rcc

Reflexive Claude Code

English | 繁體中文

A skills-driven Agentic Context Engineering workflow for Claude Code with TDD-based skill design.

Core Philosophy

The agent maintains and refactors its own core prompts and agent system — not external documents or memory banks.

Skills-driven Agentic Context Engineering means:

• Before each task, the agent reviews the skill library for relevant capabilities
• Users explicitly trigger learning moments through skills like reflecting
• Through deliberate teaching, the agent integrates learnings into its skill library
• Skills are abstract, reusable, and link to reference directories with examples and documentation

What's New in v9.0.0

Major release aligning with current agent engineering best practices.

• Architecture-first planning: planning-agent-systems now requires drawing an architecture flowchart (DOT format) before deciding components — visualize entry points, decision branches, data flow, and parallel lanes before committing to any component
  • Maps workflows to Anthropic's six production patterns (Prompt Chaining, Routing, Parallelization, Orchestrator-Workers, Evaluator-Optimizer, Autonomous Agent)
  • Dependency-driven execution order replaces fixed order — phases assigned by dependency depth
  • Core vs enhancement classification for phased rollout
• Simplicity-first brainstorming: brainstorming-workflows adds a complexity ladder (Level 1-6) that challenges every workflow for the simplest viable approach before designing components
  • Two-layer classification: Anthropic workflow patterns + skill routing patterns
  • Walkthrough-style questions replace inventory-style questions across all role templates
  • Past failures exploration: "What have you tried before? What didn't work?"
• 10-category weakness analysis: analyzing-agent-systems adds Category 10 (Cross-Tool Migration) for detecting .cursorrules, copilot instructions, and other AI tool configs
  • New checks: skill token budget (< 2,000 tokens), disable-model-invocation, allowed-tools, dynamic context injection, compaction strategy
  • Scans .cursorrules, .github/copilot-instructions.md, .windsurfrules, .editorconfig
• Maturity-graded migration: migrating-agent-systems replaces binary detection with 4-level maturity grading (None → Seed → Partial → Established)
  • Seed level imports other AI tool configs as starting context
  • Recommends English for all prompt files for optimal model performance
• Progressive disclosure: Templates and reference tables extracted to references/ across all four skills — all SKILL.md files now under 200 lines

What's New in v8.4.0

• Project context completeness: New weakness category (#9) checks for missing account names, directory conventions, deployment targets, language-specific hooks, and user-root gap analysis
  • analyzing-agent-systems now scans ~/.claude/ alongside .claude/ and compares coverage
  • reflecting classification tree adds user-root scope — cross-project learnings route to ~/.claude/CLAUDE.md or ~/.claude/rules/

What's New in v8.3.0

• Formalized routing patterns: All 20 skills now declare their routing pattern (Tree/Chain/Node/Skill Steps)
  • New routing-patterns.md reference document with selection guide and global routing map
  • Standardized ## Routing section in every SKILL.md
  • brainstorming-workflows now teaches routing pattern selection during agent system design
  • advising-architecture decision tree includes routing pattern classification
  • weakness-checklist adds 7 routing-specific checks
  • Skill Chain Reference tables added to all entry-point skills
  • writing-plugins (rcc-dev) rewritten for full Law 7 compliance

Plugins

This marketplace provides two plugins:

rcc (v9.0.0)

Core ACE workflow with TDD-based skills, task enforcement, and quality reviewers.

Skills: