Skip to content

WORLDFLOWAI.md

Latest commit

 

History

History
189 lines (143 loc) · 5.2 KB

WORLDFLOWAI.md

File metadata and controls

189 lines (143 loc) · 5.2 KB
Raw

Everything Claude Code - WorldFlowAI Setup Guide

Quick reference for using the everything-claude-code toolkit with synapse and arbiter projects.

Installed Components

Type Items
Agents architect, build-error-resolver, code-reviewer, planner, refactor-cleaner, security-reviewer, tdd-guide
Commands /build-fix, /checkpoint, /code-review, /learn, /plan, /refactor-clean, /tdd, /verify
Skills backend-patterns, coding-standards, continuous-learning, eval-harness, security-review, verification-loop
Rules coding-style, git-workflow, security, testing
Hooks memory-persistence, strategic-compact, continuous-learning-activator

Quick Start Workflows

Starting a New Feature

1. /plan              # Plan the implementation approach
2. /tdd               # Write tests first
3. /verify            # Validate changes work
4. /code-review       # Self-review before PR
5. /checkpoint        # Save progress state

Debugging Build Errors (Synapse/Rust)

/build-fix            # Analyzes cargo errors and suggests fixes

Code Quality Review

/code-review          # Comprehensive code review
/refactor-clean       # Identify refactoring opportunities

Learning & Memory

/learn                # Extract reusable knowledge from current session
/checkpoint           # Save session state for later resumption

Project-Specific Guidance

Synapse (Rust Workspace)

Best agents for synapse:

  • build-error-resolver - Rust compile errors can be cryptic
  • architect - Multi-crate workspace decisions
  • security-reviewer - LLM data handling requires scrutiny

Typical workflow:

# In synapse directory
claude

# Plan feature
> /plan

# After implementation
> cargo build 2>&1 | head -50  # If errors...
> /build-fix

# Before PR
> /code-review
> cargo +nightly fmt --check && cargo clippy --all-targets -- -D warnings && cargo test

Key synapse patterns:

  • Use parking_lot::{Mutex,RwLock} not std
  • Max 100 char line width
  • Clippy pedantic + nursery enabled
  • Conventional commits required

Arbiter (ML/Python)

Best agents for arbiter:

  • tdd-guide - ML code benefits from test-driven approach
  • architect - Pipeline architecture decisions
  • eval-harness - Model evaluation patterns

Typical workflow:

# In arbiter directory
claude

# Plan experiment/feature
> /plan

# Test-driven development
> /tdd

# Validate
> /verify

Key arbiter patterns:

  • Recall-focused metrics (safety critical)
  • PII/Org sensitivity handling
  • Model lifecycle management

Hooks (Automatic)

These run automatically - no action needed:

Hook When What it does
SessionStart New session Loads recent session context
PreCompact Before /compact Saves state before context reduction
Stop Session end Persists learnings, runs continuous-learning
PreToolUse (Edit/Write) Every ~50 edits Suggests running /compact

Memory Persistence

Sessions automatically save state to ~/.claude/sessions/. To resume:

# Start new session, previous context loads automatically
claude

# Or explicitly reference a session file
> @~/.claude/sessions/2026-01-23-feature-x.tmp

Strategic Compaction

After ~50 tool calls, you'll see a suggestion to run /compact. This helps maintain context quality during long sessions.

When to Use Each Agent

Situation Agent/Command
Planning new feature /plan → planner agent
Designing system architecture architect agent
Fixing Rust compile errors /build-fix → build-error-resolver
Pre-PR quality check /code-review → code-reviewer
Security-sensitive changes security-reviewer agent
Writing tests first /tdd → tdd-guide agent
Cleaning up code /refactor-clean → refactor-cleaner
Validating changes /verify → verification-loop skill

Updating

cd ~/dev/worldflowai/everything-claude-code
git pull

# Symlinks auto-update, but hooks need re-copy if changed:
cp -r hooks/memory-persistence ~/.claude/hooks/
cp -r hooks/strategic-compact ~/.claude/hooks/
chmod +x ~/.claude/hooks/*/*.sh

File Locations

~/dev/worldflowai/everything-claude-code/   # Source repo
~/.claude/agents/                            # Agent symlinks
~/.claude/commands/                          # Command symlinks
~/.claude/skills/                            # Skill symlinks
~/.claude/rules/                             # Rule symlinks
~/.claude/hooks/                             # Hook scripts (copied)
~/.claude/hooks.json                         # Hook configuration
~/.claude/sessions/                          # Session memory files

Troubleshooting

Commands not working:

ls -la ~/.claude/commands/  # Check symlinks exist and point correctly

Hooks not firing:

cat ~/.claude/hooks.json    # Verify config is valid JSON
ls -la ~/.claude/hooks/     # Check scripts are executable

To reset installation:

rm -rf ~/.claude/agents ~/.claude/commands ~/.claude/rules
rm -rf ~/.claude/hooks/memory-persistence ~/.claude/hooks/strategic-compact
rm ~/.claude/hooks.json
# Then re-run installation