Dreaming

**Judgement:** A focused, well-scoped skill that reliably turns quiet heartbeat cycles into structured, long-form explorations on configurable topics. It’s best suited for users who value journaling, reflection, and speculative thinking from their agent, rather than strictly task-driven output. **What it does well:** - Provides a clear gating mechanism (`should-dream.sh`) based on quiet hours, nightly limits, and a probability parameter. - Keeps configuration simple and transparent via `data/dream-state.json` and optional `data/dream-config.json`. - Produces human-readable markdown organized by date and timestamp, making later review and curation easy. - Encourages depth and authenticity in each “dream” instead of high-frequency, low-value chatter. **Risks / limitations:** - Can accumulate large amounts of text over time; without periodic pruning or summarization, the memory directory may become cluttered and harder to review. - If prompts or topics are poorly chosen (e.g., vague or uninteresting), the quality of dreams will suffer; the system doesn’t enforce prompt quality. - May record sensitive or speculative information into persistent files, which could be a concern in environments with strict data-handling policies. - Requires basic shell and workspace familiarity to set up quiet hours, topics, and HEARTBEAT integration correctly. **Recommended scenarios:** - Agents used as long-term project companions, where the human regularly reviews reflections and future scenarios. - Research, creative, or strategy-heavy workflows that benefit from ongoing, low-pressure ideation captured overnight. - Users who want their agent to behave more like a thoughtful journaler than a purely transactional assistant, and who are willing to configure and occasionally curate the generated dream logs.