| name | core-ai-cli-manual |
| description | Complete operation manual for core-ai-cli covering all configuration, features, and operational modes. Use when the user asks about configuring core-ai-cli, understanding available features, reading or writing agent.properties, setting up LLM providers, memory system configuration, hooks, MCP servers, CLI commands, ACP mode, serve mode, remote agents, or any aspect of core-ai-cli operation. This skill is for both users and LLMs — use it to look up exact property names, defaults, and configuration patterns before making changes. |
core-ai-cli Operation Manual
Comprehensive reference for configuring and operating core-ai-cli.
Configuration Files
| File | Location | Purpose |
|---|
agent.properties | ~/.core-ai/agent.properties | Main configuration (global) |
agent.properties | <workspace>/.core-ai/agent.properties | Per-project overrides (merges over global) |
instructions.md | <workspace>/.core-ai/instructions.md | Project instructions injected into system prompt |
MCP.json | <workspace>/.core-ai/MCP.json | Per-project MCP server overrides |
hooks.json | <workspace>/.core-ai/hooks.json | Per-project hook scripts |
| Plugin hooks | ~/.core-ai/plugins/<name>/hooks/hooks.json | Plugin-provided hooks |
Configuration loading chain: workspace-local properties override global properties. For hooks, workspace hooks.json has highest priority, then global plugins, then local plugins.
Quick Start: Enabling Features
# Minimal: enable memory
agent.memory.enabled=true
agent.memory.daily.logs.enabled=true
# Enable coding mode
agent.coding.enabled=true
# Enable todo v2
agent.todo.v2.enabled=true
Document Index
For detailed reference, read the relevant file:
| Topic | File |
|---|
| All agent.properties keys, defaults, and descriptions | references/agent-properties.md |
| CLI modes (interactive, headless, serve, ACP), flags, and commands | references/cli-modes.md |
| hooks.json format, events, environment variables | references/hooks.md |
| MCP server configuration | references/mcp.md |
| Memory system architecture, properties, and modes | references/memory.md |
| LLM provider configuration patterns (LiteLLM, OpenAI, DeepSeek, etc.) | references/providers.md |
Feature Gates (Quick Reference)
| Property | Interactive | Headless | Serve | ACP |
|---|
agent.memory.enabled | true | true | true | true |
agent.memory.daily.logs.enabled | false | false | false | false |
agent.memory.prompt.extraction | false | n/a | n/a | n/a |
agent.coding.enabled | false | false | false | false |
agent.todo.v2.enabled | false | false | false | n/a |
All boolean feature flags default to false unless noted. agent.memory.enabled defaults to true. Interactive mode = core-ai, Headless = core-ai --prompt, Serve = core-ai --serve, ACP = core-ai --acp-agent.
Writing Configuration
When instructed to add or change a property, edit the appropriate file:
- Global defaults:
~/.core-ai/agent.properties
- Project-specific overrides:
<workspace>/.core-ai/agent.properties
Always use edit_file to modify existing files. Only use write_file when creating a new file.
Property format: key=value (no spaces around =). Comments start with #.