Inspiration
Public agent workflows are full of useful ideas, but public code is not automatically safe or understandable to reuse. We wanted to make reusable patterns traceable, explainable, and rights-aware.
What it does
SpecGraph analyses Python LangGraph workflows and captures their state, nodes, routers, retries, provenance, licence evidence, attribution, and reuse status. It turns that evidence into reviewable guides and an RDF knowledge graph.
How we built it
We built a Python CLI and Codex skill using static AST analysis, an RDF ontology, SHACL validation, licence detection, and a governed corpus of three LangGraph projects.
Challenges we ran into
The hard part was moving beyond graph wiring into meaningful implementation context while avoiding overclaiming runtime behaviour or reuse rights. Large-scale semantic indexing also needs tighter performance bounds.
Accomplishments that we're proud of
We produced seven implementation-aware workflow patterns from real projects, each linked to a source revision, licence, attribution, and reviewable reuse conditions. Restricted code is deliberately withheld from reusable outputs.
What we learned
A graph’s node names are not its meaning. Router logic, state fields, retry bounds, handler side effects, and provenance are what make an agent workflow understandable and reusable.
What's next for SpecGraph
We’ll expand the curated corpus, improve pattern clustering and comparison, support more LangGraph construction styles, and make the Orbit/Transcend semantic path practical for larger repositories.
Log in or sign up for Devpost to join the conversation.