claude-code-dissect
Skill1 starsChanged 7 months agoMITView on GitHub
Navigate and extract reusable design patterns from the Claude Code 2.1.88 source code for application in our project. It covers nine subsystems: startup/bootstrap, conversation engine, context compaction, tool system, permissions, hooks, multi-agent orchestration, memory, and extensions (MCP/plugins/skills/UI).
In this repository11 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- ascii-renderingSkill
- codex-collabSkill
- hermes-agent-dissectSkill
- market-dataSkill
- mogu-visualSkill
- particle-effectsSkill
- podcast-to-mdSkill
- twitter-scraperSkill
- web-searchSkill
- xiaohongshu-to-markdownSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 27
Run the extraction script (see extract-sources.js in the repo) to restore source files
- Very specific: line 34
Locate the source root: search the project for restored-src/src/Tool.ts using Glob.
- Very specific: line 36
All file paths in the reference files are relative to this SOURCE_ROOT.
- Distinctive: line 56
If the topic spans multiple areas, read the primary reference first, then follow its Neighbors section to related files.
- Distinctive: line 58
If no topic matches, grep the source root for relevant keywords before giving up.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Claude Code Source Dissection
Setup
This skill requires the Claude Code source extracted from its sourcemap. To obtain it:
npm pack @anthropic-ai/claude-code→ get the tarball- Extract
cli.js.mapfrom the tarball - Run the extraction script (see
extract-sources.jsin the repo) to restore source files
The extracted source tree should live somewhere in your project. The skill will auto-discover
it by searching for restored-src/src/Tool.ts (a unique marker file).
How to use this skill
- Locate the source root: search the project for
restored-src/src/Tool.tsusing Glob. The directory containing it (minus/Tool.ts) isSOURCE_ROOT. All file paths in the reference files are relative to thisSOURCE_ROOT. - Identify the topic from the user's request
- Read the matching reference file from
references/below - Read the actual source files listed in the reference — prepend
SOURCE_ROOT/to each path - Present your findings using the output format at the bottom
Routing Table
| Topic | Reference File | Covers |
| Startup, init, bootstrap, state, session storage, git | references/bootstrap.md | Boot sequence, global state singleton, session JSONL, git worktree |
| Conversation loop, API call, streaming, retry, messages | references/engine.md | Query loop, streaming generator, retry strategies, message normalization |
| System prompt, CLAUDE.md, compaction, prompt cache, tokens | references/context.md | 5-level prompt assembly, summarizer agent, cache strategy, token budget |
| Tool definition, execution, schema, deferral, Bash parser | references/tools.md | Tool interface, lazy schema, execution pipeline, ToolSearch, Bash AST |
| Permission, allow/deny, classifier, safety, rules | references/permissions.md | 10-step pipeline, rule matching, YOLO classifier, filesystem guards |
| Hooks, PreToolUse, PostToolUse, lifecycle events | references/hooks.md | Hook definition, matching, execution, permission hooks |
| Agent, subagent, teammate, swarm, fork, multi-agent | references/agents.md | 3 spawn modes, fork cache optimization, swarm backends, file mailbox |
| Memory, MEMORY.md, extraction, session memory | references/memory.md | 3-layer memory, auto extraction, session summary, team sync |
| MCP, plugin, skill, marketplace, IDE bridge, UI, Ink | references/extensions.md | MCP client, plugin lifecycle, skill format, IDE bridge, Ink render engine |
If the topic spans multiple areas, read the primary reference first, then follow its Neighbors section to related files.
If no topic matches, grep the source root for relevant keywords before giving up.
Output Format
Structure your response like this:
Pattern: one-line name Why it exists: 2-3 sentences on the problem it solves Data flow:
ASCII diagram showing runtime flowKey code (from the source, simplified):
// The core abstraction, stripped to essentialsFiles read: list with line numbers for critical sections How to adapt: what to keep, what to change for our project
Architecture Overview
For orientation, here's how the 9 subsystems connect:
User Input
↓
[bootstrap] → init, state singleton, session restore
↓
[engine] → query loop: message → API → stream → tool execution → continue/end
↓ ↓
[context] → prompt assembly [tools] → validate → permission → hooks → call
↓ ↓ ↓ ↓ ↓
[memory] ← compaction [agents] ←──── [permissions] ← [hooks]
↑ ↓
└──── extraction ←────── [extensions] → MCP, plugins, skills, UIEach reference file contains: architecture knowledge, key abstractions, reading order, file table with line counts, and neighbor cross-references.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.