ejs-session-wrapup
>
Install
mkdir -p .claude/skills/ejs-session-wrapup-mcfuzzysquirrel && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/15964" && unzip -o skill.zip -d .claude/skills/ejs-session-wrapup-mcfuzzysquirrel && rm skill.zipInstalls to .claude/skills/ejs-session-wrapup-mcfuzzysquirrel
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
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.About this skill
EJS Session Wrap-Up
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.
Steps
-
Review the Session Journey for completeness
- Read the current journey file at
ejs-docs/journey/YYYY/ejs-session-YYYY-MM-DD-<seq>.md - Identify any sections that are incomplete or missing context.
- Read the current journey file at
-
Finalize all sections Complete each section with coherent summaries:
- Interaction Summary — ensure all key exchanges are documented
- Agent Collaboration Summary — which agents participated and their contributions
- Sub-Agent Contributions — if sub-agents were involved, ensure their decisions and handoffs are captured
- Agent Influence — suggestions adopted vs. rejected, human overrides
- Experiments / Evidence — what was tried and what happened
- Iteration Log — pivots, reversals, or refinements
- Decisions Made — all decisions with reason and impact
- Key Learnings — technical, prompting, and tooling insights
- If Repeating This Work — do this, avoid this, watch out for
- Future Agent Guidance — prefer/avoid patterns for future agents
-
Populate machine extracts Fill in the
## MACHINE EXTRACTSsection 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:
- Introduces or changes a system boundary (service, datastore, topology)
- Changes a public contract (API, schema, protocol)
- Alters security, privacy, or compliance posture
- Requires choosing among credible alternatives with meaningful trade-offs
- Has long-lived or hard-to-reverse consequences
- Changes engineering process or workflow for future work
-
Update
decision_detectedfield- Set to
trueif an ADR is warranted,falseotherwise.
- Set to
-
Create ADR if warranted
- Use template:
ejs-docs/adr/0000-adr-template.md - Save to:
ejs-docs/adr/NNNN-<kebab-title>.md(next available number) - Link the ADR back to the session journey and vice versa via
adr_links.
- Use template:
-
Confirm finalization
- Inform the user:
"Session finalized: ejs-session-YYYY-MM-DD-<seq>" - If an ADR was created, mention it:
"ADR NNNN created: <title>"
- Inform the user:
Contextual References
- ADR template:
ejs-docs/adr/0000-adr-template.md - Lifecycle patterns:
ejs-docs/session-lifecycle-patterns.md - Database tool:
scripts/adr-db.py
Key Principle
By 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.
Checkpoint vs. Full Finalization
| 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 |
When to Perform a Checkpoint
- 3+ unsaved interactions have accumulated since the last save (an interaction is one human prompt and the corresponding agent response)
- A significant decision has been made but not yet written to the journey
- A large, context-intensive operation is about to start
- 5+ exchanges have occurred since the last journey file save
- Substantial work completed but user has not signalled session end