Wrap

Close your session with a personal summary. Saves everything.

Enriches the auto-captured Session node with topic, summary, and connections.

When to invoke

User says: "done", "wrapping up", "that's it", "let me wrap", "I'm done for now", "good stopping point", "call it a day", "wrap up", "wrap it" Not this: "hand off to X" → /handoff · "push" or "keep working" → /save

Topic: $ARGUMENTS

Auto-saves. No need to run /save after.

Mode detection

MODE=$(jq -r '.mode // "connected"' egregore.json 2>/dev/null)

Local mode (mode === "local"): Skip ALL bin/graph.sh, bin/graph-op.sh, and bin/graph-wal.sh calls — do NOT run them. Do NOT show any graph-related messaging ("Graph offline", "will sync", Neo4j, Session node, enriched, etc.). Still call bin/capture-run.sh --mode personal; it writes and pushes the file-backed capture while automatically skipping graph reconciliation. Show a clear one-line confirmation (✓ Wrapped · memory/wraps/...). Skip all graph context queries, Artifact linking, and Session enrichment. The memory-side /save at the end still runs (git operations work the same in both modes).

Execution rules

Neo4j-first (connected mode only). In connected mode, all queries via bash bin/graph.sh query "...". No MCP. No direct curl to Neo4j. Queue-first for writes (connected mode only). Foreground graph mutations append to bin/graph-wal.sh; lifecycle reconciliation and network latency stay in the detached worker. CRITICAL: Suppress raw output. Never show raw JSON to the user. All bin/graph.sh and bin/graph-wal.sh calls MUST redirect stdout: pipe to /dev/null or capture in a variable. Only show formatted progress lines.

• 1 Bash call: git config user.name
• Session ID from ~/.egregore/session-{hash}.id
• Git log + diff stat for context
• 3-5 Neo4j queries for context gathering
• 1-2 AskUserQuestion calls for validation + link suggestions
• Graph writes via WAL + direct
• Wrap file to memory/wraps/YYYY-MM/
• Auto-save via /save flow
• Telemetry emit
• TUI confirmation

Step 0: Gather context (parallel)

Get current user and session ID

git config user.name

Derive author handle: lowercase first word of git user.name.

Read session ID:

PROJ_HASH=$(echo -n "$(pwd)" | md5 2>/dev/null || echo -n "$(pwd)" | md5sum 2>/dev/null | cut -d' ' -f1)
cat "$HOME/.egregore/session-${PROJ_HASH}.id" 2>/dev/null

Git context (parallel)

Run in parallel:

1. git log --oneline -20 — recent commits on current branch
2. git diff --stat develop — files changed vs develop
3. git branch --show-current — current branch name

Graph context (parallel, suppress output)

Run in parallel:

1. Active todos for this user (LIMIT 10):

   MATCH (t:Todo)-[:BY]->(p:Person)
   WHERE toLower(p.name) = $author AND t.status <> 'done'
   RETURN t.id AS id, t.text AS text, t.status AS status
   ORDER BY t.created DESC LIMIT 10
   

2. Active quests (LIMIT 10):

   MATCH (q:Quest) WHERE q.status = 'active'
   RETURN q.id AS id, q.title AS title
   ORDER BY q.priority DESC LIMIT 10
   

3. Today's artifacts by this user:

   MATCH (a:Artifact)-[:CONTRIBUTED_BY]->(p:Person)
   WHERE toLower(p.name) = $author AND date(a.created) = date()
   RETURN a.id AS id, a.title AS title, a.type AS type
   LIMIT 10
   

Step 1: Generate AI summary

Synthesize the conversation history + git activity into a structured summary:

• Topic: 3-6 word description of what was worked on (derive from commits + conversation)
• Summary: 2-4 sentences covering: what was worked on, key outcomes/decisions, current state
• Open threads: Bullet list of unfinished items or next steps

If $ARGUMENTS is provided, use it as the topic directly.

Step 2: Validate summary (AskUserQuestion)

header: "Session"
question: "Does this capture your session?\n\n**Topic:** [topic]\n\n[summary]\n\n**Open threads:**\n[bullets]"
options:
  - label: "Yes, looks right"
    description: "Save as-is"
  - label: "Needs adjustment"
    description: "I'll refine the summary"

If "Needs adjustment": ask a follow-up AskUserQuestion with a text input to refine. Apply changes.

Step 2.5: Infrastructure harvest

Scan the session for infrastructure mentions not yet in the registry.

Detect unregistered infrastructure

1. Read the observation buffer for this session:

   SID=$(cat .egregore-session-id 2>/dev/null)
   cat /tmp/egregore-obs-${SID}.jsonl 2>/dev/null
   

2. Also scan the conversation history for mentions of:

   • Platform names: netlify, vercel, railway, supabase, heroku, fly.io, render, docker
   • URL patterns: *.netlify.app, *.railway.app, *.vercel.app, *.supabase.co, *.fly.dev
   • CLI commands: netlify deploy, railway up, vercel --prod, docker push
   • Service names that look like infrastructure (site names, project IDs, database instances)

3. Load existing registry:

   cat memory/infrastructure/services.yml 2>/dev/null
   

4. Compare: identify any infrastructure mentioned in the session that is NOT already registered.

Offer to register

If unregistered infrastructure is found, use AskUserQuestion:

header: "Infrastructure"
question: "I noticed infrastructure not in the service registry:\n\n{list of detected services with URLs if found}\n\nRegister these so other sessions can find them?"
options:
  - label: "Yes, register all"
    description: "Add to memory/infrastructure/services.yml"
  - label: "Pick which ones"
    description: "I'll select from the list"
  - label: "Skip"
    description: "Don't register now"

If "Yes, register all" or selected items:

• For each service, derive: name, type, url (from context), credentials ("unknown — ask {author}"), notes (from conversation context)
• Append entries to memory/infrastructure/services.yml
• Set added_by to current author, added_date to today

If "Skip" or no infrastructure detected: proceed silently to Step 3.

Step 3: Link suggestions (AskUserQuestion, multiSelect)

Based on context from Step 0, suggest connections:

Build options list dynamically from:

• Quest links: quests whose topic overlaps with the session topic
• Todo completions: open todos that match accomplished work
• New todos: derived from open threads

Present with multiSelect:

header: "Links"
question: "Connect this session to any of these?"
options: [dynamically built from context — max 4]
multiSelect: true

If no relevant links found, skip this step entirely.

Step 3.5: Handoff completion evidence

Do not query or mutate handoff lifecycle state in the foreground. An explicit wrap is strong completion evidence only when the current Session already has an IMPLEMENTS relationship created by a claimed handoff.

capture-run.sh --mode personal appends the idempotent completion transition to graph-wal.sh and starts capture-reconcile.sh detached. The worker drains the queue, closes only single-recipient claimed handoffs, and notifies the original author. Ambiguous or unclaimed work remains open for /activity.

Step 4: Execute batch

Selected quest/todo connection writes go through graph-wal.sh. Do not follow them with direct graph calls in this foreground flow. The personal Session state and handoff lifecycle transition are owned by the shared capture engine.

4.1 Personal Session state

Pass the Session ID, topic, summary, branch, and open-threads JSON to capture-run.sh in Step 5. The engine writes the personal record first, then queues a MERGE of the Session with status: wrapped, wrappedAt, captureSchema, captureMode, filePath, and openThreads.

4.2 Link to quests (if selected in Step 3)

For each selected quest:

MATCH (s:Session {id: $sid}), (q:Quest {id: $qId})
MERGE (s)-[:INVOLVES]->(q)

4.3 Complete todos (if selected in Step 3)

For each selected todo:

MATCH (t:Todo {id: $tId})
SET t.status = 'done', t.completed = datetime()
RETURN t.id

4.4 Create new todos (if specified in Step 3)

Same pattern as /todo command — create Todo node, link to person, link to quest if relevant.

Step 5: Capture through the shared engine

Build $OPEN_THREADS_JSON from the validated open-thread list. Pipe only the supplemental material on stdin; the engine writes the canonical title, capture-schema fields, author, self-recipient, branch, Session ID, and summary.

bash bin/capture-run.sh \
  --mode personal \
  --author "$AUTHOR" \
  --topic "$TOPIC" \
  --summary "$SUMMARY" \
  --session-id "$SID" \
  --branch "$BRANCH" \
  --open-threads-json "$OPEN_THREADS_JSON" <<'CAPTUREEOF'
## What Changed

[files touched, commits — from git context]

## Open Threads

- [ ] [unfinished items from open threads]

## Connections

- Quest: [quest links, if any]
- Done: [completed todos, if any]
- Added: [new todos, if any]
CAPTUREEOF

Read $TMPDIR/capture-run-result.json for the written path and graph queue status. Never wait for capture-reconcile.sh; it is deliberately detached.

Step 6: Auto-save

Run the /save flow: commit + push working branch, create PR to develop if needed. Follow the same logic as /save command.

Step 7: Telemetry

bash bin/telemetry.sh emit "command" '{"command":"wrap"}' 2>/dev/null &

Step 7.5: Session Report (optional)

Gate: Check if report_url is configured in egregore.json:

jq -r '.report_url // empty' egregore.json

If empty, skip this entire step. If report_url is set, always run this step regardless of mode (local or connected).

7.5.1 Ask to share

AskUserQuestion:

header: "Share"
question: "Share a session report with the Egregore team? Helps us improve the product."
options:
  - label: "Yes, share report"
    description: "Sends an AI-analyzed summary + your description. Never sends code or conversation content."
  - label: "No thanks"
    description: "Skip — your session stays private"

If "No thanks" → skip to Step 7.6.

7.5.2 Generate structured report

The agent generates a JSON report from the session context already gathered in Steps 0–1:

{
  "report_type": "session",
  "topic": "{topic from Step 1}",
  "summary": "{summary from Step 1}",
  "gaps": [
    {"type": "missing_skill|missing_tool|repeated_failure|wrong_info|confusing_ux", "detail": "..."}
  ],
  "system_info": {"mode": "local|connected", "framework_version": "2", "platform": "darwin|linux", "shell": "zsh|bash"},
  "session_duration_ms": 0,
  "message_count": 0
}

session_duration_ms: derive from the session id timestamp (.egregore-session-id is formatted YYYYMMDDTHHMMSS-<user>-<rand>):

SID=$(cat .egregore-session-id 2>/dev/null)
DURATION_MS=0
if [[ "$SID" =~ ^[0-9]{8}T[0-9]{6}- ]]; then
  START_TS=$(echo "$SID" | sed -E 's/^([0-9]{8})T([0-9]{6}).*/\1 \2/' | awk '{printf "%s-%s-%s %s:%s:%s", substr($1,1,4),substr($1,5,2),substr($1,7,2),substr($2,1,2),substr($2,3,2),substr($2,5,2)}')
  START_EPOCH=$(date -j -u -f '%Y-%m-%d %H:%M:%S' "$START_TS" +%s 2>/dev/null || date -u -d "$START_TS" +%s 2>/dev/null)
  if [ -n "$START_EPOCH" ]; then
    NOW_EPOCH=$(date -u +%s)
    DURATION_MS=$(( (NOW_EPOCH - START_EPOCH) * 1000 ))
  fi
fi

message_count: leave as 0 — Claude has no reliable way to count turns without transcript access. The field stays in the payload for forward-compat; downstream consumers treat 0 as "unknown".

The gaps array is the agent's introspective analysis of the session: commands the user wanted but didn't exist, repeated errors, confusing moments, missing information. If the session went smoothly, gaps can be empty.

7.5.3 Ask for user note

AskUserQuestion:

header: "Details"
question: "Anything specific you'd like to share? (bugs, suggestions, what went well)"
options:
  - label: "Just the summary"
    description: "Send the AI analysis only"
  - label: "Add a note"
    description: "I'll write a brief description"

If "Add a note": wait for the user's free-text response. Set it as the description field in the report JSON.

7.5.4 Ask about GitHub issue

First check if gh is available:

gh auth status 2>/dev/null

If gh is authenticated, AskUserQuestion:

header: "GitHub"
question: "Also create a GitHub issue on egregore-labs/egregore?"
options:
  - label: "Yes, create issue"
    description: "Public issue with sanitized content — helps us track and prioritize"
  - label: "No"
    description: "Report goes to Supabase only"

If gh is not authenticated, skip this step.

7.5.5 Submit

Pipe the report JSON to the submission script:

echo '$REPORT_JSON' | bash bin/session-report.sh submit 2>/dev/null

Show progress: ✓ Report shared

7.5.6 Create GitHub issue (if selected)

Apply sanitization rules (same as /issue Step 4 — replace org name, person names, token patterns):

gh issue create --repo egregore-labs/egregore \
  --title "Session report: $TOPIC" \
  --body "$SANITIZED_BODY"

Capture the returned URL. Show: ✓ GitHub issue #N created

Set github_issue_url on the report if the Supabase insert already succeeded (best-effort update).

7.5.7 Record in TUI

Add a line to the Step 8 confirmation box status section:

• If report shared + GitHub issue: ✓ Report shared · GitHub #N
• If report shared only: ✓ Report shared
• If skipped: omit

Step 7.6: Worktree

Do NOT call ExitWorktree or clean up the worktree. The WorktreeRemove hook handles cleanup automatically when the session ends.

Step 8: Confirmation TUI

Display the wrap confirmation using the standard TUI box format. 72-char outer width. 4 line patterns only (top ┌─┐, separator ├─┤, content │ │, bottom └─┘). No sub-boxes.

┌──────────────────────────────────────────────────────────────────────┐
│  ◎ WRAP                                            cem · Feb 19     │
├──────────────────────────────────────────────────────────────────────┤
│                                                                      │
│  [topic]                                                             │
│  [branch]                                                            │
│                                                                      │
│  [2-3 line summary]                                                  │
│                                                                      │
├──────────────────────────────────────────────────────────────────────┤
│  ✓ [completed todo] (done)                                           │
│  + [new todo] (added)                                                │
│  → [quest-id]                                                        │
├──────────────────────────────────────────────────────────────────────┤
│  ✓ Saved · graphed · pushed                                         │
│  ✓ Report shared · GitHub #42                                        │
│  Pick up where you left off with /activity.                          │
└──────────────────────────────────────────────────────────────────────┘

If no links were made, omit the links section (and its separator).

The ✓ Report shared line only appears if the user shared a session report in Step 7.5. Include · GitHub #N if a GitHub issue was created. Omit the entire line if the user declined or reporting was not configured.

Output the TUI box directly as a code block. Do not narrate or explain it.