How rohitg00/pro-workflow works
Pro Workflow is an AI-coding workflow system built around persistent correction memory and research wikis. Its main user-to-result loop captures a correction as a learning, stores it in SQLite, then surfaces relevant learnings and wiki material in later sessions. Separate capabilities include wiki authoring and research, hybrid retrieval, multi-LLM deliberation, and skill optimization. The graph combines sampled implementation relationships with README-documented product flows; links for unsampled hooks and wiki internals are limited to those documented flows.
Main entry points: trainer.ts Optimizer trainer · store.ts Learning store · prompt-submit.js Prompt injection · reflect.ts Patch proposal
Components 27
Session memory
- Capture correction
scripts/learn-capture.js - Session startup
scripts/session-start.js - Prompt injection
scripts/prompt-submit.js - Learning store
src/db/store.ts - SQLite store
src/db/index.ts - Learning search
src/search/fts.ts
Knowledge plane
- Wiki authoring
skills/wiki-builder/scripts/wiki-cli.js - Wiki retrieval
skills/wiki-query/scripts/query.js - Auto research
skills/wiki-research-loop/scripts/research-loop.js - Research sources
skills/wiki-research-loop/scripts/source-fetchers - Vector retrieval
src/search/embeddings.ts - LLM council
skills/llm-council/scripts/council.js - Wiki viewer
skills/wiki-viewer/scripts/render.js
Skill optimization
- Optimizer trainer
src/optimizer/trainer.ts - Patch aggregation
src/optimizer/aggregate.ts - Patch application
src/optimizer/apply.ts - Patch limits
src/optimizer/clip.ts - Run persistence
src/optimizer/store.ts - Patch proposal
src/optimizer/reflect.ts - Slow update
src/optimizer/slow.ts - Skill evaluation
src/optimizer/validate.ts - LLM client
src/optimizer/llm.ts - Response parsing
src/optimizer/parse.ts
Workflow controls
- Quality hooks
scripts
Other components
- Coding agent
- LLM providers
- Web and research APIs
Connections 30
- Coding agent → Capture correction: submits correction
README.md - Capture correction → Learning store: saves learning
README.md - Session startup → Learning store: loads learnings
README.md - Session startup → Coding agent: surfaces learnings
README.md - Coding agent → Prompt injection: submits prompt
README.md - Prompt injection → Learning search: searches learnings
README.md - Prompt injection → Wiki retrieval: injects wiki hits
README.md - Learning store → SQLite store: initializes database
src/db/store.ts - Learning search → Learning store: searches records
src/search/fts.ts - Coding agent → Wiki authoring: authors wiki
README.md - Coding agent → Wiki retrieval: queries wiki
README.md - Coding agent → Auto research: starts research
README.md - Auto research → Research sources: fetches sources
README.md - Coding agent → Vector retrieval: requests hybrid search
README.md - Vector retrieval → LLM providers: requests embeddings
src/search/embeddings.ts - Coding agent → LLM council: requests deliberation
README.md - Coding agent → Wiki viewer: browses wiki
README.md - Optimizer trainer → Run persistence: persists run
src/optimizer/trainer.ts - Optimizer trainer → Patch aggregation: aggregates patches
src/optimizer/trainer.ts - Optimizer trainer → Patch application: applies patches
src/optimizer/trainer.ts - Optimizer trainer → Patch limits: limits patches
src/optimizer/trainer.ts - Optimizer trainer → Patch proposal: proposes edits
src/optimizer/trainer.ts - Optimizer trainer → Slow update: runs update
src/optimizer/trainer.ts - Optimizer trainer → Skill evaluation: evaluates skill
src/optimizer/trainer.ts - Patch proposal → LLM client: requests completion
src/optimizer/reflect.ts - Patch proposal → Response parsing: parses response
src/optimizer/reflect.ts - Skill evaluation → LLM client: requests evaluation
src/optimizer/validate.ts - Skill evaluation → Response parsing: parses outcomes
src/optimizer/validate.ts - LLM client → LLM providers: calls provider
src/optimizer/llm.ts - Coding agent → Quality hooks: triggers safeguards
README.md