← Nav Continuity Layer → Agent Platform Phase 0 · Pictures · Spec
1Principles
Five minutes lost, every session.
- Every conversation starts cold — institutional memory evaporates between sessions
- Manual re-briefing: find the plan, find the receipts, paste the context
- 5 minutes per session x 4 sessions/day = 20 minutes of daily friction
| Problem | Question | Decision |
|---|---|---|
| Context evaporates between sessions | Can the system carry the cognitive load of briefing? | Build the briefing compiler |
Picture
Agent at terminal startup — blank screen, pasting changelogs from yesterday's session while the clock ticks from 0:00 to 5:00
2Performance
Under thirty seconds or it failed.
| Metric | Target | Now |
|---|---|---|
| Session start time | <30 seconds | <30s |
| Re-briefing copy-paste | Zero over 10 sessions | Zero |
| Briefing accuracy | 100% (no hallucinated context) | 100% |
| Script execution time | <5 seconds | <3s |
| Problem | Question | Decision |
|---|---|---|
| No baseline for session start quality | How do we measure "productive in 30 seconds"? | Script timer + manual audit log |
Picture
Split timer — left: old workflow showing 5:00 manual briefing, right: new workflow showing 0:28 automatic briefing with green checkmark
3Platform
Three hooks, one briefing, zero clicks.
| Component | Built | Wired | Working |
|---|---|---|---|
team-activation.sh | Yes | Yes | Yes |
session-bootstrap-context.sh | Yes | Yes | Yes |
WM-NAV.md + working-memory.json | Yes | Yes | Yes |
| Problem | Question | Decision |
|---|---|---|
| Three scripts, unclear data flow | What's the single pipeline? | Hook -> Read -> Synthesize -> Inject |
Picture
Architecture diagram: Hook fires -> Reader gathers plans + receipts + issues -> Synthesizer writes nav-briefing.md -> Context injected before first user input
4Process
Token budget is the design constraint.
| Sprint | What | Effort | Acceptance |
|---|---|---|---|
| S1 | Standardize memory schema | 2d | Schema documented, receipt aggregation working |
| S2 | Build briefing layout engine | 2d | Briefing includes plan, blockers, last 3 receipts |
| S3 | Attach to boot hook | 1d | Zero-click: open terminal, briefing visible |
| Problem | Question | Decision |
|---|---|---|
| Context bloat breaks the LLM window | What's the hard token budget for the briefing? | 2000 tokens max, enforced by the compiler |
Picture
A funnel — wide top showing all available context (plans, receipts, issues, memory), narrow bottom showing exactly 2000 tokens of nav-briefing.md
5Players
Every PRD ships faster from here.
| Job | Struggling Moment | Switch Trigger |
|---|---|---|
| Start productive immediately | Finding the plan, receipts, explaining context to AI | Third session in a row pasting the same context block |
| Problem | Question | Decision |
|---|---|---|
| Every PRD build starts with 5min overhead | Does faster session start compound across PRDs? | Yes — throughput increase is multiplicative |
Picture
A single developer opening their terminal — the agent greets them with today's plan, yesterday's blockers, and the first file to open
1 / 5