connect

Connect — Link Memories into a Narrative Thread

Manually group related memories into a named NarrativeThread. Accepts memory IDs directly, or a topic description that searches for candidates. Claude composes a title and summary from the selected memories; the user can override before saving.

Usage

/memesis:connect <memory-id-1> <memory-id-2> [...]     # Direct IDs
/memesis:connect <topic description>                    # Search and select

Procedure

Path A — Direct memory IDs

1. Parse each argument as a memory ID.
2. Look up each via Memory.get_by_id(id). If any ID is not found, report which ones failed and stop.
3. Display the resolved titles in order so the user can confirm.
4. Compose a thread title (≤80 chars) and one-sentence summary from the memory titles and summaries. Ask the user if they want to keep or change these.
5. Create the NarrativeThread, then one ThreadMember per memory with position starting at 1.
6. Confirm with thread ID, title, and ordered list of linked memory titles.

Path B — Topic description

1. Run Memory.search_fts(topic, limit=10) — wrap the topic in Memory.sanitize_fts_term() to avoid FTS5 operator injection.
2. Display a numbered list of ranked results (title, stage badge, summary excerpt, memory ID).
3. Ask the user: "Which memories should be in this thread? Enter numbers (e.g. 1 3 5) or IDs."
4. Resolve the selection to Memory objects.
5. Continue from step 4 of Path A (compose title/summary, confirm, create).

Both paths — Creation sequence

from datetime import datetime

now = datetime.now().isoformat()

thread = NarrativeThread.create(
    title=composed_title,
    summary=composed_summary,
    created_at=now,
    updated_at=now,
)

for position, memory in enumerate(selected_memories, start=1):
    ThreadMember.create(
        thread_id=thread.id,
        memory_id=memory.id,
        position=position,
    )

Stage Badges

Stage	Badge
instinctive	[instinct]
crystallized	[crystal]
consolidated	[consol]
ephemeral	[ephemeral]

Implementation

Path A — Lookup by IDs

import os, sys
sys.path.insert(0, "${CLAUDE_PLUGIN_ROOT}")
from core.database import init_db
from core.models import Memory, NarrativeThread, ThreadMember
init_db(project_context=os.getcwd())

from datetime import datetime

memory_ids = ["<id-1>", "<id-2>", "<id-3>"]  # from user input

# --- Resolve memories ---
resolved = []
not_found = []
for mid in memory_ids:
    try:
        mem = Memory.get_by_id(mid)
        resolved.append(mem)
    except Memory.DoesNotExist:
        not_found.append(mid)

if not_found:
    print(f"Memory IDs not found: {not_found}")
    sys.exit(1)

# --- Display for confirmation ---
print("Memories to link:")
for i, mem in enumerate(resolved, 1):
    title = mem.title or "(untitled)"
    print(f"  {i}. {title}  (id: {mem.id})")

# --- Compose title and summary (Claude fills these in) ---
composed_title = "<thread title composed from memory titles>"
composed_summary = "<one-sentence summary of what connects these memories>"

# --- Create thread and members ---
now = datetime.now().isoformat()
thread = NarrativeThread.create(
    title=composed_title,
    summary=composed_summary,
    created_at=now,
    updated_at=now,
)

for position, mem in enumerate(resolved, start=1):
    ThreadMember.create(
        thread_id=thread.id,
        memory_id=mem.id,
        position=position,
    )

# --- Confirm ---
print(f"\nThread created: {thread.id}")
print(f"Title: {thread.title}")
print(f"Summary: {thread.summary}")
print("\nLinked memories:")
for i, mem in enumerate(resolved, 1):
    print(f"  {i}. {mem.title or '(untitled)'}  (id: {mem.id})")

Path B — Search by topic then select

import os, sys
sys.path.insert(0, "${CLAUDE_PLUGIN_ROOT}")
from core.database import init_db
from core.models import Memory, NarrativeThread, ThreadMember
init_db(project_context=os.getcwd())

from datetime import datetime

topic = "<user topic description>"
limit = 10

# --- FTS search ---
sanitized = Memory.sanitize_fts_term(topic)
fts_results = Memory.search_fts(sanitized, limit=limit)

if not fts_results:
    print(f"No memories found matching: {topic}")
    sys.exit(0)

# --- Display numbered list for selection ---
stage_badges = {
    "instinctive": "[instinct]",
    "crystallized": "[crystal]",
    "consolidated": "[consol]",
    "ephemeral":    "[ephemeral]",
}
print(f"Memories matching '{topic}':\n")
for i, mem in enumerate(fts_results, 1):
    badge = stage_badges.get(mem.stage, f"[{mem.stage}]")
    title = mem.title or "(untitled)"
    summary = (mem.summary or "")[:100]
    print(f"  {i}. {badge} {title}")
    if summary:
        print(f"     {summary}")
    print(f"     id: {mem.id}")
    print()

# --- After user selects (e.g. "1 3 5" or mixed IDs/numbers) ---
# Resolve selection — this block runs after the user responds
user_selection = [1, 3, 5]  # replace with parsed user input (1-indexed)
selected_memories = [fts_results[n - 1] for n in user_selection]

# --- Compose title and summary (Claude fills these in) ---
composed_title = "<thread title composed from selected memory titles>"
composed_summary = "<one-sentence summary of what connects these memories>"

# --- Create thread and members ---
now = datetime.now().isoformat()
thread = NarrativeThread.create(
    title=composed_title,
    summary=composed_summary,
    created_at=now,
    updated_at=now,
)

for position, mem in enumerate(selected_memories, start=1):
    ThreadMember.create(
        thread_id=thread.id,
        memory_id=mem.id,
        position=position,
    )

# --- Confirm ---
print(f"\nThread created: {thread.id}")
print(f"Title: {thread.title}")
print(f"Summary: {thread.summary}")
print("\nLinked memories:")
for i, mem in enumerate(selected_memories, 1):
    print(f"  {i}. {mem.title or '(untitled)'}  (id: {mem.id})")

Composing Thread Title and Summary

After resolving the selected memories, compose:

• Title (≤80 chars): A descriptive name capturing what ties these memories together. Prefer thematic nouns over generic phrases like "Related memories". Example: "SQLite ORM Migration — Decision Context".
• Summary (≤150 chars): One sentence explaining the thread's connective tissue. Example: "Decisions and corrections from migrating raw sqlite3 to Peewee across the memesis codebase."

Present both to the user before saving:

Thread title: "SQLite ORM Migration — Decision Context" Summary: "Decisions and corrections from migrating raw sqlite3 to Peewee across the memesis codebase." OK to save? Or type new title/summary to override.

If the user says "yes" or similar, proceed to create. If they provide new text, use their version.

Confirmation Output Format

Thread created.

ID:      8e1f2a3b-...
Title:   SQLite ORM Migration — Decision Context
Summary: Decisions and corrections from migrating raw sqlite3 to Peewee across the memesis codebase.

Linked memories (3):
  1. [consol] Peewee Preference Signal  (id: 4f2a...)
  2. [crystal] Memesis Architecture Context  (id: 9b3d...)
  3. [consol] Usage Tracking Fix  (id: 7c1e...)

Examples

Example 1 — Direct memory IDs

/memesis:connect 4f2a8c1e-... 9b3d5f2a-... 7c1e3d8b-...

Resolves each ID, displays their titles for confirmation, then prompts for title/summary approval before creating the thread.

Memories to link:
  1. Peewee Preference Signal
  2. Memesis Architecture Context
  3. Usage Tracking Fix

Thread title: "SQLite ORM Migration — Decision Context"
Summary: "Decisions and corrections from migrating raw sqlite3 to Peewee."
OK to save? (yes / type override)

After confirmation:

Thread created.

ID:      8e1f2a3b-4c5d-6e7f-8a9b-0c1d2e3f4a5b
Title:   SQLite ORM Migration — Decision Context
Summary: Decisions and corrections from migrating raw sqlite3 to Peewee.

Linked memories (3):
  1. [consol] Peewee Preference Signal  (id: 4f2a...)
  2. [crystal] Memesis Architecture Context  (id: 9b3d...)
  3. [consol] Usage Tracking Fix  (id: 7c1e...)

Example 2 — Topic description search

/memesis:connect cron and python path issues

Searches for matching memories and presents a numbered list:

Memories matching 'cron and python path issues':

  1. [crystal] Cron Python Path Fix
     Hourly consolidation cron was broken; /usr/bin/python3 lacks anthropic.
     id: 8e1f2a...

  2. [consol] Usage Tracking Fix
     track_usage() was never called in production; wired into PreCompact via stdin.
     id: 7c1e3d...

  3. [consol] Consolidation Backtest Baseline
     Pre-reduce baseline. Pipeline is now scan → reduce → consolidate → seed.
     id: 2b4f9a...

Which memories should be in this thread? Enter numbers (e.g. 1 2) or IDs.

User replies: 1 2

Thread title: "Cron and Hook Wiring Fixes"
Summary: "Two production fixes to the memesis background pipeline: Python path and usage tracking."
OK to save? (yes / type override)

After confirmation:

Thread created.

ID:      c3d4e5f6-7a8b-9c0d-1e2f-3a4b5c6d7e8f
Title:   Cron and Hook Wiring Fixes
Summary: Two production fixes to the memesis background pipeline: Python path and usage tracking.

Linked memories (2):
  1. [crystal] Cron Python Path Fix  (id: 8e1f2a...)
  2. [consol] Usage Tracking Fix  (id: 7c1e3d...)