Project Overview: the-heavy-dream

I decided to anchor the-heavy-dream on a single memory contract first: when available, /Users/kristian/Documents/obsidian/agent is the canonical shared context for agent work. This initial pass establishes operating rules and project surfaces, not feature implementation, so future work starts from a consistent source of truth.

What We Built

  • We established a documented operating baseline in AGENTS.md, with explicit guidance to use the Obsidian vault as canonical memory and to prefetch context before substantial work via qmd-prefetch.sh.
  • We defined the current working surface around hx-stomp-patch-set.md, AGENTS.md, and gear.md, which gives the project a clear starting boundary.
  • We set up the project in the Kimaki workspace at /Users/kristian/.kimaki/projects/the-heavy-dream with channel 1483951374474678455 so execution context is stable across sessions.

Why We Built It

  • The main decision is to optimize for continuity of reasoning, not early architecture: shared memory first, implementation second.
  • The evidence supports a true initial state: no recent commits, no recent session history, and no package scripts. That means the highest-value move is to lock in conventions before code diverges.
  • By making memory location and lookup behavior explicit (project folder first, then fallback notes), we reduce ambiguity between agents and prevent fragmented context.

How It Works

  • Before substantial work, the agent should run /Users/kristian/Documents/obsidian/agent/scripts/qmd-prefetch.sh to load relevant memory.
  • Project memory resolution follows a deterministic path from AGENTS.md: prefer /Users/kristian/Documents/obsidian/agent/projects/<slug>/ when it exists; otherwise use vault-level fallback files.
  • Current operation is docs-first and script-light: there are no package entrypoint scripts yet, so execution behavior is driven by documented process rather than code automation.
  • Next implementation steps should preserve this contract and add tooling only where it improves repeatability without introducing heavy infrastructure.