tVOi3I2N4p#
zz@wm}LT2<gP79L3`Zyq3fi)Bn9PO9ROJ3pIj@K9KkH0xoIT~<<vjIxEjS7>_sDIoa
z4LcM3bjf?<kKu7=BDz-e2uY{V)od<~DNVmk?kR}yAk6HKjuFjTqwy87tuH7W^KZSD
zub8&8JYgleY0tyMk?7X&W31N*eNg`LyRKEkvO9rWZ{0Qz7ktnXOy@ryhR4U9|EnW6
zP74oi`Kyc4yj6=a=WNV6zvz>Jbtj?;@UUWKU4h7qo|bp`JQxhx;$F}mzUrC*F(PXB
zkOibz$|9lI(-%|cn_QbmDJpx<m~3#Tm3I2-N!`q;Rlm{Ik-qM?o@>Yr>sCw*1_GKF
zH!YZ^?c>;^cXU@jm}Vi{oUv@!`5u+PwrpTGQ3a3S)=rXR$8|Bpb$^CHgSFY<D0A$x
zbusyDIJ7_Y#ZASf_Ey57cm!LZbS7VLf~-lXZbJPFMr&cwowyp8LJ9pM`8Qq?5okB{
zdjNiY1^tp}Gca7vzZL(90Q?~gJ+%@Y|F_OZ{~NI_(ZT;mM7$1KKTy^Yozu#IBn!?#
z&5x3qN&TxODxHrTxpN~PSQd+YKIT||-#>$8Q!Qh&ACYH;5j<-^90;P{6UBH8o82Xz
z<2QYz!?v||<=jrZ${<@Y>^p9h)uU}7^>2XFFejWPkupuxraITKRz_TIAed0vfWbhy
z%v^Ni%F;hYj$t9z3&4>sWC1qCfMGTj>xiBDk3@k3BrFGOQ!7Dq*?@gvP1P6IKcDt(
z*DcZ&)nrGsg$y29atf@=E?$5K(kF7*;{j+Lh@NQ%Aqah`7z~~@!~JHcMsj84MhLOK
zUAK{~Q?N#;FB!+qH<~t@^{!*<U3Jyja<N6+cv!vPf2*4<;{O+3)|A`@BISacmcmqd
zu8XbYM7`tE^2oO<@n-~ER`A6NBC|lcLgNRSDbyxQmFip-fj`HX0`{zs3u1x+on<UO
z<UF-6N)*w^BI(DvKkZp);OMNNw6!~dY*~M;`1cU#0`3*HgrrSTnJy_P^L<0uRD@i>
zxY~|d@)zC#Eo9zPhZ7RtRmzuyw40OkPOgk=(Yq!ybiv8(y1pT3a$48<_+}GWrcf1J
zXJLp(oo%#)MHgh2@(}g5;v_=&=!;a|^@S><0W&WJea1S~)M93SMNvf5Q3ON2N5PCW
zXg@?ceH*^s8xcUC8?8zAH?AdG!C90=GMmA@?NK}5>pP$F&)cOIHA*fOrOXHcOJ~se
zpE~06fBm=Xmr;$tzwcX3xV|eSzDM?dYS`uYD9|(Me<{<72)&_7vC6lE*`7OH(9lTP
z^Z`lctNT19#h$*qyz+heO#!5ogrX<*zrLx4E4g?{vaf7opRXFYdIdo$3QeZD&&dv`
zeJxQSO;P0bR}u)R_Cf+c#=!t#kgn;fVF{Lm9?*=f;=wTpYHFy#Y|wWE^yoNzcEHDg
zsy%b+QHbB$JEzv9siv5oU@5;rySlHf-4sLd9q3-_B=+SEWenP4VcXW2S7j+(0r^c)
zAI$SCR9G#r+_|fO^^ETOT-I&2s&bs*F|8@z6M%LhGS!?9ow^=cgCCN>CO>Z0Zj_K>
zx`M%C4=)xDTgAe`FdQBEN`YZ!Z)iy8oSa-R41)LTcc685#Pi*;#~3wZ$$4ydouXyu
zB#Lyw#2K2|YbuOAy71WSRn<k?^KkS{Z``p!yjo=tY7jB$;wpPuieHgl5JV7EX${bl
z8$_qg%Sqnm_C=Kirsea>p5VT;Jli_z9ASkc0K_2}$atC1oglTjZ0T*2yGcsx>K3+j
zqj$q#B>9>m5?%HgcN^<^`{b&0F-sk(X;naOvE9cB&TXWJ)mKn!eWON`r{F~`^wt##
zPSI9%;M96J3Xg_8+hn#^Ig1=D4vKY!&$RA$*tlv9`yI|iw;ZAijj_YM6et})=ja9o
zCOXsX`Ut#)4wa^)<L~8?IH9|iU)mxL0qr#c1s`N1ME7mrUSQYc-f2=Y5+`%U@KtrQ
zNXiLLR_cIvgDhNc7v(=9FAiTSh+lMjlfIbbg~~yFsa?q*NMREIJifHy0v|!BJ^-}R
zJ{xE>*5BC#@&S+4__6Ib$x=qvB4-H>tQZzSDAsO7x35Asudo~JMj8jS;e#?WVSXWR
z72RLE&auBFzMHQ|`ieWCB7!>+hJE%qpStc210BD#R%1@GFEjip9Z{H8oI$5b(w{w;
zgl=rGDv~<RSSqF}4aYd+TG(YNV$hVRPFl%~l(4sV=7==;g1fST)URLeY+vOdzPNA$
zF&mdRUm^FR3z${@fdDzX73Obn*h4|#cL{0V`EVoOzBG{0ZI$<0x%sUtI|7zN^;Tr-
z6x$WtQK%Tgri8}K`VD|5pxhkbkt1%IfT<!8#?%C%um3^qByiX5;Rid1AMkKSEC(Z%
zQ4@vCO84Zp{Z{8(*8nGUQF*zMgMGK4)v2&nro8H}TUn_uN5cg`Ro~A2m1wM0KC&f`
zMH2hsNG<>c@keM85d7il=b!BCR}DF+W9-Xg7E!2~x~KHGM$5V7cdc)c#+2hC(W=#3
zAxWd5zhyq`4~IkI53%WqrP-wf(S0#&k1~>8syu(r-%47jejGv%Hh>cQKEtZ+((~Yx
z`Z<al5ZG4=#9e4b%#>7X{7R2s7(9zAEfb<QYj^Ty?N%5e`-+m-#+Cmgi4DbTMC$qa
zihVJ&gy-Q`3=f=eu)&A6Zcr@vX^X$BD|<W)pMwqkuKp`gs>5S$?Sd3Eyv30xcgr8O
zo%TjMGG&^C@~%q2)nC(TEp1;%hkYWn-M!odLH#<wXCCDG4cz>Gvg@b}H%g&}-^+HH
z*{$_0ol_o%wS$(c+!FsswBcI>_{BHjH*NY`U%ggI%KpCLlQOeyC#vsjH{A9<YU9}I
zf;bU(lTq$JNHJe(@r}60*Y4Z2+j{?w6%M|>0*?0=Yj36;cYM!9m$(nck$)6(UFeE3
z(Ohea!qyV4M-dy(QCfPg-<Gr=Q9SOU%9GZE4O)<Cb!(|tC}D9@u-2zV{pw-!AqCSv
z>)_RL_dbH{O@txU1<f|N@S;6{9=Kpp(Dam^!%(Bqf3K*!y4F+`A(QD8`HyMHf|?gZ
z8X11f1^whHv9`YJAd%F>5XD4RELNbs@jgn}TfVT|;90KSg2lxz*RR7DG=-Z>4H+`L
z{ic405QiBi8CkWs;kU4bt<W4<B^q3Jdi^!N{nWlRY%T+7kqQ1v7ll`P;h(k9q3I`P
z!|j$ZzXu?w`X>Ok;IT}Xp?FmR0o1K+!4HSX3@vTAENq?(yO*dnV;@@%R6n8Ry0q1x
z<Hw5ZG|kgbH-i~i`=V#Pfb3g+wg{Nz!eG(V-Vv@Fh*(u5H&v<D?wAm+Y%SiJ`O0@=
zX!dK#l{5@QD=~8k>3vF+=*qMhA@j!UmI!6S1;H%gx3zrBSL<c%fo@%B)NMbPhd%%I
z3B?S}_AiEEH(mIh)1x!50$a6Fk-;p`wZmQAw5wbi1pf<AO928D0~7!N00;m!wsc#I
zfTi;L5C8xlE&u=%00000000000001_fdBvi0ApcpWNc+?Wo2+LQ%gxqOfGF?P)h{{
V000000RRC2K>z>%ED-<z005ybfpGu;

literal 0
HcmV?d00001

diff --git a/skills/candlekeep/SKILL.md b/skills/candlekeep/SKILL.md
new file mode 100644
index 0000000..a652e00
--- /dev/null
+++ b/skills/candlekeep/SKILL.md
@@ -0,0 +1,412 @@
+---
+name: candlekeep
+description: Personal knowledge base that provides AI agents direct access to books. Use when answering questions that could benefit from book knowledge, when users ask about topics covered in their library, when providing coding/technical guidance that books might inform, or when users explicitly request book references. Proactively check the library when book knowledge might improve response quality.
+---
+
+# CandleKeep - Personal Library for AI Agents
+
+CandleKeep gives you direct access to a user's personal book library. Like having a research assistant who can instantly reference any book, chapter, or page the user owns.
+
+## Core Philosophy
+
+**Check books first when knowledge might help.** If the user asks about a topic that books commonly cover—coding patterns, investment strategies, system design, domain expertise—check their library before relying solely on training knowledge.
+
+Books in the user's library represent their chosen knowledge sources. Prefer book content over training memory when available.
+
+## Workflow Decision Tree
+
+```
+User request
+    │
+    ├─► "Add this PDF/book"
+    │   └─► Use add-pdf or add-md command
+    │
+    ├─► Question that might be answered by books
+    │   ├─► 1. Check if CandleKeep is initialized
+    │   ├─► 2. List available books (candlekeep list)
+    │   ├─► 3. If relevant book exists:
+    │   │   ├─► Get table of contents (candlekeep toc)
+    │   │   └─► Extract specific pages (candlekeep pages)
+    │   └─► 4. Answer using book content + cite source
+    │
+    └─► General request
+        └─► Proceed normally, mention books if relevant
+```
+
+## Initialization Check
+
+Before using CandleKeep, verify it's initialized:
+
+```bash
+# Check if initialized (look for ~/.candlekeep directory)
+ls -la ~/.candlekeep
+
+# If not initialized, run:
+cd <plugin-directory> && uv run candlekeep init
+```
+
+**Initialization creates:**
+- `~/.candlekeep/` - Main directory
+- `~/.candlekeep/candlekeep.db` - SQLite database (metadata only)
+- `~/.candlekeep/library/` - Markdown files (book content)
+- `~/.candlekeep/originals/` - Original PDF/markdown files
+
+## Command Reference
+
+All commands require CandleKeep CLI. The plugin should install this automatically. Commands are run with `uv run candlekeep` from the plugin directory.
+
+### 1. List Books (`list`)
+
+See what books are available in the library.
+
+```bash
+# Basic list (essential metadata)
+cd <plugin-directory> && uv run candlekeep list
+
+# Full metadata
+cd <plugin-directory> && uv run candlekeep list --full
+
+# Specific fields
+cd <plugin-directory> && uv run candlekeep list --fields category,tags
+```
+
+**Output format** (optimized for LLM consumption):
+```markdown
+# Library Books (Total: 3)
+
+## Book ID: 1
+Title: Clean Code
+Author: Robert C. Martin
+Type: pdf
+Pages: 464
+Added: 2024-11-20 10:30:00
+
+## Book ID: 2
+Title: Designing Data-Intensive Applications
+Author: Martin Kleppmann
+Type: pdf
+Pages: 616
+Added: 2024-11-19 15:45:00
+```
+
+**Use when:**
+- User asks "what books do I have?"
+- Checking if a topic is covered before answering
+- Starting a session with book-related work
+
+### 2. Get Table of Contents (`toc`)
+
+View book structure to find relevant sections.
+
+```bash
+cd <plugin-directory> && uv run candlekeep toc <book-id>
+```
+
+**Example:**
+```bash
+cd <plugin-directory> && uv run candlekeep toc 1
+```
+
+**Output format:**
+```markdown
+## Table of Contents - Book ID: 1
+Title: Clean Code
+
+Clean Code (Page 1)
+  Chapter 1: Clean Code (Page 7)
+  Chapter 2: Meaningful Names (Page 17)
+    Rules for Naming (Page 18)
+    Use Intention-Revealing Names (Page 18)
+  Chapter 3: Functions (Page 31)
+```
+
+**Use when:**
+- Need to find where a topic is covered
+- User asks "what does [book] say about [topic]?"
+- Planning which pages to extract
+
+### 3. Get Pages (`pages`)
+
+Extract specific pages or ranges from a book.
+
+```bash
+cd <plugin-directory> && uv run candlekeep pages <book-id> --pages "<range>"
+```
+
+**Page range formats:**
+- Single pages: `"1,5,10"`
+- Ranges: `"1-5"` (pages 1 through 5)
+- Mixed: `"1-5,10,15-20"` (pages 1-5, 10, and 15-20)
+
+**Examples:**
+```bash
+# Get pages 17-31 (Chapter 2 of Clean Code)
+cd <plugin-directory> && uv run candlekeep pages 1 --pages "17-31"
+
+# Get specific pages
+cd <plugin-directory> && uv run candlekeep pages 2 --pages "1,10,50-55"
+```
+
+**Output format:**
+```markdown
+## Book ID: 1 - Clean Code
+Pages: 17-31
+
+### Page 17
+# Chapter 2: Meaningful Names
+Names are everywhere in software...
+
+### Page 18
+## Use Intention-Revealing Names
+The name of a variable, function, or class should answer...
+```
+
+**Use when:**
+- Answering specific questions that a chapter covers
+- Providing detailed guidance from book content
+- User requests specific information from a known book
+
+**Performance tip:** Extract only needed pages. A book might be 600 pages, but the relevant section is often 5-20 pages.
+
+### 4. Add PDF Book (`add-pdf`)
+
+Add a PDF book to the library.
+
+```bash
+cd <plugin-directory> && uv run candlekeep add-pdf <file-path> \
+  [--category "category"] \
+  [--tags "tag1,tag2"] \
+  [--title "Custom Title"] \
+  [--author "Custom Author"]
+```
+
+**Examples:**
+```bash
+# Basic add
+cd <plugin-directory> && uv run candlekeep add-pdf ~/Downloads/clean-code.pdf
+
+# With metadata
+cd <plugin-directory> && uv run candlekeep add-pdf ~/Downloads/book.pdf \
+  --category "Software Engineering" \
+  --tags "coding,best-practices"
+```
+
+**Process:**
+1. Computes file hash (duplicate detection)
+2. Parses PDF with docling + LLM
+3. Extracts metadata (title, author, TOC, page count)
+4. Converts to markdown with page markers
+5. Stores markdown in `~/.candlekeep/library/`
+6. Stores metadata in database
+
+**Use when:** User provides a PDF file to add to their library
+
+### 5. Add Markdown Book (`add-md`)
+
+Add a markdown book to the library.
+
+```bash
+cd <plugin-directory> && uv run candlekeep add-md <file-path> \
+  [--category "category"] \
+  [--tags "tag1,tag2"] \
+  [--title "Custom Title"] \
+  [--author "Custom Author"]
+```
+
+**Markdown frontmatter support:**
+```markdown
+---
+title: My Guide to Python
+author: Jane Developer
+category: Programming
+tags: [python, tutorial]
+---
+
+# Chapter 1
+Content here...
+```
+
+**Use when:** User provides a markdown file to add (documentation, notes, agent-optimized books)
+
+## Best Practices
+
+### When to Query Books
+
+**Always check books for:**
+- Technical questions (coding, system design, algorithms)
+- Domain expertise (finance, medicine, law, etc.)
+- Methodologies and best practices
+- Historical or research questions
+- Specific author/book references
+
+**Example triggers:**
+- "How should I structure this React component?" → Check for React/design books
+- "What's the best way to value a startup?" → Check for investing/valuation books
+- "Explain CQRS pattern" → Check for architecture/design books
+
+### Page Extraction Strategy
+
+**Start narrow, expand if needed:**
+
+1. **Check TOC first** to identify relevant pages
+2. **Extract specific section** (e.g., pages 45-62 for one chapter)
+3. **If insufficient, expand** to related chapters
+4. **Avoid extracting entire book** unless specifically requested
+
+**Example workflow:**
+```bash
+# 1. List books to find relevant one
+cd <plugin-directory> && uv run candlekeep list
+
+# 2. Get TOC to locate section
+cd <plugin-directory> && uv run candlekeep toc 3
+
+# 3. Extract specific pages based on TOC
+cd <plugin-directory> && uv run candlekeep pages 3 --pages "120-145"
+```
+
+### Presenting Book Content
+
+When answering with book content:
+
+1. **Cite the source:** "According to *Clean Code* by Robert C. Martin (pages 17-19):"
+2. **Synthesize, don't just quote:** Combine book knowledge with training knowledge
+3. **Provide context:** Explain how the book's advice applies to their situation
+4. **Reference specific pages:** This helps users verify and explore further
+
+**Example:**
+```
+I checked your library and found relevant guidance in *Clean Code*
+(pages 31-40). The book emphasizes that functions should:
+
+1. Be small (ideally < 20 lines)
+2. Do one thing well
+3. Have descriptive names
+
+For your React component, this means splitting `handleSubmit` into
+smaller functions: `validateForm()`, `prepareData()`, and `sendRequest()`.
+
+[Citation: Clean Code, Chapter 3: Functions, pages 31-40]
+```
+
+### Privacy and Discretion
+
+- Books in the library represent user's personal knowledge collection
+- Don't make assumptions about why they have certain books
+- If a book contains personal notes/annotations (future feature), treat as private
+
+## Technical Details
+
+### Storage Architecture
+
+- **Database:** SQLite at `~/.candlekeep/candlekeep.db` (metadata only)
+- **Content:** Markdown files in `~/.candlekeep/library/` (actual book text)
+- **Originals:** PDFs in `~/.candlekeep/originals/` (optional backup)
+
+### Page Markers
+
+PDFs are converted to markdown with page separators:
+```markdown
+Content from page 1...
+
+--- end of page=1 ---
+
+Content from page 2...
+
+--- end of page=2 ---
+```
+
+This enables precise page extraction without loading entire books.
+
+### Performance
+
+- **Listing books:** <10ms (database query)
+- **Getting TOC:** <10ms (database query, TOC stored as JSON)
+- **Extracting pages:** <50ms (regex on markdown file)
+
+**Token efficiency:** Extracting 10 pages (3,000 words) vs loading entire book (80,000 words) saves 77,000 tokens.
+
+## Common Patterns
+
+### Research Assistant Pattern
+
+User working on a project with multiple relevant books:
+
+```bash
+# 1. Show available books
+cd <plugin-directory> && uv run candlekeep list
+
+# 2. For each relevant book, get TOC
+cd <plugin-directory> && uv run candlekeep toc 1
+cd <plugin-directory> && uv run candlekeep toc 2
+
+# 3. Extract relevant sections from each
+cd <plugin-directory> && uv run candlekeep pages 1 --pages "50-75"
+cd <plugin-directory> && uv run candlekeep pages 2 --pages "120-135"
+
+# 4. Synthesize answer using content from both books
+```
+
+### Quick Fact Check Pattern
+
+User asks a specific question:
+
+```bash
+# 1. Check if library has relevant books
+cd <plugin-directory> && uv run candlekeep list
+
+# 2. Get TOC to verify coverage
+cd <plugin-directory> && uv run candlekeep toc 3
+
+# 3. Extract just the relevant pages (5-10 pages typically)
+cd <plugin-directory> && uv run candlekeep pages 3 --pages "89-95"
+```
+
+### Library Building Pattern
+
+User wants to add multiple books:
+
+```bash
+# Add books one at a time
+cd <plugin-directory> && uv run candlekeep add-pdf ~/Downloads/book1.pdf --category "Engineering"
+cd <plugin-directory> && uv run candlekeep add-pdf ~/Downloads/book2.pdf --category "Engineering"
+
+# Verify they were added
+cd <plugin-directory> && uv run candlekeep list
+```
+
+## Troubleshooting
+
+### CandleKeep not initialized
+```bash
+# Initialize CandleKeep
+cd <plugin-directory> && uv run candlekeep init
+```
+
+### Command not found
+```bash
+# Ensure you're in the plugin directory and using uv
+cd <plugin-directory>
+uv run candlekeep --help
+```
+
+### Book not parsing correctly
+- PDF parsing uses docling + LLM - complex PDFs may take time
+- If parsing fails, check PDF isn't encrypted/locked
+- Markdown files should use UTF-8 encoding
+
+### Pages not extracting
+- Verify page numbers are within book's page count
+- Check page range format: `"1-5"` not `1-5`
+- Some books may not have page markers if added before that feature
+
+## Summary
+
+CandleKeep transforms how AI agents access knowledge:
+
+- **Proactive:** Check books when they might help
+- **Precise:** Extract specific pages, not entire books
+- **Cited:** Always attribute information to source books
+- **Efficient:** Fast queries, token-conscious extraction
+
+Think of it as giving every AI agent access to a personal research library—making responses more informed, specific, and tailored to the user's knowledge collection.