Repository / project instructions
DocumentedAGENTS.md and AGENTS.override.md provide repository and project guidance in the documented Codex context assembly.
View source detailAgent intelligence
OpenAI Codex
Source-backed Codex CLI instruction intelligence separates the product, model context, limited prompt excerpt, and supporting evidence.
Codex CLI is documented as an agentic coding interface whose harness can supply workspace context, update plans, use terminal and patch tools, and receive repository guidance through AGENTS.md. The selected evidence does not establish the complete runtime instruction assembly or a persistent memory mechanism.
Scope: Codex CLI · GPT-5.2 · reviewed Aug 10, 2026
Reviewed product surfaces
Product Surface Schematic
Each layer links to the detailed evidence map. Its state applies only to the reviewed evidence.
Reviewed surface Codex CLI
- Base promptDocumented
- Dynamic harness contextDocumented
- Project instructionsDocumented
- Tools and approvalsDocumented
- MemoryNot established
- Documented
- Not established
Source-backed answers
Key behaviors
Each answer is scoped to the selected evidence. “Not established” means the evidence does not support the claim; it does not assert absence.
Planning
DocumentedThe public excerpt names making and updating plans as a communication capability.
View prompt evidenceTools and approvals
DocumentedThe excerpt lists terminal commands and patches, with configuration-dependent approval escalation.
View prompt evidenceDynamic workspace context
DocumentedThe excerpt says the harness can provide workspace files and other context.
View prompt evidenceMemory / persistent guidance
Not establishedThe limited excerpt does not establish a memory mechanism.
Review evidence scopeEditorial guidance
What this means in practice
These conservative editorial implications come from the linked evidence. They add no vendor claims or generic prompt advice.
Editorial implication
Put durable repository guidance in the documented instruction files.
AGENTS.md and AGENTS.override.md are the established repository/project guidance mechanism in this source set.
Supporting evidenceEditorial implication
Treat tool access and approvals as runtime-dependent.
Plan updates and terminal/patch calls are established, but actual tool availability and approval escalation depend on configuration.
Supporting evidenceEditorial implication
Do not assume persistent memory from the selected evidence.
The selected excerpt does not establish a separate durable memory mechanism for Codex.
Supporting evidenceUseful next actions
Continue your research
Compare agent behaviorsReview the same evidence-backed questions across all six products.
Inspect instruction evidenceA limited excerpt from the source is available for evidence-led reading; the complete runtime instruction assembly is not represented.
How does Codex compare with Claude Code?Compare prompt visibility, project context, tools, permissions, and memory through paired first-party evidence.
How can another CLI replace its base prompt?Inspect Gemini CLI's documented system-prompt override and export workflow.
Review source and provenance detailsVerify origin, retrieval, completeness, publication limits, and rights.
Product scope
Codex CLI, its model context, and its harness are distinct.
- Organization
- OpenAI
- Product
- Open-source agentic coding interface
- Surface
- Codex CLIA terminal coding-agent surface described in the pinned source file.
- Model context
- GPT-5.2This model name comes from the selected public prompt file and is not generalized to every Codex runtime.
Evidence status
Coverage and review
The review maps a limited excerpt from a pinned first-party repository file. It does not generalize that excerpt to every Codex configuration or runtime.
- Coverage
- partial
- Primary source
- Tier B · first-party open source
- Review freshness
- datedCoverage reaches the pinned repository revision selected here; the latest Codex product instruction state is not established.
- Reviewed
- Aug 10, 2026
Detailed instruction layers
Instruction-system map
The map distinguishes base instructions, dynamic harness context, project rules, tools, approvals, and unestablished memory behavior.
Dynamic harness contextdocumented
The excerpt says the harness can provide workspace files and other context.
EvidenceProject instructionsdocumented
AGENTS.md and AGENTS.override.md provide repository and project guidance in the documented Codex context assembly.
Source detailTools and approvalsdocumented
Terminal and patch calls are named; availability and approval flow are configuration-dependent.
EvidenceSupporting instruction evidence
Prompt and snapshot evidence
A limited excerpt from the source is available for evidence-led reading; the complete runtime instruction assembly is not represented.
Inspect instruction evidence- Selected snapshot
- gpt_5_2_prompt.md · limited public excerpt
- Evidence coverage
- Through Aug 10, 2026gpt_5_2_prompt.md · limited public excerpt
- Latest known vendor version
- Not established by selected evidence
- Current product state
- Not established by selected evidence
- Publication
- Limited excerpt and analysis
- Instruction page
- Available and indexablePinned first-party open-source excerpt with provenance and publication notes.
Evidence-backed capabilities
Capabilities with evidence
Each capability links to its supporting instruction section and keeps its stated confidence level.
| Capability | What the source establishes | Confidence | Evidence |
|---|---|---|---|
| Workspace context | The public excerpt says the harness can provide workspace files as context. | high | Section |
| Plan updates | The public excerpt names making and updating plans as a communication capability. | high | Section |
| Terminal and patch calls | The excerpt lists terminal commands and patches, with configuration-dependent approval escalation. | high | Section |
Version evidence
Snapshots and meaningful changes
The selected evidence pins one source revision. It does not establish a release timeline from repository history.
- selected gpt_5_2_prompt.md · limited public excerptReviewed Aug 10, 2026Instruction evidence
Provenance
Source ledger
Every source lists its origin, retrieval date, completeness, publication limits, and rights separately.
Tier B · first-party open source
OpenAI Codex prompt file
- Publisher
- OpenAI
- Retrieved
- Aug 10, 2026
- Upstream ref
- a603d7ca5c0fc5f965abfba09366ae5fb091350e
- Rights posture
- Limited excerpt from Apache-2.0 first-party repository source.
The public file contains more content than this excerpt package; runtime assembly and configuration-dependent additions are not established.
Open primary source Tier A · vendor-published
Unrolling the Codex agent loop
- Publisher
- OpenAI
- Retrieved
- Aug 11, 2026
- Upstream ref
- OpenAI technical article on Codex context assembly
- Rights posture
- Original summary of official OpenAI documentation; no instruction-file contents copied.
Establishes AGENTS.md and AGENTS.override.md aggregation from the Git/project root to the working directory, subject to the documented size limit; it does not describe every runtime instruction input.
Open primary source