Understand what your Agentforce agent actually did — visually.
Try it now: https://msrivastav13.github.io/AgentLens/
Load a trace from the Agentforce DX extension and AgentLens shows you:
- Agent Graph — which sub agents talked to each other, how many times, and in what order
- Finite State Machine — the internal orchestration flow for each sub agent: LLM calls, tool executions, variable mutations, and handoffs
- Step-by-Step Inspector — walk through every event with full detail: system prompts, model output, tool inputs/outputs, and variable diffs
- Graph Analysis Report — one-click PDF with degree distribution, connectivity analysis, betweenness centrality, and algorithm explanations
- Open https://msrivastav13.github.io/AgentLens/ (or
index.htmllocally — no server required) - Paste or upload the plan response JSON from the Agentforce DX extension
- Click a sub agent to explore its state machine and step through the trace
Agentforce traces are large JSON blobs. Reading them raw is painful. AgentLens turns them into something you can actually navigate — so you can debug handoff loops, understand why an LLM chose a tool, or figure out where latency is hiding.
Zero dependencies. Runs entirely in the browser. Works offline.
MIT