rohitg00/pro-workflow

Pro Workflow – Persistent, Self‑Correcting AI‑Coding Assistant

What it is – Pro Workflow is a plugin‑style toolkit that augments LLM‑based coding assistants (Claude Code, Cursor, Gemini‑CLI, etc.) with a single SQLite store that remembers every correction you make, builds searchable research wikis, and runs automated research loops. It turns a session‑based LLM into a persistent coding partner that learns from you and can retrieve past knowledge automatically.

---

Core ideas

Idea	How it works
Self‑correction memory	When you tell Claude “don’t mock the database in tests”, the /learn‑rule command stores that correction as a rule in SQLite (FTS5‑indexed). On every new session the rule is auto‑loaded, so the same mistake never repeats.
Knowledge plane (wikis)	You can create a persistent wiki with /wiki init <slug>. The wiki is stored on disk, indexed with FTS5, and can be auto‑grown by an auto‑research loop that crawls the web, arXiv, GitHub, etc., respecting a budget you set.
Hybrid retrieval	Wiki queries use BM25 plus optional vector embeddings (OpenAI/Voyage) with reciprocal‑rank‑fusion, giving both keyword and semantic matches.
Multi‑LLM council	A three‑phase deliberation runs across several providers (Anthropic, OpenAI, OpenRouter, …). Failures of one provider don’t abort the run; the transcript is saved as a wiki page.
Quality‑gate hooks	Hook scripts fire on events like SessionStart, UserPromptSubmit, PreToolUse, etc., adding lint‑style checks, secret scanning, cost tracking, and token‑efficiency analysis before the LLM runs.

---

What you get out of the box

• 41 ready‑to‑use skills (e.g., wiki‑builder, smart‑commit, cost‑tracker).
• 8 built‑in agents for planning, reviewing, debugging, cost analysis, etc.
• 23 slash commands such as /learn‑rule, /wrap‑up, /wiki, /develop, /smart‑commit.
• 33 hook scripts covering session lifecycle, file changes, permission denials, and git operations.
• Cross‑agent support – works natively in Claude Code and can be installed in 30+ other agents via the skills add CLI.

---

Typical workflow (60‑second tour)

# Capture a correction once
/learn-rule "Don't mock the DB in tests"

# End the session – learns are persisted
/wrap-up

# Start a research wiki on a topic
/wiki init agent-memory --title "Agent Memory"
/wiki seed agent-memory "memory consolidation in agents"
/wiki research agent-memory --max-pages 5 --budget-usd 0.50

# Ask the wiki during a later prompt – the hook auto‑injects the top hits
/wiki ask "what is episodic memory" --wiki agent-memory

The system automatically loads the rule and relevant wiki snippets the next time you start a Claude session, so you spend less time re‑explaining the same concepts.

Installation

Target	Command
Claude Code (native)	/plugin marketplace add rohitg00/pro-workflow && /plugin install pro-workflow@pro-workflow
Other agents (Cursor, Gemini‑CLI, etc.)	npx skills add rohitg00/pro-workflow
Manual	Clone, npm install && npm run build, then copy skills/, commands/, hooks/ into the agent’s ~/.<agent>/ directories.
After install run /doctor to verify the SQLite store and hooks are active.

---

Who benefits?

• Developers who use LLM coding assistants and get frustrated by repeated corrections.
• Teams that need a shared, searchable knowledge base for research, design decisions, and bug‑fix patterns.
• Anyone wanting cost‑aware LLM usage – the built‑in cost‑tracker and budget‑capped research loops keep API spend visible.

---

Where to learn more

• Docs – https://rohitg00-pro-workflow.mintlify.app/
• Reference guides – references/ folder (modern‑workflows‑2026, context‑engineering, agent‑teams, etc.)
• Demo commands – try /insights, /doctor, /list inside any supported agent.

---

Bottom line – Pro Workflow turns a stateless LLM coding assistant into a persistent, self‑improving partner by persisting corrections, building searchable wikis, and automating research, all driven from a single SQLite database.