v2.0 // ZERO_DEPS // OPEN_SOURCE

LIGHTWEIGHT HARNESS FOR
CLAUDE CODE

> Security hooks (0 tokens) + dev pipeline + smart defaults. Install in one command, save 85% token overhead_

VIEW_SOURCE QUICK_START
neural_claude_code v2.0.0
$ curl -fsSL .../install.sh | bash
[ok] 5 security hooks installed (0 tokens)
[ok] 5 skills: forge, init, git-save, overseer, slop-scan
[ok] 5 compact rules (~135 tokens total)
[ok] outputStyle: concise
---
Token overhead: ~635/msg (down from ~4300)
Run /init to generate your CLAUDE.md_
>
21
TOTAL_FILES
~635
TOKENS/MESSAGE
0
DEPENDENCIES

SECURITY_HOOKS

> Bash scripts outside context. Zero token cost. They enforce, not guide._

DANGEROUS_ACTIONS_BLOCKER

Blocks rm -rf /, force push to main, DROP TABLE, npm publish.

PROMPT_INJECTION_DETECTOR

Detects role override attempts and manipulation patterns in real-time.

OUTPUT_SCANNER

Warns when API keys (Anthropic, OpenAI, AWS, GitHub, Stripe) appear in output.

SENSITIVE_FILE_GUARD

Blocks read/write to .env, credentials.json, SSH keys, .npmrc.

PRE_COMPACT

Saves git state and active plans before context compaction. Recoverable context.

DESIGN_PRINCIPLE

If behavior can be enforced with a hook (0 tokens), don't write a rule. Hooks enforce. Rules guide.

SKILLS

> On-demand tools. Zero tokens when idle. Load only when invoked._

/forge

scan, plan, execute, review, ship

 /forge --full

3-agent deliberation + 4-layer review

 /init

Auto-generate project CLAUDE.md

 /git-save

Conventional commits + safety

 /overseer

Security + quality code review

 /slop-scan

Detect AI slop and tech debt

INSTALLATION

> Get started in 3 commands_

ONE_COMMAND_INSTALL

1. Install everything 
$ curl -fsSL https://raw.githubusercontent.com/brolag/neural-claude-code/main/install.sh | bash
2. Generate project CLAUDE.md 
> /init

Installs 5 hooks, 5 skills, 5 rules. Merges into existing settings.json without overwriting.

> READY TO OPTIMIZE?

21 files. ~635 tokens/msg. Zero dependencies. Production-grade guardrails in one command.

CLONE_REPO
v2.0.0