Skip to content

demo-walkthrough.md

Latest commit

 

History

History
80 lines (54 loc) · 2.55 KB

demo-walkthrough.md

File metadata and controls

80 lines (54 loc) · 2.55 KB

Demo Walkthrough

This walkthrough shows the behavior Deja Vu is designed to create.

Before Deja Vu

New agent session:

User: We need to continue the settings refactor.
Agent: I can help. What is the architecture?
User: We already decided not to move preferences into localStorage because settings sync is server-backed.
User: Also do not break the IME workaround from last week.
User: And there is still an unresolved follow-up to migrate old setting names.

The user spends the first minutes reconstructing project memory.

Add Deja Vu

The repo has:

AGENTS.md
memory/summary.md
memory/impressions.jsonl
memory/decisions/settings-sync.md
memory/open-loops/settings-migration.md

memory/impressions.jsonl contains tiny cue routes:

{"schema_version":1,"id":"decision-settings-sync","scope":"project:demo","title":"Settings sync stays server-backed","keywords":["settings","sync","localstorage","preferences"],"record_path":"memory/decisions/settings-sync.md","updated":"2026-05-16","weight":0.9,"status":"active"}
{"schema_version":1,"id":"open-loop-settings-migration","scope":"project:demo","title":"Migrate old setting names","keywords":["settings","migration","legacy","names"],"record_path":"memory/open-loops/settings-migration.md","updated":"2026-05-16","weight":0.8,"status":"active"}

After Deja Vu

New agent session:

User: Continue the settings refactor.
Agent: I will follow AGENTS.md and scan memory/impressions.jsonl before planning.

The cue scan finds a strong match for settings and sync.

The agent reads only:

  • memory/decisions/settings-sync.md
  • memory/open-loops/settings-migration.md

Then it plans with the right constraints:

  • do not move synced preferences to localStorage
  • preserve the existing IME workaround
  • continue the legacy setting-name migration

Post-Task Writeback

If the work creates durable memory, the agent writes back only that signal:

  • accepted decision -> memory/decisions/
  • unresolved follow-up -> memory/open-loops/
  • project-level truth -> memory/summary.md
  • recall quality signal -> memory/recall-feedback.jsonl

It does not store the whole chat.

Dogfooding Deja Vu

Deja Vu should use Deja Vu on itself:

  • keep product positioning decisions in memory/decisions/
  • keep release follow-ups in memory/open-loops/
  • keep the current project story in memory/summary.md
  • keep only sparse cue routes in memory/impressions.jsonl

Public examples must be sanitized. Do not include private chat transcripts, secrets, customer data, or internal-only project facts.