Neura Market
Pro Workflow logo

Pro Workflow

Free

Claude Code learns from your corrections: self-correcting memory that compounds over 50+ sessions. Context engineering, parallel worktrees, agent teams, and 17 battle-tested skills.

FreeFree tier
Type
Open Source
Visit Website GitHub (2.2k stars)

About Pro Workflow

Pro Workflow is a collection of battle-tested patterns and components for Claude Code, developed by power users who ship production code daily. It includes a self-correcting memory loop that compounds improvements over 50+ sessions via a [LEARN] rule workflow (Correct → Propose → Approve → Persist), persistent FTS5-indexed wikis with an auto-research loop (budget-capped BFS over web, arXiv, GitHub, and custom fetchers), a multi-LLM council with provider-agnostic 3-phase deliberation (Anthropic, OpenAI, OpenRouter, Fireworks, etc.), parallel worktrees for zero dead time, quality gates, and context discipline for managing the 200k token budget. The package provides 41 skills, 8 agents, 24 hook events, 23 commands, 3 contexts, 12 rules, and 37 scripts. It also features a modular CLAUDE.md split memory (AGENTS.md, SOUL.md, LEARNED.md), hooks automation with 24 events (e.g., PreToolUse, PostToolUse, SessionStart, SessionEnd), and integration with SQLite + FTS5 for full-text search.

Key Features

Self-correcting memory loop (Correct → Propose → Approve → Persist) that compounds improvements over 50+ sessions
Persistent FTS5-indexed wikis with 9 flavors (research, paper, domain, product, person, organization, project, codebase, incident)
Auto-research loop: budget-capped BFS over web, arXiv, GitHub, and custom fetchers with convergence detection and kill switch
Multi-LLM council: provider-agnostic 3-phase deliberation (Anthropic, OpenAI, OpenRouter, Fireworks, custom)
Parallel worktrees for zero dead time with native claude -w support
41 skills, 8 agents, 24 hook events, 23 commands, 3 contexts, 12 rules, and 37 scripts
Modular CLAUDE.md split memory: AGENTS.md, SOUL.md, LEARNED.md, plus Context Discipline rules (Read before edit, Compact at boundaries, Subagents isolate)
24 hook events for automation: PreToolUse (edit tracking, quality gates), PostToolUse (console.log, secrets scan), SessionStart/Load, SessionEnd, PromptSubmit (drift detection), and more
Quality gates on task completion, subagent lifecycle tracking, and drift detection at 6+ edits
Support for SQLite + FTS5 full-text search with wiki-query (BM25 retrieval with snippets, hybrid BM25 + vector RRF)

Pros & Cons

Pros
  • Open source and free to use
  • Comprehensive set of battle-tested patterns and components (41 skills, 8 agents, etc.)
  • Self-correcting memory that compounds accuracy over time
  • Persistent FTS5-indexed wikis with auto-research loop for knowledge management
  • Multi-LLM council enables diverse model deliberation without provider lock-in
  • Extensive hooks automation (24 events) for fine-grained workflow control
  • Modular context management with split CLAUDE.md files and context discipline rules
  • Well-documented with guides, rules, and script support
Cons
  • Requires Claude Code as a dependency (not a standalone tool)
  • Steep learning curve for new users due to the large number of patterns and components
  • May be overkill for simple or small-scale projects
  • Documentation is detailed but can be overwhelming to navigate initially

Best For

Managing complex software projects with structured memory and context disciplineAutomating literature reviews and research wikis via budget-capped auto-research loopsEnforcing code quality gates and pre-commit linting in Claude Code workflowsBuilding multi-LLM deliberation pipelines for decision-making and analysisMaintaining persistent project documentation and learning logs across sessionsReducing token waste with compact context boundaries and subagent isolationAutomating error handling and learning from test failures

FAQ

What is Pro Workflow?
Pro Workflow is a set of battle-tested patterns, skills, agents, hooks, and rules for Claude Code, designed to improve memory, context management, research, and automation. It is open source and free to use.
How does the self-correction loop work?
When a user corrects a mistake, Pro Workflow proposes a [LEARN] rule. After the user confirms, it persists the rule into LEARNED.md and the SQLite database, compounding accuracy over many sessions.
What is the auto-research loop?
The auto-research loop performs a budget-capped breadth-first search over pluggable fetchers (web, arXiv, GitHub, custom) to compile wiki pages with citation and novelty scoring. It halts on budget cap, depth cap, or after three consecutive non-novel pages, and can be stopped manually via a STOP file.
What providers does the LLM council support?
The multi-LLM council is provider-agnostic and supports Anthropic, OpenAI, OpenRouter, Fireworks, and custom models. It uses Promise.allSettled so one bad provider does not abort the run.
How are wikis stored and queried?
Wikis are stored in a persistent SQLite database with FTS5 full-text indexing. Queries use BM25 retrieval with snippets or hybrid BM25 + vector RRF, and support commands like /wiki ask, related, show, and hybrid.
What is split memory?
Split memory is a modular approach where CLAUDE.md imports separate files: AGENTS.md for workflow rules, SOUL.md for style preferences, and LEARNED.md for auto-populated learnings. This prevents instruction drift and keeps context manageable.