Reading the process trace: reasoning, actions, outputs, gates, warnings
Camelia's work is transparent. As she streams a response, HIIE parses it into a "process trace" — an ordered timeline of what she thought and did — so you can see the full story behind any answer.
The five event kinds
The trace is built from five kinds of events:
- Reasoning — her prose: her narration and thinking between actions.
- Action — a tool call she executed (for example, a line beginning with "› Camelia →").
- Output — an artifact she produced, such as a chart, a flow, an image, or a CAD model.
- Gate — an approval she requested (a human-in-the-loop popup).
- Warning — an error or limit she hit.
What you see vs. what's behind the scenes
The user-facing answer is her prose only. Actions, gates, produced artifacts, and the internal directives that generate them are stripped from the displayed reply and surfaced in the trace instead. That keeps the conversation readable while still letting you audit every step.
Why it's reliable
The trace is parsed continuously and idempotently as her response grows, and incomplete trailing directives (where a closing tag hasn't streamed yet) are dropped rather than leaking raw markup into the prose. So you won't see half-rendered tool syntax in the chat — only clean prose plus a structured timeline.
Using the trace
Open the process trace when you want to understand why Camelia produced a result: which files she read, which sub-agents or tools she called, what she generated, where she paused for your approval, and whether she ran into any limits. It's the same data that feeds the conversation node graphs, giving you both a step-by-step log and a visual map of the run.