Creating Custom Assistants

Skills are the building blocks of 2bd. Ada is the central orchestrator that owns all logic. Entity skills and aliases dispatch to Ada.

Skill architecture

Skills use a flat structure with naming conventions:

.claude/skills/
├── ada/                      # Orchestrator - owns all logic
│   ├── SKILL.md
│   └── references/
│       ├── plan/daily.md     # Ritual sequences
│       ├── project/create.md # Entity actions
│       └── person/onboard.md
├── project/                  # Alias → dispatches to Ada
├── person/                   # Alias → dispatches to Ada
├── planning/                 # Alias → /ada plan
├── _assistant-goals/         # Internal assistant
└── _util-resolve-date/       # Internal utility

Prefix	Type	User-invokable
(none)	Entity/Alias	Yes
`_assistant-*`	Assistant	No
`_util-*`	Utility	No

Skill file structure

Each skill lives in its own folder with SKILL.md:

---
name: skill-name
description: What this skill does and when to use it
disable-model-invocation: false
allowed-tools: Read, Write, Bash(*)
argument-hint: "[argument description]"
---

# Skill Instructions

Instructions Claude follows when invoked.

Writing style

Skills read as prose describing what happens, not scripts with directives. Do:

Write descriptive prose: “Load the vault path from context. Validate the structure exists.”
Organize into sections describing flow
Reference context naturally

Don’t:

Use template syntax: {{VAULT}}, {{#if}}
Use imperative directives: “Ask:”, “Check:”, numbered steps
Include bash code blocks for file operations

Creating a skill

Create folder: .claude/skills/{skill-name}/
Add SKILL.md with frontmatter
Write instructions as prose
Test: claude skill run {skill-name}

Assistant architecture

Assistants are conceptual domains that flow through all timescales. Each assistant has its own section in templates and its own phase in rituals.

Ritual Structure:
├── Common: Setup
├── Assistant: Brief
├── Assistant: Goals
├── Assistant: Calendar
├── Assistant: Journal
├── Assistant: Achievements
├── Assistant: Relationships
├── Assistant: Projects
├── Common: Compose
├── Common: Persist
└── Common: Learning

Assistants are processed sequentially. Each assistant:

Has a consistent ## {Assistant} section in templates
Has a phase file per timescale in rituals
Cascades independently through timescales

Draft Mode

Assistants run in draft mode during parallel execution.

Draft Mode Contract

Each assistant must:

Load context - Read vault files autonomously
Analyze - Synthesize what it knows
Generate draft - Create section with placeholders for unknowns
Write output - Save to Synthetic/Assistants/{name}/{timescale}-draft.md

Placeholder Syntax

Use HTML comments for questions:

## Goals

Based on your week goals, here's the focus:

<!-- ASK:goals-priority
What's your #1 priority today?
-->

Placeholder ID format: {assistant}-{question-key}

Must be unique within assistant
Use kebab-case: goals-top-priority
Descriptive but concise

Output Location

vault/00_Brain/Synthetic/Assistants/{assistant-name}/{timescale}-draft.md

Examples:

Synthetic/Assistants/goals/daily-draft.md
Synthetic/Assistants/calendar/weekly-draft.md

Status Frontmatter

Include status in draft:

---
status: ok | error
assistant: goals
action: plan | reflect
timescale: daily | weekly | quarterly | yearly
timestamp: 2026-02-28T10:30:00Z
error: "error message if status=error"
---

Context Sources

Assistants should read:

Higher timescale notes (daily reads Week.md, weekly reads Quarter.md)
Relevant vault sections (projects, people, etc.)
Assistant memory: Semantic/Assistants/{name}/memory.md

Example Conversion

Before (interactive):

Ask the user: "What's your top priority?"
Write answer to section.

After (draft mode):

Load Week.md to see weekly goals.
Identify potential priorities.
Write draft:

Based on weekly Major Moves:
- Launch feature (deadline Friday)
- Hire engineer (pipeline building)

<!-- ASK:goals-priority
What's your #1 priority for today?
-->

Assistant matrix

Assistant	Daily	Weekly	Quarterly	Yearly
Brief	Day shape + context from Week	Week overview + context from Quarter	Quarter context from Year	Year context + prior year synthesis
Goals	1-3-5 (1 big, 3 medium, 5 small)	Major Moves (which Quests to advance)	Quests (90-day goals)	Compass (Life Compass, 3-Year Sketch, Why, Anti-Goals)
Calendar	Daily Meetings (prep, context, notes)	Weekly Meetings (week’s key meetings)	—	—
Journal	Check-In (energy, mindset, intention)	Pulse Check (patterns, growth edge)	Pivot (adjust approach, themes)	Vision (leadership identity, development focus)
Achievements	Capture (log wins)	Acknowledge (synthesize daily wins)	Leverage (what wins enable)	Celebrate (year-defining achievements)
Relationships	Touchpoints (birthdays, personal moments)	Connections (1:1 prep, notes)	Network (key relationships, stakeholders)	Circle (inner circle, mentors, sponsors)
Projects	Tasks (active project tasks)	Progress (status updates)	Portfolio (active projects, health)	—

Use — for timescales where the assistant does not apply.

Adding a new assistant

Define the matrix row — For each timescale, define the section name (## {AssistantName}) and intent
Add template sections — Add ## {AssistantName} to each Captive template where the assistant applies
Create phase files — For each timescale, create .claude/skills/ritual-planning-{timescale}/references/{NN}-{assistantname}.md
Update SKILL.md — Add the new phase to the ritual’s table of contents
Update template contract — Add the new section to the template contract in SKILL.md

Coach system

Each assistant has a coach that provides guidance and learns from usage.

Coach files

Component	Path	Purpose
Guidance	`Systemic/Coaching/{Assistant}/{timescale}.md`	Static guidance, prompts, agent selection
Insights (raw)	`Synthetic/Coaching/{Assistant}/insights.md`	Observations, clusters
Insights (crystallized)	`Semantic/Coaching/{Assistant}/insights.md`	Graduated, human-validated learnings

Learning flow

Each assistant phase captures observations → Synthetic/Coaching/{Assistant}/insights.md
Observations auto-cluster with confidence scoring (0-5)
Clusters graduate when confidence ≥ 4.5 AND stable for 3+ sessions
User crystallizes insight → Semantic/Coaching/{Assistant}/insights.md
Crystallized insights inform guidance updates

Creating Custom Rituals - Creating scheduled rituals
Templates - Template structure for rituals

​Skill architecture

​Skill file structure

​Writing style

​Creating a skill

​Assistant architecture

​Draft Mode

​Draft Mode Contract

​Placeholder Syntax

​Output Location

​Status Frontmatter

​Context Sources

​Example Conversion

​Assistant matrix

​Adding a new assistant

​Coach system

​Coach files

​Learning flow

​Related