agi-merge / waggle-dance

Knowledge work automation with AI agents

Home Page:https://waggledance.ai

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

off-by-one execution slowdown

jondwillis opened this issue · comments

commented

There is a bug which manifest such that n-1 tasks are actually executed when their parent dependencies are not yet completed

so for plans with linear paths, e.g. 1-0, 1-c, 2-0, 2-c, a bug occurs. first, 1-0 waits fully for the plan to complete before being started, whereas it should start immediately after 1-c has been added. then, 1-0 and 1-c executes normally. however, 2-0 never starts.

similarly, for larger plans, e.g. 1-0, 1-1, 1-2, 1-c, the n-1 phenomenom manifests again, until either 1-0 or 1-1 complete, after which 1-2 is freed up to execute. i assume the system will get stuck again after 1-c, however, other errors terminate the system first.

  • Add sanity checks for the structure of the "DAG" checkForCircularDependencies checkForUnreachableTasks
  • Prompt engineer: ask nicely to avoid invalid DAGs
  • Auto-repair graph (server-side?)
  • Work on moving away from "busy/active waiting" (await sleep in WaggleDanceAgentExecutor) to promises