Codex Advisor

What it does

1. Finds the decision.Ordinary architecture and interfaces, or evidence-unresolved system boundaries, recovery, trust, and competing diagnoses.
2. Challenges it read-only.The advisor gets a bounded packet, no tools, and no write access.
3. Returns the call.Your main agent accepts, changes, or rejects the recommendation.

Choose your models

Advisor uses the live advisor.toml bundled with the installed plugin. Automatic Standard remains Terra at high effort and automatic Specialist remains Sol at high effort.

[standard]
model = "gpt-5.6-terra"
effort = "high"

[specialist]
model = "gpt-5.6-sol"
effort = "high"

# advisor-astra is a separate explicit-only role pinned to gpt-6-astra/high.

advisor-astra is reserved for explicit requests on the most complex uses. It is never selected automatically, uses more of your Codex allowance, and remains subject to account and runtime availability. Editing the bundled advisor.toml in place applies to the next consultation for Standard or Specialist. Plugin updates may replace edits to the bundled file.

Standard covers ordinary bounded architecture, interface, data-model, and generic-advisor decisions. Specialist applies only when targeted evidence still leaves a cross-system, compatibility/concurrency, competing-diagnosis, security/trust, recovery, irreversible-migration, or data-loss decision unresolved. Project importance, security adjacency, or an ordinary architecture question alone stays Standard.

Long consultations now drain the nonterminal shell handoff, require terminal exit, validate exactly one schema-v3 result envelope, and return that receipt to the enclosing call. This retains the 1.4.4 result-delivery repair without adding fallback.

Install from OpenAI

Codex Advisor is available in the official OpenAI Plugins Directory.

Install it there, then start a new Codex thread so Advisor is available to the session.

Compatibility

Consultations send bounded packets through your own authenticated Codex/OpenAI account; Zero Delta operates no relay, hosted backend, or intermediary service.

Resources