clawport-cli

Port any claw-ecosystem AI agent config to any other claw format.

OpenClaw → ZeroClaw → OpenFang → TitanClaw → and back again. Exact field mapping, automatic skill name normalization, and agent persona migration.

Install

npm install -g clawport

Usage

# Identify which claw clone a directory contains
clawport detect ./my-agent-dir

# List all supported clones
clawport list

# Dry-run: see what maps cleanly and what doesn't
clawport validate ./my-agent-dir

# Port two directories (auto-discovers config files and output path)
clawport port ./my-openclaw-dir ./my-zeroclaw-dir

# Export with explicit target clone and output file
clawport export ./my-openclaw-dir --to zeroclaw -o config.toml

# Export to openfang
clawport export ./my-zeroclaw-dir --to openfang -o ~/.openfang/config.toml

# Include persona files (MEMORY.md + agent.yaml/toml)
clawport export ./my-openclaw-dir --to openfang --persona-dir ./personas -o out/

# Export to canonical JSON (interchange / debug format)
clawport export ./my-agent-dir --to canonical -o config.json

# Write to stdout
clawport export ./my-agent-dir --to zeroclaw --stdout

Supported Clones

32 adapters across JSON, TOML, YAML, and ENV formats.

| Clone | Format | Schema Version | | ---------- | ------ | -------------- | | aionui | JSON | v1.4.2 | | andyclaw | JSON | v0.3.1 | | bashobot | ENV | v1.1.0 | | carapace | TOML | v0.2.1 | | copaw | JSON | v0.5.1 | | freeclaw | JSON | v0.2.3 | | grip-ai | JSON | v1.0.2 | | ironclaw | TOML | v0.12.0 | | kafclaw | JSON | v1.3.2 | | lightclaw | TOML | v0.6.0 | | memubot | JSON | v2.3.0 | | mimiclaw | ENV | v0.2.0 | | moltis | TOML | v0.10.6 | | n8nclaw | JSON | v0.5.0 | | nanobot | JSON | v1.2.0 | | nanoclaw | JSON | v0.4.1 | | nullclaw | JSON | v1.2.0 | | openclaw | YAML | a3f812c | | openfang | TOML | 0.3.1 | | opengork | ENV | v0.1.0 | | ouroboros | JSON | v0.3.0 | | picobot | JSON | v0.1.4 | | picoclaw | JSON | v0.2.0 | | rowboat | JSON | v0.8.0 | | ruvector | TOML | v0.4.0 | | safeclaw | JSON | v2.1.0 | | smallclaw | JSON | v0.3.2 | | thepopebot | JSON | v1.1.0 | | tinyclaw | JSON | v0.0.7 | | titanclaw | TOML | v1.0.3 | | zclaw | ENV | v0.4.0 | | zeroclaw | TOML | b91e44f |

Canonical Schema

All adapters map into a shared canonical format. Fields that cannot be mapped are placed in unmapped[] and emitted as comments in the target file — never silently dropped.

| Section | Fields | | ---------- | ---------------------------------------------------------------------------------------------------------------------- | | agent | name, model, provider, system_prompt, temperature, max_tokens, max_context, top_p, frequency_penalty, presence_penalty | | memory | backend, path, connection_string, embedding_model, vector_dims | | channels | type + platform-specific auth/addressing (bot_token_env, guild_id, chat_id, workspace, room_id, …) | | skills | name, enabled, config |

Skill Name Normalization

clawport normalizes skill names bidirectionally at adapter boundaries. Verb-first names (read_file, search_web) are normalized to canonical noun-first form (file_read, web_search) and denormalized back on write. Unknown skill names pass through unchanged.

Persona Migration

Pass --persona-dir to migrate MEMORY.md and agent.yaml/agent.toml alongside the config. The source adapter reads the persona directory; the target adapter writes persona files in its native format. Adapters without persona support skip this step silently.

Accuracy Model

clawport uses exact field mapping — every field in a source config is either:

• Mapped to a canonical field, or
• Explicitly flagged as unmapped with a reason and original value

There is no fuzzy matching. Fields unsupported by the target adapter are also flagged in the output so nothing is lost silently.

Architecture

src/
├── cli.ts              # CLI (detect, list, validate, export)
├── detect.ts           # fingerprint clone from directory
├── types.ts            # CanonicalConfig, Adapter interface
├── skills.ts           # bidirectional skill name alias map
├── persona.ts          # MEMORY.md + agent config migration
└── adapters/           # one file per clone, exact field mappings
    └── write-helpers.ts  # shared esc(), tomlVal(), unmappedCanonicalExtras()

Adding a New Adapter

1. Create src/adapters/<clone>.ts implementing the Adapter interface
2. Declare the exact config shape as a TypeScript interface, pinned to a commit
3. Implement parse() — map every source field to canonical or push to unmapped[]
4. Implement write() — emit supported canonical fields; call unmappedCanonicalExtras() for extension fields the format doesn't support
5. Register in src/adapters/index.ts

Every field must be explicitly handled or explicitly flagged. No silent drops.

Contributing

Pull requests for new adapters are welcome. Accuracy over approximation: every field mapping must be verified against the source clone's actual config schema, pinned to a specific commit.

If clawport saved you time: