ckb
Code knowledge base: detect, build, and maintain a chaptered project knowledge base; read the index first, then the relevant chapters.
Skill by 5235aa-web
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/5235aa-web-ckb
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/ckb
What's inside
- Skill name
ckb- Description
- Code knowledge base: detect, build, and maintain a chaptered project knowledge base; read the index first, then the relevant chapters.
- License
- AGPL-3.0
- Instructions
- 162 lines in SKILL.md · view on GitHub
Files (2)
- CKB_SPEC.md
- SKILL.md
SKILL.md (first lines)
# CKB — Code Knowledge Base
> Code knowledge base: detect the project knowledge base at the start of a conversation, build it first if missing, then work.
> Structure is a directory plus chapter files: read `content.md` first to locate chapters, then read only the chapters you need.
> Full specification: `CKB_SPEC.md`.
> **Document positioning**: chapters and `VERSION` are current-state facts (present tense, objective, neutral) — not incident reports, change logs, or modification narratives. Describe what is true now, never the edit that produced it.
---
## Trigger Rules (highest priority)
**At the start of every conversation, run detection immediately after the first valid task request:**
```
1. read_file "knowledgesource/content.md" (relative to the project root) — the read itself is the probe
2. Decide:
├─ Read succeeds → chapter list available → read the relevant chapter files → continue
├─ File not found → fall back to the legacy single-file layout:
│ read_file "knowledgesource/PROJECT_GRAPH.md"
│ ├─ Succeeds → legacy format (v1.x): use it as the authoritative reference and tell the user
│ │ "the knowledge base is in the legacy single-file format; consider migrating to directory + chapter files"
│ └─ Fails → no knowledge base → ask the user whether to build one; enter build mode on approval (see below)
└─ Other errors → report to the user
```
**📖 Read on demand**: detection reads only the directory file (`content.md`), not every chapter. During work, use the per-chapter summary in the directory to locate the relevant chapter and read only that file (e.g. `03-Gateway-System.md`). Read multiple chapters only when the task needs a global view (e.g. a full refactor).
**⚠️ Never use glob to detect the knowledge base**: in this environment glob under-reports directory-prefixed relative patterns (e.g. `knowledgesource/content.md`, `knowledgesource/*`) — it can return zero results while the file exists, which would falsely trigger a rebuild. Always probe with `read_file` or `get_file_info`.
## Build Mode Constraints
While building, it is **strictly forbidden** to:
- answer task requests
- run code (execute_code / execute_shell)
- modify files (write_file / edit_file) — except the knowledge base files being generated
- judge or speculate about the project code
**Allowed operations only**: index, read, trace, and generate knowledge base files (`content.md` + chapter files + `VERSION`).
---
## Build Pipeline (6 stages)
### Stage 1: Full index
```
mcp__codebase_memory__index_repository(repo_path=<project root>, mode="full")
```
Wait for completion; record `total_nodes` + `total_edges`.
### Stage 2: Architecture extraction
```
mcp__codebase_memory__get_architecture(project, aspects=["all"])
```
Extract: languages, clusters, hotspots, boundaries.
### Stage 3: File enumeration and classification
```
glob "**/*.py" "**/*.ts" "**/*.vue" "**/*.json" "**/*.yaml" "**/*.md"
```
Exclude: `__pycache__/`, `node_modules/`, `dist/`, `.git/`, `.pebble/snapshots/`.
Classify: entry / core service / tool / data layer / configuration / types / test / stub / deprecated.
### Stage 4: Relationship tracing
For key calls of each core module:
```
mcp__codebase_memory__trace_path(project, function_name, direction="both", depth=2)
```
Use `depth=3` for high-fan-in hotspots.
### Stage 5: Sampling confirmation
- Pick 1-2 files per directory and read the first 30 lines to confirm
- Entry files, `__init__.py`, and YAML configuration must be read in full
- Cross-check: index results vs actual code
### Stage 6: Generate the knowledge base
Plan the chapter structure, generate chapters, then the directory:
```
1. Plan chapters (5-15):
- Standalone global chapters: 01-Project-Overview; 02-Module-Topology
- One chapter per functional system: based on clusters + directory layout + actual responsibilities
- File index, critical paths, and boundary notes go into the owning system chapter
2. Generate chapter files: knowledgesource/{NN}-{Chapter-Name}.md
- Name: English Title Case, hyphen-separated (e.g. Gateway-System)
- Index: two decimal digits (01, 02, ... 10, 11, 12 ...)
- Shared chapter format: file index (table), critical paths (numbered steps), topology diagram (ASCII), boundary notes (grouped)
3. Generate the directory file: knowledgesource/content.md
4. Write knowledgesource/VERSION
```
**content.md format**:
```
# {Project} Code Knowledge Base — Directory
> Last updated YYYY-MM-DD | {change summary} | {statistics}
01-Project-Overview (positioning, stack, entry points)
02-Module-Topology (module layers and call relationships)
03-{System-Name} ({one-line summary})
...
> Reading guide: read this directory first, then open only the chapters you need.
```
**Knowledge base layout**:From 5235aa-web/pebbleagent, AGPL-3.0.
More from 5235aa-web/pebbleagent
- 0.600score
brag
5235aa-web
把当前项目做成一段 15-25 秒的、可直接发布的宣传视频(launch video),自带动效、配乐和分享文案,最终渲染成 mp4。当用户说"做个发布视频""把项目做成视频""生成宣传视频/预告片""brag 一下"时使用。直接读取项目代码理解产品,不需要先跑起网站或截图。底层用 Hyperframes 渲染。
SkillAFreshness: A★ 2Claude Code +9 - 0.597score
blender-character-workflow
5235aa-web
根据人物设定图在 Blender 制作或修复可编辑角色,组织参考配准、脸部与头发塑形、多视图验收,以及按需扩展全身和绑定。适用于二次元人物还原、完整脸部制作和角色工程交付;不用于单纯生成二维人像或无关的 Blender 场景。
SkillAFreshness: A★ 2Claude Code +9 - 0.589score
deck-guizang-editorial
5235aa-web
电子杂志 × 电子墨水; 10 个版面 + 5 套调色板 (墨水/靛蓝瓷/森林墨/牛皮纸/沙丘)
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
deck-magazine-web
5235aa-web
电子杂志 × 电子墨水风, WebGL 流体背景 + 衬线 display
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
dating-web
5235aa-web
消费感配对仪表板: 信号 ticker + KPI + 30 天柱状 + 趋势
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
deck-hermes-cyber
5235aa-web
黑底 + CRT 网格扫描线 + $ 命令行标题 + 薄荷绿大字 + 三档 tag
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
data-report
5235aa-web
把 CSV/Excel/JSON 数据转成漂亮的可视化报告页
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
deck-course-module
5235aa-web
暖纸背景 + Playfair, 左侧学习目标常驻, 含 MCQ 自测页
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
deck-blueprint
5235aa-web
奶油纸 + 锈红 + 蓝图网格 mask + 黑边硬卡片 + pipeline 盒
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
deck-dir-key-nav
5235aa-web
8 页单色背景, 160px display + 4px accent + Mono 箭头列表
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
deck-graphify-dark
5235aa-web
深夜渐变 + 漂浮 orbs + SVG 力导向图谱 + JetBrains Mono
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
article-magazine
5235aa-web
Substack / Medium 高级感长文排版, 适合公众号、博客发布
SkillAFreshness: A★ 2Claude Code +9
Signals
- Stars
- 2
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 17, 2026 (12 days ago)
- Last release
- Oct 23, 2026 (6 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- AGPL-3.0
- Repository
- github.com/5235aa-web/pebbleagent
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from 5235aa-web
- 0.600score
brag
5235aa-web
把当前项目做成一段 15-25 秒的、可直接发布的宣传视频(launch video),自带动效、配乐和分享文案,最终渲染成 mp4。当用户说"做个发布视频""把项目做成视频""生成宣传视频/预告片""brag 一下"时使用。直接读取项目代码理解产品,不需要先跑起网站或截图。底层用 Hyperframes 渲染。
SkillAFreshness: A★ 2Claude Code +9 - 0.597score
blender-character-workflow
5235aa-web
根据人物设定图在 Blender 制作或修复可编辑角色,组织参考配准、脸部与头发塑形、多视图验收,以及按需扩展全身和绑定。适用于二次元人物还原、完整脸部制作和角色工程交付;不用于单纯生成二维人像或无关的 Blender 场景。
SkillAFreshness: A★ 2Claude Code +9 - 0.589score
deck-guizang-editorial
5235aa-web
电子杂志 × 电子墨水; 10 个版面 + 5 套调色板 (墨水/靛蓝瓷/森林墨/牛皮纸/沙丘)
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
article-magazine
5235aa-web
Substack / Medium 高级感长文排版, 适合公众号、博客发布
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
deck-course-module
5235aa-web
暖纸背景 + Playfair, 左侧学习目标常驻, 含 MCQ 自测页
SkillAFreshness: A★ 2Claude Code +9 - 0.584score
dating-web
5235aa-web
消费感配对仪表板: 信号 ticker + KPI + 30 天柱状 + 趋势
SkillAFreshness: A★ 2Claude Code +9
Cite this page
ckb skill. Agent Mart. https://agentmart.dev/skills/5235aa-web-ckb (last updated 2026-10-24)