WenJunDuan/Rlues - Page 6

session-manager

会话状态管理，保证工作流跨会话持久化

Source text: Chinese

code-quality

Code quality assurance with Linus-style review principles. Enforces simplicity, good naming, proper error handling, and "good taste". Integrates with verification-loop for automated checks.

context7

Intelligent library documentation fetcher using Context7 MCP tools. Auto-activates when: writing code with external libraries, needing API docs, setting up frameworks, or requesting code examples with specific packages. Fetches up-to-date, version-specific…

continuous-learning

Auto-extract reusable patterns from coding sessions. Activates on session end or when /learn command is invoked. Identifies successful patterns, debugging solutions, and architectural decisions worth preserving for future sessions.

cunzhi

寸止协议 - Pause and wait mechanism for human confirmation at critical decision points. Prevents AI from proceeding autonomously past important checkpoints. Core safety and collaboration mechanism.

Source text: Chinese

experience

Experience management for pattern retrieval and deposit. Searches past solutions before development, deposits new patterns after completion. Works with continuous-learning skill for auto-extraction.

knowledge-base

External knowledge base reader. Loads project docs, dev standards, company requirements from .knowledge/ directory. Auto-invoked during requirement analysis, design, and code review phases.

memory

Dual-track memory system combining file-based state (.ai_state/) with Memory MCP for cross-session persistence. Handles session lifecycle, context restoration, and state synchronization.

multi-ai

Multi-AI coordination for Claude Code, Codex CLI, and Gemini CLI. Handles task delegation, context handoff, and result synchronization across different AI engines. Works with orchestrator.yaml.

phase-router

Intent recognition and workflow routing. Analyzes user input to determine task type and routes to appropriate agent. Core decision layer.

riper

RIPER workflow core - Research, Innovate, Plan, Execute, Review. Five-phase development methodology with verification loops. Extended to nine steps for comprehensive project lifecycle.

service-analysis

Service understanding and analysis. Provides service overview, business analysis, architecture analysis, and operations analysis. Use when onboarding to a codebase or understanding service boundaries.

strategic-compact

Intelligent context window management with strategic compaction suggestions. Monitors context usage and suggests /compact at logical boundaries. Saves state before compaction to enable recovery.

verification-loop

Checkpoint-based verification for code quality assurance. Use for: running verification loops, saving checkpoints, evaluating implementations against acceptance criteria. Commands: /verify, /checkpoint.

context7

Library documentation fetcher using Context7 CLI. Provides up-to-date, version-specific documentation for external libraries and frameworks. Uses `npx ctx7` CLI instead of MCP for lower context overhead.

continuous-learning-v2

Instinct-based learning system with confidence scoring. Automatically learns patterns from sessions and evolves them into skills. Supports import/export for team sharing.

cunzhi

寸止协议 - Pause and wait mechanism using cunzhi MCP for human confirmation at critical decision points. Prevents AI from proceeding autonomously past important checkpoints.

Source text: Chinese

eval-harness

Verification loop evaluation framework. Provides graders, metrics, and pass@k evaluation for code quality assurance. Integrates with verification-loop for systematic quality gates.

iterative-retrieval

Progressive context refinement for subagents and complex tasks. Retrieves context incrementally based on actual needs rather than loading everything upfront. Reduces context waste and improves relevance.

agent-teams

Agent Teams orchestration for Claude Code. Coordinates multiple Claude instances working in parallel on shared codebase. Requires experimental flag CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1. New in v8.0, replaces multi-ai skill for Claude Code platform.

Source text: Chinese

code-quality

Linus-style code quality enforcement. Checks taste, simplicity, and engineering rigor.

Source text: Chinese

context7

Library documentation fetcher using Context7 CLI (npx ctx7). Provides up-to-date, version-specific docs for external libraries. CLI version replaces Context7 MCP for lower context overhead.

Source text: Undetermined

continuous-learning

v1 basic learning. Extracts patterns from sessions and stores them in the experience skill.

Source text: Chinese

continuous-learning-v2

v2 Instinct-based learning system. Builds confidence through repeated observations and evolves high-confidence patterns into skills.

cunzhi

寸止确认机制。通过 cunzhi MCP 工具强制暂停等待用户确认。 真正的暂停，不是输出文字假装等待。

Source text: Chinese

eval-harness

Evaluation framework for measuring code agent effectiveness. v8.0: Added Agent Teams collaboration quality metrics.

experience

Experience management. Stores learned patterns, mistakes, and best practices from past development sessions.

iterative-retrieval

Large context iterative retrieval. For tasks requiring information from many files, retrieves in batches to avoid overwhelming the context window.

Source text: Mixed languages

knowledge-base

Project knowledge management. Stores and retrieves project-specific documentation, standards, and technical references.

Source text: Chinese

model-router

Intelligent task-to-model routing based on benchmark data. Routes tasks to Claude Code or Codex CLI based on task type and model strengths. New in v8.0.

phase-router

P.A.C.E. v2.0 complexity routing engine. Analyzes task complexity and routes to optimal path (A/B/C/D) with effort level mapping. v8.0: Added Path D for Agent Teams, effort parameter integration.

riper

RIPER execution cycle: Research → Innovate → Plan → Execute → Review. Core workflow engine for all P.A.C.E. paths.

service-analysis

Service boundary analysis. Identifies service boundaries, dependencies, and integration points in the codebase.

Source text: Chinese

smart-archive

Intelligent context archiving for the 1M token era. Replaces strategic-compact. Works with server-side Compaction API. Archives completed RIPER phases to .ai_state/ instead of emergency compression.

Source text: Chinese

verification-loop

Verify-fix-retry loop with cross-model validation support. v8.0: Added --cross mode for dual-model verification.

Source text: Chinese

agent-teams

Path D parallel execution with Lead-Teammate architecture and context handoff

archive

Context threshold management and done task archival for long sessions

brainstorm

R/D phase enhancement: Socratic questioning, ADR writing, and design confirmation via cunzhi

Source text: Chinese

code-quality

Linus-style six-dimension code review with official plugins and security scanning

Source text: Mixed languages

context7

Library documentation query via ctx7 CLI with deepwiki fallback

Source text: Undetermined