Slash Commands

Mega Brain extends Claude Code with 40+ slash commands for knowledge management, agent orchestration, and pipeline processing.

Quick Reference

Command	Alias	Category	Description
`/jarvis`	-	System	Activate JARVIS orchestrator
`/agents`	`/a`	Agents	Show agent status and memory
`/conclave`	-	Deliberation	Multi-agent council session
`/inbox`	`/i`	Pipeline	List pending files
`/process-jarvis`	-	Pipeline	Run 5-phase processing
`/extract-knowledge`	-	Knowledge	Extract insights from source
`/create-agent`	-	Agents	Create new cargo agent
`/save`	-	Session	Save current session
`/resume`	-	Session	Resume previous session

JARVIS (Orchestrator)

/jarvis

Activate the JARVIS meta-agent orchestrator. Syntax:

/jarvis [subcommand]

subcommand

string

Optional subcommand: status, resume, checkpoint, rollback, diagnose, log, decisions, suggest, pending, explain, pause, force, clear-pending

Behavior:

Loads state from /.claude/jarvis/STATE.json
Displays system status with visual timeline
Auto-resumes if process in progress
Provides context preservation and error recovery

Subcommands:

Show /jarvis status

Show detailed system state with metrics, pending items, and recent decisions.

Show /jarvis resume

Continue processing from last checkpoint.

Show /jarvis checkpoint

Create manual snapshot of current state to /.claude/jarvis/CHECKPOINTS/.

Show /jarvis rollback {CP-ID}

Restore from specific checkpoint ID.

Show /jarvis diagnose

Run complete system health analysis.

Files:

/.claude/jarvis/STATE.json - Current state (never edit manually)
/.claude/jarvis/CONTEXT-STACK.json - Context history (max 50)
/.claude/jarvis/DECISIONS-LOG.md - All autonomous decisions
/.claude/jarvis/PENDING.md - Items requiring resolution
/.claude/jarvis/SESSION-*.md - Per-session logs

Source: .claude/commands/jarvis.md:1-213

Agents

/agents

Show status of all agents and their memory freshness. Syntax:

/agents [--outdated | --knowledge "person" | --category CATEGORY]

--outdated

flag

List agents with memory not updated in 3+ days

--knowledge

string

Find agents with knowledge about specific person

--category

string

Filter by category: SALES, C-LEVEL, OPERATIONS, MARKETING

Output:

═══════════════════════════════════════
           AGENTS STATUS
═══════════════════════════════════════
🤖 AGENTES ATIVOS: 22

C-LEVEL:
  CRO      │ MEMORY: 2026-03-01 │ 12 pessoas
  CFO      │ MEMORY: 2026-02-28 │ 8 pessoas
  CMO      │ MEMORY: 2026-03-02 │ 15 pessoas
  COO      │ MEMORY: 2026-03-01 │ 10 pessoas

SALES:
  CLOSER   │ MEMORY: 2026-03-05 │ 25 pessoas
  BDR      │ MEMORY: 2026-03-04 │ 18 pessoas
  ...

Source: .claude/commands/agents.md:1-162

/create-agent

Create new cargo agent from detected role. Syntax:

/create-agent <role-name>

role-name

string

required

Name of the role (e.g., “ACCOUNT-EXECUTIVE”, “PRODUCT-MANAGER”)

Behavior:

Checks role mention threshold (10+ weighted mentions)
Generates agent files: AGENT.md, SOUL.md, MEMORY.md
Updates AGENT-INDEX.yaml
Assigns DNA from knowledge base

Source: .claude/commands/create-agent.md:1-12

Conclave (Multi-Agent Deliberation)

/conclave

Run full council session with debate, critic, devil’s advocate, and synthesizer. Syntax:

/conclave "<decision or question>"

decision

string

required

Strategic decision or question to deliberate

Example:

/conclave "Should we change closer commission from 10% to 15%?"
/conclave "Invest R$500k in Q1 team expansion?"

Phases:

Constitutional Foundation - Invoke core principles (Empiricism, Pareto, Inversion, Antifragility)
Debate - Multi-agent discussion between relevant roles (CRO, CFO, etc.)
Critic - Methodological quality assessment (score 0-100)
Devil’s Advocate - Challenge assumptions, identify risks
Synthesis - Unified recommendation with confidence score

Output:

═══════════════════════════════════════
DECISÃO RECOMENDADA: [recommendation]
CONFIANCA: 85%

RISCOS RESIDUAIS:
• Risk 1: Mitigation strategy
• Risk 2: Mitigation strategy

PRÓXIMOS PASSOS:
1. Action - Owner - Deadline
2. Action - Owner - Deadline

Anti-loop Rule: Council runs once per query. If confidence < 60%, escalates to human decision. Source: .claude/commands/conclave.md:1-197

/debate

Run debate between specific agents (without full council). Syntax:

/debate <agent1>,<agent2>[,<agent3>] "<topic>"

agents

string

required

Comma-separated agent IDs (e.g., “cro,cfo,cmo”)

topic

string

required

Topic or decision to debate

Example:

/debate cro,cfo "Reduce CAC by 30% in Q2"

Source: .claude/commands/debate.md:1-10

Pipeline Processing

/inbox

List files pending processing in inbox. Syntax:

/inbox [--all | --pending | --person "Name" | --organize [--execute]]

--all

flag

Show both processed and pending files

--pending

flag

Show only pending files (default)

--person

string

Filter by person name

--organize

flag

Auto-organize files into correct folders (dry-run)

--execute

flag

Execute auto-organization (requires —organize)

Auto-Organization:

Detects person from filename/content
Classifies content type (PODCASTS, COURSES, MASTERMINDS, etc.)
Standardizes naming (UPPERCASE, [youtube link])
Moves to {PERSON}/{TYPE}/ structure

Output:

═══════════════════════════════════════
           INBOX STATUS
═══════════════════════════════════════
📥 AGUARDANDO PROCESSAMENTO: 5

1. COLE-GORDON/PODCASTS/farm-system.txt
   ├─ Added: 2 days ago | ~5,400 words
   └─ Command: /process-jarvis "inbox/..."

Source: .claude/commands/inbox.md:1-297

/process-jarvis

Run 5-phase pipeline processing on source file. Syntax:

/process-jarvis "<file-path>"

file-path

string

required

Path to source file in inbox (e.g., “inbox/COLE-GORDON/PODCASTS/farm-system.txt”)

Phases:

Chunking - Split into semantic chunks (500-1500 words)
Entity Resolution - Normalize persons, themes, roles
Insight Extraction - Extract DNA layers, frameworks, heuristics
Narrative Synthesis - Create coherent narratives per person/theme
Cascade - Update agent memories, dossiers, and knowledge base

Checkpoints:

Each phase creates checkpoint in /.claude/jarvis/CHECKPOINTS/
Auto-resume from last checkpoint on interruption
Rollback support via /jarvis rollback

Output Artifacts:

/processing/chunks/{source-id}.json - Chunks with metadata
/processing/canonical/CANONICAL-MAP.json - Entity mappings
/processing/insights/INSIGHTS-STATE.json - Extracted insights
/processing/narratives/NARRATIVES-STATE.json - Synthesized narratives
/knowledge/dossiers/persons/{person}.md - Updated dossiers
/agents/{category}/{agent}/MEMORY.md - Updated agent memories

Source: .claude/commands/process-jarvis.md:1-144

/ingest

Ingest new material (URL, file, or folder) into inbox. Syntax:

/ingest <source> [--type TYPE]

source

string

required

URL, file path, or folder path to ingest

--type

string

Content type: podcast, course, mastermind, blueprint, vsl, script

Supported Sources:

YouTube URLs (auto-transcription)
Google Drive files/folders
Local files (.txt, .md, .pdf, .docx)
Audio files (.mp3, .wav, .m4a)

Behavior:

Downloads/copies to inbox
Extracts transcriptions (audio/video)
Detects person and content type
Registers in INBOX-REGISTRY.md

Source: .claude/commands/ingest.md:1-4

/extract-knowledge

Extract insights from source and populate knowledge base. Syntax:

/extract-knowledge "<file-path>" [--mode MODE]

file-path

string

required

Path to source file or “auto” to use NARRATIVES-STATE.json

--mode

string

Extraction mode: legacy (direct from file) or narratives (from pipeline output)

Modes:

/extract-knowledge "inbox/source.txt"
# Reads raw file directly
# Extracts insights without pipeline

Extraction Categories:

01-ESTRUTURA-TIME - People structure, roles
02-PROCESSO-VENDAS - Selling process, techniques
03-CONTRATACAO - Hiring, onboarding
04-COMISSIONAMENTO - Compensation, commissions
05-METRICAS - Performance metrics
06-FUNIL-APLICACAO - Lead qualification, pipeline
07-PRICING - Pricing strategy
08-FERRAMENTAS - Tools, tech stack
09-GESTAO - Management practices
10-CULTURA-GAMIFICACAO - Culture, motivation
99-SECUNDARIO - Other insights

Output Structure:

/knowledge/SOURCES/{SOURCE-NAME}/{CATEGORY}/{topic}.md

Source: .claude/commands/extract-knowledge.md:1-508

Knowledge Management

/dossiers

View or update person/theme dossiers. Syntax:

/dossiers [--person "Name" | --theme "topic" | --list]

Source: .claude/commands/dossiers.md:1-6

/view-dna

View DNA structure for person or agent. Syntax:

/view-dna <person-or-agent-name> [--layer LAYER]

name

string

required

Person or agent name

--layer

string

Specific DNA layer: L1 (PHILOSOPHIES), L2 (MENTAL-MODELS), L3 (HEURISTICS), L4 (FRAMEWORKS), L5 (METHODOLOGIES)

DNA Layers:

L1: PHILOSOPHIES - Core beliefs, worldview
L2: MENTAL-MODELS - Thinking frameworks
L3: HEURISTICS - Practical rules, shortcuts
L4: FRAMEWORKS - Structured methodologies
L5: METHODOLOGIES - Step-by-step implementations

Source: .claude/commands/view-dna.md:1-8

Session Management

/save

Save current session state. Syntax:

/save [session-name]

session-name

string

Custom session name. If omitted, uses timestamp.

Saved State:

JARVIS state and context stack
Processing phase and position
Pending decisions and items
Conversation history

Location: /.claude/jarvis/SESSION-{timestamp}.md Source: .claude/commands/save.md:1-9

/resume

Resume previous session. Syntax:

/resume [session-id]

session-id

string

Specific session ID to resume. If omitted, resumes most recent.

Source: .claude/commands/resume.md:1-5

System Commands

/system-digest

Generate system health and metrics digest. Syntax:

/system-digest [--full]

--full

flag

Include detailed breakdown per agent and knowledge domain

Metrics:

Total agents: minds (5), cargo (14), conclave (3)
Knowledge base: sources, dossiers, themes
Pipeline status: processed files, pending items
Agent memory freshness
Recent errors and warnings

Source: .claude/commands/system-digest.md:1-8

/verify

Run 6-level verification protocol. Syntax:

/verify [--level LEVEL] [--fix]

--level

string

Specific level: L1 (syntax), L2 (schema), L3 (references), L4 (consistency), L5 (completeness), L6 (semantic)

--fix

flag

Auto-fix detected issues where possible

Verification Levels:

L1: Syntax - File structure, YAML validity, JSON schemas
L2: Schema - Compliance with core schemas
L3: References - Foreign key integrity, broken links
L4: Consistency - Cross-file consistency, entity canonicalization
L5: Completeness - Required fields, minimum thresholds
L6: Semantic - Logical coherence, narrative quality

Source: .claude/commands/verify.md:1-9

Notes

All slash commands are defined in .claude/commands/ and auto-indexed by skill_router.py.

Commands that modify state (create-agent, process-jarvis) create automatic backups in /.claude/jarvis/CHECKPOINTS/.

Command Reference

Core Modules

Agents

Slash Commands

Slash Commands

Quick Reference

JARVIS (Orchestrator)

/jarvis

Agents

/agents

/create-agent

Conclave (Multi-Agent Deliberation)

/conclave

/debate

Pipeline Processing

/inbox

/process-jarvis

/ingest

/extract-knowledge

Knowledge Management

/dossiers

/view-dna

Session Management

/save

/resume

System Commands

/system-digest

/verify

Notes

Build docs developers (and LLMs) love

Command Reference

Core Modules

Agents

​Slash Commands

​Quick Reference

​JARVIS (Orchestrator)

​/jarvis

​Agents

​/agents

​/create-agent

​Conclave (Multi-Agent Deliberation)

​/conclave

​/debate

​Pipeline Processing

​/inbox

​/process-jarvis

​/ingest

​/extract-knowledge

​Knowledge Management

​/dossiers

​/view-dna

​Session Management

​/save

​/resume

​System Commands

​/system-digest

​/verify

​Notes

Build docs developers (and LLMs) love

Slash Commands

Quick Reference

JARVIS (Orchestrator)

/jarvis

Agents

/agents

/create-agent

Conclave (Multi-Agent Deliberation)

/conclave

/debate

Pipeline Processing

/inbox

/process-jarvis

/ingest

/extract-knowledge

Knowledge Management

/dossiers

/view-dna

Session Management

/save

/resume

System Commands

/system-digest

/verify

Notes