vipin wiki

Codex Automation Model Policy

2 min read

Codex Automation Model Policy

Rule

Cron automations should run with:

  • model = "gpt-5.5"
  • reasoning_effort = "xhigh" when supported; otherwise use the strongest accepted reasoning level and record the fallback

EXTRACTED: The user explicitly rejected low- and medium-reasoning automation settings and asked future agents to treat high-intelligence automation as a hard rule. On 2026-06-04, this rule was tightened for Codex cron automations to prefer xhigh reasoning where the automation schema supports it.

INFERRED: This applies to scheduled cron automations where the automation schema exposes model and reasoning_effort. Heartbeat automations may not expose these fields; they should not be treated as noncompliant solely for lacking them.

Operating Guidance

  • When creating a new cron automation, set model to gpt-5.5 and reasoningEffort / reasoning_effort to xhigh when supported.
  • When updating an existing cron automation, preserve schedule, prompt, workspace, environment, and status unless the user asks to change them, but keep or restore gpt-5.5/xhigh where supported.
  • Do not use minimal, low, or medium reasoning for recurring scheduled work.
  • If a runtime rejects xhigh, keep gpt-5.5, use the strongest accepted reasoning level, and record the downgrade in the relevant automation note or wiki log.
  • If an automation appears cheaper or lighter than usual, still prefer xhigh/high reasoning unless the user explicitly gives a narrow one-off exception.

2026-05-18 Audit

EXTRACTED: On 2026-05-18, all local cron automations under C:\Users\admin\.codex\automations were audited. Seven cron automations were upgraded from low or medium reasoning to high, while cron automations already at gpt-5.5/high were left unchanged.

EXTRACTED: The final audit found zero noncompliant cron automations. One paused heartbeat automation had no model or reasoning fields and was considered not applicable to the cron-only model policy.

2026-06-04 WEIPING_WIKI Update

EXTRACTED: The active maintain-vipinknowledge cron automation under the D-drive Codex home was updated to keep the historical ID while using the current WEIPING_WIKI identity in the name/prompt and reasoning_effort = "xhigh".

INFERRED: Older audit entries that say high describe the state at that time. They do not override the current preference for xhigh where supported.

Counterpoints And Gaps

  • AMBIGUOUS: Heartbeat automations are controlled by the thread heartbeat mechanism and may not expose model or reasoning fields. This policy should not invent unsupported fields for heartbeats.
  • INFERRED: High reasoning costs more latency and compute, but the user’s stated preference is to preserve quality for recurring automation rather than optimize for cheap low- or medium-reasoning runs.

Graph View

Backlinks