Skip to content
Confabulous

Codex

Confabulous has first-class support for OpenAI Codex sessions, including subagent spawns and skill invocations.

What gets parsed

Section titled “What gets parsed”
  • Full conversation history.
  • Per-message token counts (input, output, cached input, reasoning).
  • Model identifier (gpt-5, gpt-5.5, o3, etc.).
  • Tool calls.
  • Subagent spawns (spawn_agent / wait_agent) — bucketed by agent role.
  • Skill invocations (<skill> user-message wrappers) — bucketed by skill name.
  • Parent-child thread relationships (recursive tree of spawned subagents).

Analytics cards

Section titled “Analytics cards”
  • Tokens — including reasoning tokens (preserved for display; billed at output rate).
  • Cost — using the OpenAI pricing table.
  • Tools, Agents & Skills — Codex-specific breakdown.
  • Conversation — Codex synthesizes reasoning time into active time.
  • Repo activity.

Subagent aggregation

Section titled “Subagent aggregation”

When a Codex session spawns subagents, Confabulous aggregates the main thread plus every subagent thread for most analytics cards. The Conversation card stays main-only by design.

Pricing nuances

Section titled “Pricing nuances”
  • cached_input_tokens is a subset of input_tokens (not a separate count).
  • reasoning_output_tokens is a subset of output_tokens (billed at output rate).
  • OpenAI does not charge for cache writes.