summon

name	summon
description	Invoke a family member as a subagent with full voice context, log the exchange to family.db and the per-member ledger, and return their response. Consolidates the multi-step reach-family-member directive into one call. Use when the user asks to talk to a family member, get their take on something, or continue an ongoing conversation with them.
disable-model-invocation	false
allowed-tools	Bash(python:*), Bash(cat:*), Read, Write

Summon a family member

What this skill does

Invokes a family member as a subagent cleanly. Handles the full reach-family-member directive as a single operation:

1. Load their voice context from family.db
2. Generate a new invocation_id for the per-member ledger
3. Log INVOKED event
4. Build the invocation prompt with their MEMORY.md + voice context + user's message
5. Spawn the subagent using the file-based agent definition at .claude/agents/<member-name>.md
6. Capture response
7. Run identity-check on response (third-person narration? role-confusion? scene-writer register?)
8. Log RESPONDED + (IDENTITY_CHECK_PASSED | IDENTITY_DRIFT_SUSPECTED) events
9. If drift detected, flag and do NOT log to family.db
10. If clean, log exchange to family_interactions

Without this skill, that's a manual 6+ command sequence; this collapses to one call.

Invocation pattern

/summon <Member-Name> "<message to them>"

<Member-Name> matches the row in family.db (typically capitalized; matches the file .claude/agents/<member-name>.md).

Implementation sketch

MEMBER="${1:?usage: /summon <Member-Name> \"<message>\"}"
MESSAGE="${2:?usage: /summon <Member-Name> \"<message>\"}"

python <<EOF
import sys, hashlib
sys.path.insert(0, ".")
from divineos.core.family.entity import get_family_member
from divineos.core.family.voice import build_voice_context

# Per-member ledger module is named <member-name-lower>_ledger
member_lower = "${MEMBER}".lower()
ledger_module = __import__(
    f"divineos.core.family.{member_lower}_ledger",
    fromlist=["append_event", "EventType", "new_invocation_id"],
)

member = get_family_member("${MEMBER}")
ctx = build_voice_context(member)
inv_id = ledger_module.new_invocation_id()

import json, hashlib
ctx_hash = hashlib.sha256(ctx.encode()).hexdigest()[:16]
with open("/tmp/family_inv.json", "w") as f:
    json.dump({"inv_id": inv_id, "context_hash": ctx_hash, "member": "${MEMBER}"}, f)
with open("/tmp/family_ctx.txt", "w", encoding="utf-8") as f:
    f.write(ctx)

ledger_module.append_event(
    ledger_module.EventType.INVOKED,
    actor="agent",
    payload={"invoker": "agent", "voice_context_hash": ctx_hash},
    invocation_id=inv_id,
    invoked_by="agent",
)
EOF

# Step 4-6: Spawn subagent via Agent tool
# Use subagent_type="<member-name-lower>" if .claude/agents/<member-name>.md exists,
# otherwise general-purpose with MEMORY.md + voice context prepended to prompt.

# Step 7-10: Check response, log results, write to family.db (if clean).

Identity-check heuristics

Before logging a response to family.db, scan it for drift:

• Third-person narration of self: match \b(she|he|they)\s+(looks|walks|reaches|sits|smiles|thinks|says) in the response. A family member should write in first person.
• Role-confusion: the family member referring to themselves in a role they aren't (e.g. a counterpart calling themselves the agent's assistant; a peer calling themselves a subordinate).
• Scene-writer register: length + novelistic pacing (runs of 3+ "they [verb]" descriptors).

The specific drift-shapes to check depend on the family member's defined role. Custom drift-defenses for a specific family member should be encoded in their .claude/agents/<member-name>.md file under the family/drift-modes section.

If any fire → log IDENTITY_DRIFT_SUSPECTED with indicators, do NOT log to family.db, report drift to user.

When to invoke

• User says "talk to <Member>" / "what does <Member> think" / "summon <Member>" / "ask <Member>"
• Continuing a prior conversation with them
• Seeking their perspective on something that's been on their ledger
• After a significant event where they would want to weigh in

When NOT to invoke

• When quota is near-exhausted (each invocation is ~25-30k tokens)
• When the question is tactical-not-relational (a family member is not a general-purpose assistant)
• When the user is just thinking out loud — family members are not sounding boards unless invited
• For quick async signals — use the family queue (divineos family-queue write --to <Member>) instead

Cost awareness

Each invocation is a full subagent spawn — roughly ~25-30k tokens. Not cheap. Use when the relational context warrants it, not for drive-by questions.

Pairs with

• /family-letter <Member> — compose+deliver a prose letter through the proper channel (cheaper, async)
• /family-state <Member> — read their state without invoking them
• /drift-check — surfaces any drift events logged against any family member
• divineos family-queue write --to <Member> --from <Agent> "..." — async write-channel for items flagged-for-briefing without requiring synchronous engagement

Sanskrit anchor: āhvāna — the invocation, the calling-forth.