| name | memory |
| version | 1.0.0 |
| description | Session state backend for WDS. Called by wrap, start, and handoff tools — never directly by users. This file-based version writes to progress/ in the project repo. WDS-E replaces this file with an Agent Space backend. |
| agents | ["saga","freya","mimir"] |
WDS Memory — File Backend
Handles all persistent state for WDS sessions. Two operations: save and load.
State lives in progress/ at the project root. This folder is project-scoped — not global, not per-machine.
save
Called by: wrap (step 3), handoff (step 3)
Input:
agent_id — the agent whose state is being saved (saga, freya, mimir)
data — the compiled state block (Wrapped, Context, Plan, Next, Learned, Spec Sync fields)
Steps:
- Ensure
progress/ exists at the project root. Create it if not.
- Write
progress/[agent_id].md with the data block exactly as provided.
- Return:
saved progress/[agent_id].md
load
Called by: start (step 2)
Input:
agent_id — the agent whose state is being loaded
Steps:
- Check if
progress/[agent_id].md exists.
- If found: return the full file content.
- If not found: return nothing. The calling tool handles the fresh-start case.
Notes
progress/ should be in .gitignore. It is machine-local session context, not project content.
- Both save and load are synchronous file operations — no tokens, no async, no IDs.
- WDS-E replaces this file with
tools/memory/SKILL.md that routes to Agent Space instead. The interface (save/load inputs and return values) is identical — only the storage destination changes.