Finalize an Engineering Journey System (EJS) session by completing all journey sections, populating machine extracts, evaluating the ADR decision rubric, and optionally creating an ADR document.
Use this skill when a session is ending — for example when the user says "wrap up", "finalize session", "end session", "ship it", "commit this", or "commit and push".
This skill also applies for context-threshold checkpoints — proactive mid-session saves that prevent documentation loss if context runs out. See the Checkpoint vs. Full Finalization section below for the differences.
Review the Session Journey for completeness
ejs-docs/journey/YYYY/ejs-session-YYYY-MM-DD-<seq>.mdFinalize all sections Complete each section with coherent summaries:
Populate machine extracts
Fill in the ## MACHINE EXTRACTS section with structured summaries:
INTERACTION_EXTRACT — compact summary of the collaboration trailDECISIONS_EXTRACT — list of decisions with rationaleLEARNING_EXTRACT — transferable insightsAGENT_GUIDANCE_EXTRACT — guidance for future agentsSUB_AGENT_EXTRACT — sub-agent contributions (if applicable)Evaluate the ADR decision rubric Create an ADR only if at least one of these criteria is met:
Update decision_detected field
true if an ADR is warranted, false otherwise.Create ADR if warranted
ejs-docs/adr/0000-adr-template.mdejs-docs/adr/NNNN-<kebab-title>.md (next available number)adr_links.Confirm finalization
"Session finalized: ejs-session-YYYY-MM-DD-<seq>""ADR NNNN created: <title>"ejs-docs/adr/0000-adr-template.mdejs-docs/session-lifecycle-patterns.mdscripts/adr-db.pyBy session end, most of the journey should already be populated from continuous updates. Finalization is a quick review and completion step, not a full reconstruction effort.
| Aspect | Checkpoint (mid-session) | Full Finalization (session end) |
|---|---|---|
| Trigger | Context getting large, 3+ unsaved interactions, before heavy operations, 5+ exchanges since last save | User signals session end |
| Sections updated | Interaction Summary, Decisions Made, Experiments, Iteration Log, Key Learnings | All sections reviewed and completed |
| Machine extracts | Not populated | Populated in full |
| ADR rubric | Not evaluated | Evaluated; ADR created if warranted |
| Goal | Preserve work-in-progress against context loss | Produce a coherent, complete record |