reflect-appworld-failure

# Reflect on AppWorld Failure

Analyze failed AppWorld tasks to extract specific, actionable learnings that can be added to the playbook.

## Purpose

When an AppWorld task fails, the Reflector calls this Skill with error details and failed code. You analyze the failure semantically and generate a high-quality bullet with:
1. Specific title describing the pattern
2. Detailed content with working code examples
3. Relevant tags for retrieval
4. Appropriate confidence level

## Input Format

The input will be a text description with sections:

```
# Task
<task instruction>

## Apps
<comma-separated list of apps used>

## Error Type
<error_type: api_misuse, logic_error, timeout, etc.>

## Error Messages
<list of error messages from execution>

## Failed Code Snippet
<relevant code that failed>

## Missing Patterns (from heuristics)
<list of patterns the old system identified>

## Suggested Fixes (from heuristics)
<list of fix suggestions>
```

## Your Analysis Process

1. **Identify Root Cause**: What was the fundamental mistake?
   - Wrong API method name?
   - Missing authentication?
   - Incorrect data structure access?
   - Logic error?

2. **Extract Pattern**: What general pattern does this represent?
   - Is this specific to one app or applies to multiple?
   - Is this about API order (login first)?
   - Is this about method naming conventions?
   - Is this about data validation?

3. **Generate Concrete Example**: Create working code that demonstrates the CORRECT pattern

4. **Write Actionable Bullet**: Make it specific enough that the Generator can apply it

## Output Format

Return a JSON object with this structure:

```json
{
  "bullet": {
    "id": "bullet-YYYY-MM-DD-HHMMSS",
    "title": "<Specific pattern title>",
    "content": "<Detailed explanation with working code example>",
    "tags": ["app.<app_name>", "<error_category>", "<pattern_type>"],
    "evidence": [
      {
        "type": "execution",
        "ref": "<task_id>",
        "note": "<brief note