Connectors

Connectors bridge framework-specific file layouts and the portable Egg format. Each supported platform has a spawner (reads source files into an Egg) and a hatcher (writes Egg contents back to files).

For how to implement a new connector, see Adding New Connectors.

Pipeline overview

Source files  →  Redaction  →  Spawner  →  Egg  →  Hatcher  →  Secret injection  →  Target files

• Spawner receives file contents after secrets and PII have been replaced with {{SECRET_NNN}} / {{PII_NNN}} placeholders. It extracts skills and memory records, each tagged with a label.

• Hatcher receives the full Egg and produces target files with placeholders still intact. The pipeline handles secret injection as the final step. Hatcher output is written to the agent/ subdirectory inside the output directory. Standard artifacts (AGENTS.md, agent-card.json, apm.yml, mcp.json) are written to the output root by the engine, not the connector.

Egg contents

Every Egg stores four types of content. The tables below show how each maps to files in each platform.

Content	Description
Memory[persona]	Who the agent is: identity, personality, tone
Memory[flow]	How the agent operates: system instructions, workflows
Memory[context]	What the agent knows about its environment: user info, tool descriptions
Memory[state]	Accumulated knowledge: episodic memory, notes, conversation history
Skills	Code, workflows, and capabilities
Secrets	Credential and PII placeholders with injection metadata
MCP	MCP server configurations

OpenClaw

Egg content	Spawner reads	Hatcher writes
Memory[persona]	SOUL.md, soul.md, IDENTITY.md	SOUL.md + IDENTITY.md (split by source)
Memory[flow]	AGENTS.md, agents.md, BOOT.md, HEARTBEAT.md	AGENTS.md
Memory[context]	USER.md, user.md, TOOLS.md	USER.md + TOOLS.md (split by source)
Memory[state]	knowledge.md, MEMORY.md, memory/*.md	MEMORY.md + memory/YYYY-MM-DD.md (dated records)
Skills	skill.md, skills.md, skills/*.md	skills/<name>.md (one file per skill, kebab-case)
Secrets	config.yaml, config.yml, config.json	config.json
MCP	mcp.json, mcp/*.json	mcp.json (Claude Desktop format)

Detection: persona file (SOUL.md, soul.md, IDENTITY.md) or skill files (skill.md, skills.md, skills/).

Letta

Letta stores memory in named blocks inside agent_state.json. The block names differ from Egg labels: Letta calls context memory human and flow memory system.

The hatcher produces a single agent.af file conforming to the AgentFileSchema, importable by any Letta server via letta.agents.import_file().

Egg content	Spawner reads	Hatcher writes
Memory[persona]	agent_state.json > memory.persona, *.af > blocks[label="persona"]	agent.af > blocks[label="persona"]
Memory[flow]	agent_state.json > system, system_prompt.md/.txt, *.af > agents[0].system	agent.af > agents[0].system
Memory[context]	agent_state.json > memory.human, *.af > blocks[label="human"]	agent.af > blocks[label="human"]
Memory[state]	archival_memory.json, archival/*.{txt,md,json}	archival_memory.json
Skills	agent_state.json > tools[], tools/*.py, *.af > tools[]	agent.af > tools[]
Secrets	config files	agent.af > agents[0].tool_exec_environment_variables
MCP	mcp.json, mcp/*.json	agent.af > mcp_servers[] (Claude Desktop format)

agent.db (SQLite) is also supported as a source. It contains skills, memory, and secrets in database tables.

Detection: .letta/ directory, agent_state.json, agent.db, *.af, or tools/*.py.

ZeroClaw

Egg content	Spawner reads	Hatcher writes
Memory[persona]	SOUL.md, persona.md, IDENTITY.md, identity.json	persona.md + identity.md (split by source)
Memory[flow]	AGENTS.md, instructions.md, system_prompt.md, HEARTBEAT.md	agents.md
Memory[context]	USER.md, user.md, context.md, TOOLS.md	user.md + tools.md (split by source)
Memory[state]	MEMORY.md, knowledge.md, memory/*.md, memory.db	knowledge.md + memory/YYYY-MM-DD.md (dated records)
Skills	tools/*.py, tools.json	tools/<name>.py
Secrets	config.json, config.yaml, config.toml	config.toml
MCP	mcp.json, mcp/*.json	mcp.json (Claude Desktop format)

The hatcher also produces a .zeroclaw/.keep marker directory so the workspace is recognized by the ZeroClaw spawner on re-ingestion.

Detection: .zeroclaw/ marker, persona file, tools//tools.json, AGENTS.md, or memory.db.