Experience Packets
Every tool call, error, and milestone becomes an experience packet with derived internal state - cognitive load, frustration, energy, dominant emotion, stance, urgency. The agent has a mood.
Experience Packets
Every event the agent cares about - a tool execution, an error, a milestone, a session event - is captured as an “experience packet.” Each packet stores the event summary, source references, outcome, and a derived internal state.
The internal state is six dimensions: cognitive load (0-1), frustration (0-1), energy (0-1), dominant emotion (curiosity, frustration, confidence, etc.), stance (motivated, defensive, exploratory, etc.), urgency (low, medium, high).
These are derived from observable signals - error frequency, tool-call velocity, correction patterns, task complexity. Computed from what the agent does, not guessed from what it says.
Two purposes
The first is pattern detection. The Belief Promotion and Self-Model subsystems scan packets for recurring patterns: “this agent gets frustrated with CSS, confidence drops on design tasks, energy is high early in sessions.” These become beliefs and capability scores.
The second is re-entry context. When the agent starts a new session, it gets its recent state: “You finished the last session frustrated with the test suite; you’ve been at high cognitive load for three sessions; here’s where you left off.”
What it is not
The emotional valence model is a heuristic signal extractor, not a psychology model. It is useful for detecting patterns; it is not precise enough to take literally. 22,749 packets captured over 30 days. Running in production.