Codex operating model
Prompt blocks for planning, discovery, implementation, and verification. Each note links back to the OpenAI developer documentation that justifies the pattern.
/plan
Do not edit files yet.
Goal: verify whether README.md accurately describes the current project.
Use targeted repo inspection only. Identify:
- claimed architecture/setup/test commands in README
- files that prove or disprove each claim
- stale/missing README sections
- smallest verification commands to run
Return:
1. findings with file references
2. recommended README updates
3. commands you would run next
Stop before editing.Do not edit files yet.
I need to understand the issuance process and downstream verification impact.
Spawn read-only explorer agents:
1. issuance-flow: trace creation/issuance entrypoints and persistence
2. verification-flow: trace downstream verification consumers
3. data-contracts: inspect schemas/types/events/API contracts
4. tests-risk: find relevant tests and missing regression coverage
Wait for all agents. Consolidate into:
- end-to-end flow
- files/symbols involved
- invariants we must preserve
- implementation plan for feature X
- smallest test plan
No code changes yet.Implement only Milestone 1 from the approved plan.
Touch only the listed files unless you find a blocker.
Run the smallest relevant test suite.
Report:
1. files changed
2. behavior changed
3. tests run
4. remaining risks or follow-ups