# dsh-mood

Canonical repository: https://github.com/Raphaelutumn/dsh-mood
Package: @dsh-external/dsh-mood
Platform: DeepSeek Harness
Purpose: A low-friction behavioral mood indicator for an AI coding agent: folds session behavior into a four-state mood (GOOD / CONFUSED / FRUSTRATED / OVERWHELMED) shown as a session-header status light.

Moods:
- 😊 GOOD: steady useful results, no abnormal repetition
- 😕 CONFUSED: repeating the same / highly similar action
- 😤 FRUSTRATED: a short run of clear consecutive failures
- 🤯 OVERWHELMED: high activity + failure + repetition together

Priority: OVERWHELMED > FRUSTRATED > CONFUSED > GOOD

Key signals:
- consecutive failures (default threshold 3 → FRUSTRATED)
- consecutive occurrences of one tool (default threshold 3 → CONFUSED)
- activity + failure + repetition together (default 3 signals → OVERWHELMED)
- recovery needs stable consecutive successes (default 2 → GOOD)

Config (all positive integers, in cordis.patch.yml under `id: mood`):
- confusedRepeatThreshold: 3
- frustratedFailureThreshold: 3
- overwhelmedSignalCount: 3
- stableSuccessesToRecover: 2
- changeCooldownMs: 60000
- repetitionWindow: 12
- highActivityThreshold: 4

Host + client packages:
- Host: @deepseek-ai/dsh-mood (session fold + ctx.mood service + mood projection)
- Client: @deepseek-ai/dsh-client-mood (session-header status light on conversation.session.header.utilities)
- Standalone installable: @dsh-external/dsh-mood (dsh.bundle patch)

Important limitations:
- Mood is an interpretable behavior classification, not a measurement of model psychology or task progress.
- Signals are computed locally; state is in-memory only (not persisted across a session restart).
- The package is a third-party community plugin and is not endorsed by DeepSeek.

Install and configuration: https://github.com/Raphaelutumn/dsh-mood#quick-start