Skip to content

Latest commit

 

History

History
107 lines (84 loc) · 4.85 KB

File metadata and controls

107 lines (84 loc) · 4.85 KB

Acknowledgments

Attune AI Enhancement Contributors

Hook System & Markdown Agent Format

The hook system and markdown-based agent definition patterns in this framework were inspired by the excellent work in everything-claude-code by Affaan Mustafa (@affaan-m).

Affaan's repository represents 10+ months of battle-tested Claude Code configurations, and we learned valuable architectural patterns from studying his approach:

  • Event-driven hooks (PreToolUse, PostToolUse, SessionStart, etc.)
  • Markdown agents with YAML frontmatter for portable, human-readable definitions
  • Strategic context compaction for managing token windows
  • Continuous learning patterns for extracting reusable knowledge

What We Learned vs. What We Built

Pattern Learned Empathy Implementation
Hook event types Python/Pydantic-based HookConfig with async execution
Markdown agent format MarkdownAgentParser integrated with UnifiedAgentConfig
Markdown commands CommandRegistry with YAML frontmatter parsing and alias resolution
Session persistence Integration with Empathy's state management and trust levels
Context compaction CompactionStateManager with SBAR handoff preservation
Continuous learning SessionEvaluator and PatternExtractor for knowledge retention

Our implementation is original Python code designed to integrate with Attune AI's existing architecture (LangGraph, Pydantic, multi-tier model routing), but the conceptual patterns we learned from Affaan's work were invaluable.

License

everything-claude-code is released under the MIT License, which permits derivative works with attribution. We provide this attribution in good faith and gratitude.

Links


Claude Code & Workflow Philosophy

Boris Cherny (@bcherny), the creator of Claude Code, was also a significant inspiration for this project. His candid posts about how he personally uses Claude Code offered practical insights that shaped how Attune AI approaches agentic workflows:

  • Plan Mode first — Boris advocates using Plan Mode for every non-trivial task, iterating on the plan before switching to execution. This directly influenced Attune's Socratic discovery pattern and our emphasis on scoping before running.
  • Parallel sessions — Running multiple Claude Code sessions concurrently (5+ in terminal, 5-10 on claude.ai/code) demonstrated that orchestrating many agents in parallel is not just possible but productive. This validated Attune's multi-agent architecture.
  • Verification loops — Boris stresses giving Claude a way to verify its own work, noting it can 2-3x the quality of results. This principle is embedded in Attune's validation-first workflow design.
  • Shared CLAUDE.md — His team checks CLAUDE.md into git and collaboratively maintains it as a living document. Attune adopted this pattern directly for project-level instructions.

Boris created Claude Code and then generously shared how he actually uses it — the hard-won, non-obvious workflow patterns that only come from building something and living with it. His detailed posts about plan-first execution, verification loops, and parallel agent sessions represent months of real-world refinement. Studying that work significantly influenced Attune's design, and in several cases taught us lessons that led to meaningful changes in our approach.

Similarly, Affaan's 10+ months of battle-tested Claude Code configurations represent a depth of practical experience that can't be shortcut. The patterns we learned from his hook system, markdown agent format, and context compaction strategies shaped core parts of Attune's architecture.

One of Attune AI's goals is to make the kind of best practices that Boris and Affaan discovered through months of independent work accessible to a wider audience. Without a framework to codify them, each team would have to discover and implement them on their own. Attune packages these proven workflows into reusable, structured tools so that any developer can benefit from them out of the box.

Links


Thank you, Affaan and Boris, for sharing your knowledge with the community.