How Darc Works

Understand how Darc extracts high-value context from agent sessions and turns it into reusable assets.

Resolution Arcs

When a coding agent successfully resolves a complex issue, Darc analyzes the entire prompt chain and action history to summarize the "Resolution Arc".

Captures the exact constraint that solved the bug.
Excludes dead-ends and hallucinations to keep context pristine.
Formats as a crisp Markdown summary ready for agent ingestion.

ARCWebSocket Memory Leak

"The fix requires closing the ping interval explicitly when the connection drops, otherwise the V8 GC holds the context indefinitely."

Applied in src/socket.ts

<PromptContext>
  <Context source="Darc">
    <Constraint>
      Do not use standard JSON.stringify 
      for large payloads due to node 
      crash history in previous session.
      Use fast-json-stringify instead.
    </Constraint>
  </Context>
</PromptContext>

Continuity Across Sessions

Start new tasks with the compounding knowledge of past tasks. Darc dynamically injects extracted context constraints ensuring AI teammates never repeat historical mistakes.