← Member help · Chat

Project intent, posture, and session continuity

Audience: Members and builders using Chat OmniLore mode and/or CoTrinity.
As-of: 2026-08-12T15:30Z

Dual-seat “predictive coach” helpers for starting and continuing projects — without Core blackbox prediction tools.

What these features are (and are not)

| Is | Is not |
|----|------------|
| Short intent card + ranked next steps | Automated deep research |
| Risk card from changed file names | Core predict_warmup / Gartner suite |
| Posture rollup from your intents, handoffs, session feed | Machine-learning “user model” |
| Continuity via handoff + session feed | Full chat transcript as product SoT |

Law: Product/fabric facts still come from SoT / product answers. Web digests stay FRONTIER DRAFT.

1. Project intent (Chat)

When: Starting a new project or clarifying a goal.

Ask (OmniLore mode):

> “I’m starting a local CLI tool for log triage — propose a project intent and next steps.”

What the agent should do:

  1. project_intent_propose (optional use_posture=true to use your rolling posture)
  2. Show a draft card (class, confidence, next actions)
  3. On your confirm: project_intent_accept → writes INTENT.md under your workspace projects/

Classes (heuristic): local_tool · member_web · forge_slice · fabric_feature · research · unknown

Propose is read-only; accept writes disk.

2. Test / change risk card

When: You have changed files (git names) or CoTrinity workspace context.

Chat:

> “Risk card for these files: src/app.py, tests/test_app.py, docs/NOTE.md”

CoTrinity (VS Code/Cursor, extension ≥ 1.0.35):

Honesty: Path heuristics only (e.g. test_*.py often HIGH). Suggested commands are not auto-run.

3. User posture (rolling focus)

Stored at workspaces/<you>/posture/posture.json — retrieval snapshot of recent goals, schedules, session-feed snippets, handoff next steps.

| Trigger | Behavior |
|---------|----------|
| user_posture_rollup | Rebuild from intents + feed + optional next_steps |
| session_handoff (with your user) | Auto soft-triggers rollup |
| project_intent_propose(use_posture=true) | Reads posture hints if present |

Read: “What’s my current posture / focus?” → user_posture_get.

Not ML prediction. Empty session feed still rolls up intents/handoff.

4. Session handoff + session feed

End of significant work (or before a fresh chat):

> “Handoff this session: &lt;durable summary&gt;. Next steps: …”

Use a real summary (not a stub). Fill next_steps so posture stays useful.

Continuity pack / facts: Ask for a continuity pack or “what do you remember about my next steps?” (MCP continuity_pack / continuity_search — not product SoT). See CONTINUITY_MEMORY.md.

Session feed: Semantic search over your prior session summaries (continuity memory — not product SoT, not LibreChat Memory UI; ops map: SESSION_FEED_AND_AGENT_MEMORY_PLANES.md). Ask:

> “Search my session feed for seat capture next steps.”

Default privacy: your sessions only.

5. Suggested flow (new project)

1. Propose intent (use_posture if you have history)
2. Accept → INTENT.md
3. Build in CoTrinity; ask for risk card on git files
4. Before leaving: session_handoff with next_steps
5. Next day: propose with use_posture, or “get posture”

Related

End.