agent_space

People /

Template mapping

How the canonical clear_assist repo maps onto each harness. Static reference — fill in TBDs as we settle each one.

Concept clear_assist (Cloud / Claude Code) Nanobot Hermes
Canonical agent instructions AGENTS.md AGENTS.md AGENTS.md
Claude Code entry shim CLAUDE.md (imports AGENTS.md)
Tool guidance tools.md TOOLS.md TOOLS.md (referenced via AGENTS.md; not auto-loaded)
Owner / user profile USER.md (was context/about-me.md) USER.md USER.md (in ~/.hermes/)
People / contacts context/people.md context/people.md context/people.md
Goals context/goals.md context/goals.md context/goals.md
Scheduling preferences context/scheduling.md context/scheduling.md context/scheduling.md
Heartbeat / cadence HEARTBEAT.md HEARTBEAT.md (referenced via AGENTS.md; not auto-loaded)
Soul / persona SOUL.md SOUL.md (in ~/.hermes/)
Long-term memory memory/MEMORY.md MEMORY.md (top-level, in ~/.hermes/ — Hermes default)
Skills / commands .claude/commands/*.md skills/<name>/SKILL.md skills/<name>/SKILL.md

Notes

  • Canonical content always lives in the cloud / clear_assist layout. Other harnesses get a translated copy at link time.
  • CLAUDE.md is a Claude-Code-only shim that imports AGENTS.md; it has no equivalent on Nanobot or Hermes.
  • "Skills" and "commands" are the same concept — a directory per skill containing a single SKILL.md (Nanobot/Hermes) vs a single .md file in .claude/commands/ (Claude Code).
  • Heartbeat, Soul, and Memory are Nanobot-specific runtime concepts; the cloud harness handles those via session context, not files.
  • Items marked TBD need a definitive answer before we wire up the translation layer.