Ecosystem Update - 2026-05-08
Highlights
- OpenAI Codex 0.129.0 shipped on 2026-05-07; the useful deltas for this setup are
/hooks, broader hook lifecycle support, workspace-aware/diff, stronger plugin sharing controls, and TUI ergonomics - Safe Quick Wins were limited to config/agent metadata: a config schema hint plus display nicknames for the highest-use custom agents. Riskier items stay in Build Queue
Quick Wins (implemented today)
-
Config schema diagnostics Codex-mdAdd the official
config-schema.jsondirective to~/.codex/config.tomlfor editor diagnostics -
Subagent display nicknames agent-patternAdd
nickname_candidatesmetadata to high-use custom agents for clearer parallel-agent TUI labels
New Tools, Skills & Patterns
-
Codex 0.129.0 upgrade check Codex-mdhttps://github.com/openai/codex/releases/tag/rust-v0.129.0 - Current local CLI is
codex-cli 0.128.0; 0.129.0 includes/hooks, hooks before/after compaction, plugin sharing/source filtering, workspace-aware/diff, and TUI improvements. Worth a deliberate upgrade plus smoke test, not an automatic ecosystem edit -
Hook lifecycle review for 0.129.0 hookhttps://developers.openai.com/codex/hooks - Current
hooks.jsonusesSessionStart,Stop, andPreCompact. After upgrading, review whether existing scripts should be wired into supportedPreToolUse,PostToolUse,PermissionRequest, or post-compaction events. Do not wire anything until the target event and existing script contract are confirmed -
Permission profile audit Codex-mdhttps://developers.openai.com/codex/config-reference#configtoml - Official config now exposes named
default_permissionsand[permissions.<name>]. The current power-user posture is intentionallydanger-full-access; a useful next step is a non-defaultreview/ciprofile, not replacing the default -
Plugin sharing and marketplace hygiene pass mcphttps://github.com/openai/codex/releases/tag/rust-v0.129.0 - 0.129.0 improves workspace sharing, source filtering, marketplace removal/upgrades, and admin-disabled handling. Check installed marketplaces after the CLI upgrade
-
Codex native memories pilot skillPilot only if it can run with
no_memories_if_mcp_or_web_search = trueand without duplicating durable memory paths -
AgentLint-style compatibility audit skillhttps://github.com/rohitg00/awesome-claude-code-toolkit - The toolkit highlights AI-agent compatibility linting. This should become a small Codex-owned audit skill only if current
codex-runtime-doctorcannot cover the same checks
Research Worth Reading
-
A Self-Improving Coding Agent- Directly relevant to the
evolve/autoconfigloop: use benchmark-scored, reversible self-edits rather than broad runtime rewrites -
Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerging Frontiers
Considered, Not Adopting
Items reviewed and explicitly declined this cycle, with the reason. Curation discipline matters more than coverage.
- Install Claude toolkit wholesale — - Overengineered and Claude-owned; adapt discrete ideas into Codex-owned surfaces instead
- Enable native Codex memories immediately — enabling a second memory system without a pilot risks duplicate recall and privacy confusion
- Add new hook wiring that needs new scripts — - The skill explicitly forbids new-script hook Quick Wins. Existing scripts must be reviewed against current hook input/output contracts first
- Upgrade to 0.130.0 alpha releases — - Current 0.130.0 artifacts are prerelease alphas from 2026-05-08. Use stable 0.129.0 for the next upgrade check
-
Change constitutional policy docs — -
~/.codex/AGENTS.mdand~/AGENTS.mdare hard-limited out of Quick Wins - Deploy website or wrapper output — - Out of scope per user instruction; wrapper owns render/deploy after completion