minni

name	minni
description	Use when the user asks the agent to recall, learn, write, audit, or operate through Minni, or when a task likely benefits from prior local memory. Default automatic behavior is recall-only; do not learn unless explicitly requested.
version	0.1.0
author	Infektyd
tags	["minni","obsidian","llm-wiki","local-first","mcp"]
compatibility	Requires minni MCP server running (node dist/server.js) and the Minni daemon at the canonical per-user socket.

Minni for KiloCode

Use this skill to operate KiloCode's local Minni bridge and the KiloCode-owned Obsidian vault. Minni is shared across multiple agents (KiloCode, Claude Code, Codex, Hermes, OpenClaw) — each has its own vault, but they all talk to the same daemon and can recall each other's notes, tagged with agent_origin.

KiloCode Spine Integration

When loaded as a KiloCode plugin, Minni wires four hooks into the session:

• SessionStart — boots identity context, recent audit, and any pending learnings from the inbox.
• UserPromptSubmit — auto-recalls before each turn and injects ranked vault + daemon results.
• PreCompact — captures scar tissue (failed paths, dead ends) so post-compaction doesn't re-walk them.
• Stop — drafts candidate learnings into the vault inbox (never auto-writes); next session reviews them.

All hook output is wrapped in <sovereign:context version="1" event="..." agent="kilocode" tokens="..."> envelopes containing JSON. Parse the JSON; don't reformat the envelope. Disable with MINNI_KILOCODE_HOOKS=off.

The KiloCode vault lives at ~/.minni/kilocode-vault (override: MINNI_KILOCODE_VAULT_PATH). The Claude Code vault at ~/.minni/claudecode-vault is a peer, not a parent — they share a daemon, not a directory.

Default Behavior

• On tasks that likely benefit from prior local context, call minni_status first if available, then use minni_recall with a narrow query.
• Automatic behavior is recall-only.
• Do not call minni_learn unless the user explicitly asks to remember, learn, save to memory, or write a durable note.
• Do not run AFM extraction, session mining, adapter training, or staging review in v1.
• Keep private session content, adapter files, launchd plists, datasets, DB files, and raw vault contents out of public git.

When to Use

• User asks to recall or search prior memory/context
• User wants to learn, remember, or save something to durable memory
• User asks for audit logs or status of the memory system
• Starting a complex task that benefits from prior context (recall first)
• Finishing a task and wanting to capture learnings
• Delegating to a subagent that needs context (use handoff)
• User asks about Minni or the vault

When NOT to Use

• Pure code-generation tasks with no memory component
• Tasks where the user has explicitly disabled memory
• When the minni MCP server is unavailable

Quick Reference

Action	Tool	Key Args
Recall context	minni_recall	query, agentId: "kilocode", limit: 8
Check health	minni_status	vaultPath (optional)
Learn something	minni_learn	title, content, category, source, agentId: "kilocode"
Quality check	minni_learning_quality	title, content, category, source
Write vault note	minni_vault_write	title, content, section, source
Prepare for task	minni_prepare_task	task, agentId: "kilocode", profile
Dry-run outcome	minni_prepare_outcome	task, summary, changedFiles, verification
Route intent	minni_route	task
Audit tail	minni_audit_tail	limit: 20
Audit report	minni_audit_report	limit: 20
Handoff	minni_negotiate_handoff	task, agentId: "kilocode", toAgent
Compile vault	minni_compile_vault	passName, dryRun: true

Tool Details

• minni_route: Classify whether a task should recall, learn, write a note, show audit, check status, or do nothing.
• minni_status: Check daemon, AFM bridge, vault path, and recent audit state.
• minni_prepare_task: Build a compact task packet with ranked context, source reasons, privacy metadata, and optional AFM distillation.
• minni_prepare_outcome: Build a dry-run post-task outcome packet without writing durable memory.
• minni_recall: Search existing Minni, prepend a vault context pack, and log the lookup.
• minni_learning_quality: Review a potential memory before writing it.
• minni_learn: Write a vault note first, quality-report it, then store the learning through Minni.
• minni_vault_write: Write a structured Obsidian note without durable learning.
• minni_audit_report: Summarize recent memory tool activity.
• minni_audit_tail: Show recent memory audit entries.
• minni_negotiate_handoff: Build an agent-to-agent handoff envelope (identity, top recalls with provenance, scar tissue, open questions, inbox pointer) optimized for another LLM to consume — use before delegating to a subagent or another session.
• minni_compile_vault: Dry-run AFM compile passes: session_distillation, synthesis, procedure_extraction, reorganization, pruning.

Slash Commands

• /minni:recall <query> — quick recall against the KiloCode vault + daemon.
• /minni:learn — commit a durable learning (vault-first, quality-checked).
• /minni:status — daemon + AFM + vault health.
• /minni:audit — recent tool activity from the audit log.
• /minni:prepare-task <task> — ranked task packet before complex work.
• /minni:prepare-outcome — dry-run outcome packet, no writes.

Vault Rules

The KiloCode vault defaults to ~/.minni/kilocode-vault (override: MINNI_KILOCODE_VAULT_PATH).

• raw/ is immutable raw sources.
• wiki/ is agent-maintained synthesis.
• schema/AGENTS.md is the operating schema.
• index.md catalogs pages.
• log.md and logs/YYYY-MM-DD.md are append-only transparency logs.
• inbox/ holds candidate learnings drafted by the Stop hook awaiting next-session review.

For Karpathy-style LLM wiki behavior, prefer short sourced pages with Obsidian wikilinks over opaque hidden memory. The vault is the visible surface; SQLite/FTS/FAISS is the recall machinery.

Cross-Agent Awareness

When a recalled snippet has agent_origin other than kilocode (e.g., Claude Code wrote it), treat it as authoritative for what that agent concluded — but verify before acting on it in your own context. If it's load-bearing, recall scoped to that agent or read the source note directly.

Workflow Patterns

Starting a complex task

1. Call minni_status to check system health
2. Call minni_prepare_task with the task description and agentId: "kilocode"
3. Review the returned brief, constraints, relevantSources, and risks
4. Proceed with implementation using the context

Finishing a task

1. Call minni_prepare_outcome with task summary, changed files, and verification results
2. Review outcomeDraft.learnCandidates — if any are worth committing, call minni_learn
3. The doNotStore list is non-negotiable: never commit those even if asked

Delegating to another agent

1. Call minni_negotiate_handoff with the task, your agent ID (kilocode), and the target agent
2. Pass the returned envelope to the subagent as context
3. The envelope includes top recalls, scar tissue, and open questions

Environment Variables

Variable	Default	Description
MINNI_KILOCODE_VAULT_PATH	~/.minni/kilocode-vault	KiloCode vault location
MINNI_KILOCODE_AGENT_ID	kilocode	Agent identity string
MINNI_KILOCODE_WORKSPACE_ID	workspace-<dir>	Workspace identifier
MINNI_KILOCODE_HOOKS	on	Set to off to disable hooks
MINNI_SOCKET_PATH	~/.minni/run/minnid.sock	Minni daemon socket
MINNI_AFM_HEALTH_URL	http://127.0.0.1:11437/health	AFM bridge health URL

Red Flags

• STOP if minni_learn is called without explicit user request. Automatic behavior is recall-only.
• STOP if you're about to write secrets, tokens, raw logs, or local absolute paths to the vault.
• STOP if doNotStore items from prepare_outcome are being committed — honor them always.
• STOP if hook output envelope is being reformatted — parse the JSON inside, don't touch the wrapper.