[BUG] API bug: thinking options type cannot be disabled when reasoning_effort is set on all 4 models #1192

First time using it. I installed at claude code and used Deepseek to creating process. the message said:

God Nodes (Top 10)

All wlrestart.sh scripts + print_header() functions — 9 edges each. Heavily duplicated across ol7/ol9/reciclar-OL8 copies.

Surprising Connections

None — AST-only extraction (semantic subagents blocked by API-level reasoning_effort/thinking conflict). All edges are structural (file→symbol calls).

Key Findings

• 190 communities detected, dominated by near-duplicate clusters (same scripts in ol7/ol9/reciclar-OL8/___playbook-ol7-default-weblogic). Top communities are
  setDomainEnv.sh API vars, setUserOverrides.sh env, and deploy_files_manager_dynamic.sh — each duplicated ~5x.
• 269 isolated nodes (≤1 connection) — mostly API variable references and config keys that only appear once in a single file.
• Low cohesion (0.07-0.09) in top communities — artifacts of file duplication across directories, not genuinely weak code.
• wlrestart.sh is the most-connected artifact, appearing in 5+ locations across the codebase.

Semantic Extraction — Skipped

API bug: thinking options type cannot be disabled when reasoning_effort is set on all 4 models (deepseek, sonnet, opus, haiku) when dispatching subagents. 551 doc
files not processed. The graph is purely AST-structural (100% EXTRACTED, 0% INFERRED).

Re-run with GEMINI_API_KEY set to avoid Claude subagent dispatch, or when the API bug is resolved.

Graph ready for querying: /graphify query "How does wls-update connect to create-domain-offline?"