team-attention/openclone/openclone-cli

openclone CLI helper

Use this skill when the user wants help running the standalone openclone CLI from a terminal rather than the Claude Code /openclone slash command.

Core workflow

1. Identify the user's target runtime:

• Local model/Ollama → read references/ollama.md.
• Codex OAuth → read references/codex-oauth.md.
• Claude Code subscription OAuth → read references/claude-code-oauth.md.
• API key or custom OpenAI-compatible endpoint → read references/openai-compatible.md.
• General install, first run, listing/resuming sessions → read references/quickstart.md.

1. Explain that the CLI uses the same markdown persona and knowledge/ files as source of truth.
2. For interactive chat behavior, memory, knowledge lookup, compaction, on-disk session persistence, and resuming prior conversations (--resume, --resume=<id>, openclone history, --no-persist), read references/conversation-and-knowledge.md.
3. If the user reports errors, read references/troubleshooting.md.

Keep answers short and concrete. Prefer copy-pasteable commands. Do not claim the CLI starts external services unless the reference says so.

Safety and expectation checks

• Ollama/local models run through an already-running Ollama server; openclone does not start it.
• Codex OAuth is opt-in and local-machine oriented; do not suggest hosted token sharing. Codex response item persistence defaults to off because the ChatGPT backend currently rejects store=true for ChatGPT-tier OAuth tokens.
• Model quality varies. Local models may follow persona/tool instructions less reliably than stronger hosted models.
• Conversation history is saved to ~/.openclone/conversations/<slug>/<sessionId>.json by default. Tell users they can inspect or delete those files directly, or pass --no-persist to skip writing them.
• The Claude Code skill and the CLI are separate entry points that share markdown clone data.