v0.40.0 - Actionable Error System (LLM-Agent-First)

🎯 Actionable Error System

Errors now include executable recovery commands that LLM agents can directly execute.

New Error Fields

Field	Description
error.fix	Primary recovery command (copy-paste ready)
error.alternatives	Array of {action, command} options
error.context	Structured error data for decision-making

Example Error Output

{
  "error": {
    "code": "E_DEPTH_EXCEEDED",
    "message": "Cannot add child to T550: max depth exceeded",
    "fix": "ct add 'My Task' --type task",
    "alternatives": [
      {"action": "Create as standalone", "command": "ct add 'My Task' --type task"},
      {"action": "Add to grandparent", "command": "ct add 'My Task' --parent T545"}
    ],
    "context": {"parentId": "T550", "parentType": "subtask", "parentDepth": 2}
  }
}

Compact JSON for Agents

Set CLEO_AGENT_MODE=1 for single-line JSON output (prevents "+N lines" truncation):

export CLEO_AGENT_MODE=1

Errors with Actionable Fixes

• E_DEPTH_EXCEEDED (exit 11)
• E_SIBLING_LIMIT (exit 12)
• E_PARENT_NOT_FOUND (exit 10)
• E_INVALID_PARENT_TYPE (exit 13)

Updated Agent Instructions

The CLAUDE.md injection now includes a CRITICAL: Error Handling section at the top, instructing agents to:

1. Check exit codes
2. Execute error.fix directly
3. Choose from error.alternatives based on intent

---

Full changelog: https://github.com/kryptobaseddev/claude-todo/blob/main/CHANGELOG.md