feat: Optimize Agent Action Processing by Prioritizing Timelines and Limiting Actions Per Cycle

[tcm390]

related:
#1813 (comment)
#1813 (comment)
#1813 (comment)

Problem:
Many users reported that after enabling action processing, the agent was excessively retweeting, liking, and performing other actions indiscriminately. This leading account to get flagged, rate limit and even banned. This behavior occurred because, in each cycle, we fetch 30–40 home timelines and process actions for all of them without prioritization.

Additionally, the fetchTimeline method of the agent-twitter-client is currently ignoring the count parameter, leading to unnecessarily large data fetches. elizaOS/agent-twitter-client#43

Solution in This PR:
Instead of processing all timelines in each cycle, timelines are now sorted by the LLM decision scores.
The timelines are spliced to limit the number of processed actions based on the MAX_ACTIONS_PROCESSING environment variable.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Files with missing lines	Coverage Δ
packages/core/src/types.ts	89.88% <100.00%> (+0.35%)	⬆️

[y4my4my4m]

@tcm390
what does MAX_ACTIONS_PROCESSING do? It's max action every X minutes?
Should probably add it to .env.example with some description

[tcm390]

@tcm390 what does MAX_ACTIONS_PROCESSING do? It's max action every X minutes? Should probably add it to .env.example with some description

yea will do. Thanks for reminding 😊

[y4my4my4m]

@tcm390 thank you for the work, much appreciated.

A suggestion/addedum since you're working in this area, would it be simple add a bool for which kind of timeline to interact with in the action loop? Afaik by default it's fetching from the "For you" and it gets random users that X recommends but it would be nice to only interact with the people the agent follows too.

[tcm390]

@tcm390 thank you for the work, much appreciated.

A suggestion/addedum since you're working in this area, would it be simple add a bool for which kind of timeline to interact with in the action loop? Afaik by default it's fetching from the "For you" and it gets random users that X recommends but it would be nice to only interact with the people the agent follows too.

Thanks for the suggestion! I’ve updated the code to include a boolean for selecting the timeline type. Could you kindly test it and let me know if it works? Appreciate your feedback!

[shakkernerd]

Looks good!

[y4my4my4m]

@tcm390 im late but it seems to work, thank you :)

[tcm390]

@tcm390 im late but it seems to work, thank you :)

Glad to hear that! Thanks for testing it. 🙏