Summary

Fixes the claude-code-action failure by simplifying the JSON schema that was too complex.

Problem

The previous fix (#21240) used a JSON schema with allOf conditionals, if/then logic, and enum constraints. This complex schema likely caused the claude-code-action to fail during the "Analyze Issue Severity" step.

Error from run: https://github.com/coder/coder/actions/runs/20149538756

Solution

Simplified the JSON schema to match patterns used in claude-code-action's test workflows:

1. Removed complex conditionals

   • No more allOf, if/then, or const checks
   • Removed enum constraints for status and severity

2. Simplified to basic schema

   • Only require: status and reasoning
   • Keep severity and next_steps as optional fields
   • Validation of values happens in the downstream parsing step

3. Improved formatting

   • Changed claude_args from single-line to multi-line with |
   • Better readability and avoids potential YAML escaping issues

Schema Comparison

Before (complex):

{
  "allOf": [
    {
      "if": {"properties": {"status": {"const": "classified"}}},
      "then": {"required": ["severity"]}
    },
    ...
  ],
  "enum": ["classified", "insufficient_info"],
  ...
}

After (simple):

{
  "type": "object",
  "properties": {
    "status": {"type": "string"},
    "severity": {"type": "string"},
    "reasoning": {"type": "string"},
    "next_steps": {"type": "array", "items": {"type": "string"}}
  },
  "required": ["status", "reasoning"]
}

References

• Similar simple schemas work in claude-code-action tests

Supersedes

• fix: use structured output for classify-issue-severity workflow #21240 (schema was too complex)
• fix: correct jq parsing in classify-issue-severity workflow #21238 (attempted to parse execution file manually)

🤖 Generated with Claude Code