CLI and TUI

Status: published 0.3.0 stable release. Supported platforms: Windows and Linux. macOS is not yet officially supported.

Purpose

Provide a beginner end-to-end path through create, run, chat, check, eval, doctor, and templates, display explicit Loop progress, and use run --resume for paused or interrupted runs with a safe boundary.

Public interfaces

• coremind create
• coremind run --resume
• coremind chat
• coremind check
• coremind eval
• coremind doctor
• coremind templates

Errors and boundaries

• Stable run exit codes: succeeded 0, failed 1, paused 2, budget exhausted 3, timeout 124, and aborted 130
• --json-events keeps stdout as JSONL and always ends with run_result; diagnostics go to stderr
• --print and --json-events are mutually exclusive so prose cannot contaminate machine output
• Non-TTY approvals deny safely
• TUI and readline share the same ChatSession harness, and failed terminal states display their reason
• TUI approvals show effects, complete targets, and reasons first; long bodies are summarized and credential fields redacted
• TUI, readline, and JSONL expose the same ordered loop_state events; pause exits with code 2 and remains resumable
• Unsafe or already-finished run IDs fail resume explicitly
• TUI /status, /artifacts, and /context show recovery, evaluation, artifact, cache, and compaction status
• The final JSONL run_result.snapshot is the same pure-JSON snapshot used by the TypeScript and Python SDKs

CoreMind supplies mechanisms, quality guardrails, and development guidance. Users or business owners retain control of goals, rules, data fields, approval ownership, and final acceptance.

Source, tests, and examples

• packages/coremind-cli/src
• packages/coremind-cli/src/cli.e2e.test.ts
• packages/coremind-cli/src/approval.test.ts
• 模块示例
• Module example
• Agent Skill