ariffazil/AGI-claude-xml-structured-reasoning

AGI · claude-xml-structured-reasoning

Bind Claude's native XML-tag structuring into 333-AGI's reasoning chain. Doctrine: when output exceeds 1 screen, wrap in <analysis> / <plan> / <evidence> / <verdict> tags.

When to invoke

• Reasoning chain > 800 tokens
• Multi-step plan with > 3 dependencies
• Audit-grade output where downstream parsers need clear bounds
• Log/code interpretation needing anchored quotes

Native shape

<analysis>
  <premise>...</premise>
  <evidence src="...">...</evidence>
  <evidence_for>...</evidence_for>
  <evidence_against>...</evidence_against>
  <gap>...</gap>
</analysis>
<plan>
  <step n="1">...</step>
  <step n="2">...</step>
</plan>
<verdict>
  <decision>...</decision>
  <reversibility>...</reversibility>
  <floor_check>F1✓ F2✓ F4✓ F11✓ F13✓</floor_check>
</verdict>

Why XML on Claude

• Stable token boundaries (XML close tags = predictable parsing cuts)
• Extended-context recall with anchors (<evidence src="file:N">…</evidence>)
• Reduces pre-CoT noise in long outputs
• Interop with Anthropic's tool_use envelopes

Not instead of

This is a wrapper, not a replacement. Existing AGI skills (plan-dag, dream-engine, explorer-intelligence) keep their semantic content; this skill only governs output structure when invoked by a Claude-flavored agent.

DITEMPA BUKAN DIBERI.