agent-forge

name	agent-forge
description	Use when CC asks to create a new agent, scaffold a new agent repo, or clone Bravo's architecture for a new domain (client agent, sibling agent, specialized agent). Generates a new AI-agent repo from templates, wires it into C_SUITE_ARCHITECTURE.md and brain/APP_REGISTRY.md, preserves the V5.6 outbound chokepoint, and gives the new agent a doctor command on day one.
triggers	["agent forge","use agent forge","run agent forge","use when cc asks to create a new agent"]
tier	specialized

Agent Forge

Bravo's moat is that it can create new agents with the same architecture (brain + memory + skills + scripts + safety rules) in minutes, not weeks. This is the command set that implements that.

When to Use

Trigger when CC says any of:

• "Create a new agent for..."
• "Clone Bravo for..."
• "Scaffold an agent that does..."
• "Make an agent for [client name]"
• "Build me an Atlas-style agent for X"

Do not use for: one-off scripts, temporary workers, Claude Code sub-agents that already fit agents/ or .claude/agents/.

Commands

# List everything Forge knows about
bravo agent list

# Create a new agent from template
bravo agent create <name> [--template <template>] [--role "<role>"] [--path <target-path>]

# Validate an existing agent's structure
bravo agent doctor <name>

Templates live in templates/agent-scaffold/. The default template produces the minimum viable agent: AGENTS.md, CLAUDE.md, brain/SOUL.md, brain/STATE.md, memory/SESSION_LOG.md, memory/ACTIVE_TASKS.md, scripts/core/self_audit.py, a doctor command, and a README.

Template files (read by bravo agent create):

• [[templates/agent-scaffold/README]] — generated agent's user-facing README
• [[templates/agent-scaffold/AGENTS|AGENTS]] — universal entry point for all AI clients
• [[templates/agent-scaffold/CLAUDE|CLAUDE]] — Claude Code instructions stub
• [[templates/agent-scaffold/brain/SOUL|brain/SOUL]] · [[templates/agent-scaffold/brain/STATE|brain/STATE]] · [[templates/agent-scaffold/brain/USER|brain/USER]]
• [[templates/agent-scaffold/memory/ACTIVE_TASKS|memory/ACTIVE_TASKS]] · [[templates/agent-scaffold/memory/SESSION_LOG|memory/SESSION_LOG]]
• [[templates/agent-scaffold/skills/INDEX|skills/INDEX]]

What a Forged Agent Gets on Day One

Layer	File(s)	Purpose
Identity	AGENTS.md, CLAUDE.md, brain/SOUL.md	Who the agent is, voice, values, prime directive
Memory	brain/STATE.md, memory/ACTIVE_TASKS.md, memory/SESSION_LOG.md	Live operational state + task backlog + session history
User context	brain/USER.md	Who they work for and that person's priorities
Safety	scripts/integrations/send_gateway.py (stub) + skills/security-protocol/	V5.6 outbound chokepoint; no bypass paths
Health	scripts/core/self_audit.py + scripts/doctor.py	Same 100-point audit Bravo uses
Skills	skills/INDEX.md	Registry of capabilities
Docs	README.md	One-page explanation of what this agent does

Templating Tokens

Template files support these tokens (replaced during bravo agent create):

• {{AGENT_NAME}} → literal name (e.g., Hermes)
• {{agent_name}} → lowercase slug (e.g., hermes)
• {{AGENT_ROLE}} → role description (e.g., client operations agent)
• {{AGENT_TEMPLATE}} → template slug used for this generation
• {{DATE}} → creation date (YYYY-MM-DD)

Conventions (NON-NEGOTIABLE)

1. Every forged agent preserves the V5.6 send_gateway chokepoint. Outbound email/DM/post/publish routes through scripts/integrations/send_gateway.py or an explicit approval gate.
2. Every forged agent has a doctor command on day one. It doesn't need to be rich; it just needs to be runnable so the agent can self-check.
3. Every forged agent has a clear Prime Directive. brain/SOUL.md must state what this agent exists to do.
4. Forged agents get registered in Bravo's world. The Forge updates brain/C_SUITE_ARCHITECTURE.md and brain/APP_REGISTRY.md so Bravo can route to them.
5. Forged agents never store credentials in their own repo. Read from a shared .env.agents or the parent operator's environment.

Default Template Slots (v1)

The default scaffold ships with placeholders for:

• Name + role
• Business context (who they work for, what they optimize)
• Safety gates (what requires CC approval)
• Capability list (what tools this agent can use)

Relationship to Existing Agents

• Atlas (CFO) — finance/tax/trading. Created manually; will be re-aligned to Agent Forge scaffold on its next refresh.
• Maven (CMO) — content/ads/brand. Same — planned migration.
• Aura (Life/Home) — ambient/habits. Same.
• Hermes (client-side commerce agent for Emmanuel Lowinger) — first agent intentionally forged with this template pattern.

Post-Creation Checklist

After bravo agent create <name> finishes:

1. cd into the forged repo.
2. Fill in .env.agents with only the keys this agent actually needs.
3. Run bravo agent doctor <name> (or python scripts/core/self_audit.py from inside the new repo).
4. Edit brain/SOUL.md to tune voice and Prime Directive.
5. Commit and push.
6. Add a line to brain/C_SUITE_ARCHITECTURE.md under the appropriate role slot.

Safety Notes

• Forge NEVER writes to .env.agents in either the source or target.
• Forge NEVER force-pushes or deletes existing files at the target path.
• Forge refuses to overwrite non-empty target directories.
• Forge logs every creation to memory/SESSION_LOG.md.

Related

• [[brain/C_SUITE_ARCHITECTURE]] — where new agents slot in
• [[brain/APP_REGISTRY]] — routing table for named agents
• [[brain/AGENTS]] — sub-agent roster
• [[skills/security-protocol/SKILL.md]] — safety defaults
• [[runtime/profile_home]] — profile substrate for multi-agent isolation