II.
Experiment overview
Reference · liveexperiment:codex-capability-gaps
Codex CLI Capability Gaps Review overview
Manual review of Codex CLI capabilities confirming documented gaps: no --thinking-level flag, no auto-memory extraction, runtime-only MCP support, and default native sandboxing.
Attributes
displayName
Codex CLI Capability Gaps Review
description
Manual review of Codex CLI capabilities confirming documented gaps:
no --thinking-level flag, no auto-memory extraction, runtime-only MCP
support, and default native sandboxing.
status
completed
result
validated
startedAt
2026-05-10
completedAt
2026-05-10
evidence
Reviewed atlas graph agent-version and agent-core-impl nodes for Codex.
Confirmed via graph data:
- No --thinking-level/--reasoning-effort flag in CLI help (codex#9381 closed)
- memory-system:codex-agents-md has autoExtraction=false, consolidation=none
- capability:supports-mcp on runtime impl only, not agentVersion
- subprocessSandboxStrategy=native-sandbox in runtime impl
All four claims remain accurate against current graph data and public
Codex CLI documentation.
Outgoing edges
tests_claim5
- claim:codex-cli-lacks-thinking-flag·TestableClaimCodex CLI lacks thinking-level flag
- claim:codex-lacks-auto-memory·TestableClaimCodex lacks auto memory extraction
- claim:codex-lacks-mcp-client·TestableClaimCodex lacks built-in MCP client
- claim:codex-sandbox-only·TestableClaimCodex defaults to sandbox-only execution
- claim:codex-has-sandbox·TestableClaimCodex has native sandboxing
Incoming edges
None.