agent-zero/plugins/_context_doctor
Alessandro 9bc33fe761 Let Responses text reach the core response dispatcher
Keep usable Responses output text intact in Context Doctor so core can
complete it through the normal response tool and policy path. Continue
repairing recognizable tool intent and retrying plain Chat Completions text.

Verified 100 focused tests and direct Docker framework dispatch. Live Codex
preset terminal tasks used native calls with 79% and 78% cache hits; the same
chat switched to Venice and completed its Chat Completions terminal task.
2026-09-06 18:57:45 +02:00
..
extensions/python Let Responses text reach the core response dispatcher 2026-09-06 18:57:45 +02:00
helpers Fix raw-text fallback classification 2026-09-05 15:18:13 +03:00
prompts Handle reasoning-only and plain-text fallback responses 2026-09-03 03:37:59 +03:00
tests Let Responses text reach the core response dispatcher 2026-09-06 18:57:45 +02:00
webui Add split-thoughts repair strategy to context doctor 2026-09-03 03:53:38 +03:00
AGENTS.md Let Responses text reach the core response dispatcher 2026-09-06 18:57:45 +02:00
default_config.yaml Add split-thoughts repair strategy to context doctor 2026-09-03 03:53:38 +03:00
hooks.py Install Context Doctor dependency after self-update 2026-08-25 23:40:31 +02:00
plugin.yaml Add Context Doctor plugin 2026-08-24 19:51:30 +03:00
README.md Add split-thoughts repair strategy to context doctor 2026-09-03 03:53:38 +03:00

Context Doctor

Repairs model tool call JSON before sending it to history and tool processing.

Behavior

  • Uses json_repair after native parsing has completed.
  • Accepts only complete Agent Zero tool calls (tool_name and object tool_args).
  • Stores repaired tool calls as compact JSON; log kvps retain streamed reasoning and add transformed fields.
  • Expands blank-line-separated thoughts into separate entries after repair when the split strategy is enabled.
  • Optionally replaces XML-like output with {} in cases where model uses native XML tool calls.