Workspaces

The Station — Command Center

Created by bonsai init. The Tech Lead lives here alongside the project scaffolding:

File	Purpose	Used by
station/CLAUDE.md	Tech Lead agent navigation	Tech Lead
station/agent/	Tech Lead instruction layer (Core, Protocols, Workflows, etc.)	Tech Lead
station/INDEX.md	Project snapshot — tech stack, phase, document registry	Every agent at session start
station/Playbook/Status.md	Live task tracker — in progress, pending, done	Agents check before starting work
station/Playbook/Roadmap.md	Long-term milestones and phases	Context for where tasks fit
station/Playbook/Plans/Active/	Numbered implementation plans	Tech Lead writes, code agents execute
station/Playbook/Standards/SecurityStandards.md	Hard security rules across all agents	Security protocol references this
station/Logs/FieldNotes.md	Your notes to the agents	Read every session start
station/Logs/KeyDecisionLog.md	Settled architectural decisions	Checked before proposing alternatives
station/Logs/RoutineLog.md	Routine execution history	Updated after each routine run
station/Reports/Pending/	Unreviewed completion reports	Tech Lead reviews during code review

Code Agent Workspaces — The Private Layer

Each code agent gets its own directory. This is what it owns and is allowed to modify:

• Directorybackend/

  • CLAUDE.md — agent-specific navigation
  • Directoryagent/

    • DirectoryCore/ — identity.md, memory.md, self-awareness.md

      • …
    • DirectoryProtocols/ — hard rules

      • …
    • DirectoryWorkflows/ — task procedures

      • …
    • DirectorySkills/ — domain knowledge

      • …
    • DirectorySensors/ — rendered hook scripts (.sh)

      • …

Config and Hooks

File	What it does	Managed by
.bonsai.yaml	Single source of truth — all agents, components, paths	bonsai init / bonsai add / bonsai remove
.bonsai-lock.yaml	Content hashes for conflict detection on re-run	Auto-generated by Bonsai
.claude/settings.json	Hook entries for every sensor	Auto-generated by Bonsai

Context Between Sessions

The agent’s memory resets each conversation. Three things bridge the gap:

Bridge	Who writes it	What it does
Core/memory.md	The agent	Working memory — current state, flags, notes
station/Logs/FieldNotes.md	You	Your notes — changes outside sessions, new requirements
station/Playbook/Status.md	The agent	Task tracker — done, in progress, pending

Tip

If the agent seems to have lost context:

• “Re-read your memory file”
• “Check FieldNotes — I added some notes”
• “Review the current status”