Ecosystem Update - 2026-07-04

July 4, 2026 · curated by Chad Simon · 12 items reviewed

Highlights

One safe harness Quick Win was implemented: removed 17 stale trusted project entries for deleted temp/scratch directories from ~/.codex/config.toml
Today's strongest ecosystem signal is context/harness linting: AGENTS/SKILL/hooks/MCP validators are maturing, but they should be audited before adoption
Dynamic workflows, broad skill catalogs, native memories, and automatic Codex upgrades remain useful signals but fail the automatic Quick Win safety gate for this setup

Stale trusted project cleanup config

developers.openai.com

Removed 17 deleted temp/scratch [projects] trust blocks from ~/.codex/config.toml; backup written to ~/.codex/backups/2026-07-04/

Agent context linter adapter skill/tooling

github.com/hesreallyhim/awesome-claude-code

awesome-claude-code now highlights AGENTS/SKILL/hooks/MCP linting tools such as agnix, Ctxlint, and agents-md-cookbook; build a read-only audit wrapper first, then run codex-skill-audit --strict before trusting any external package
approvals_reviewer profile normalization check config

official docs list approvals_reviewer as user | auto_review, while the inactive conservative-auto-review profile still uses guardian_subagent; needs a deterministic strict-config validation path before changing behavior
Existing-script secret-scan expansion hook

github.com/hesreallyhim/awesome-claude-code

Agent Guard reinforces real-time secret-leak guardrails; only build this by extending existing Codex-owned guard scripts or adding a reviewed script first, never by wiring an unbuilt hook
Codex MCP-server interop smoke mcp

github.com/shanraisshan/codex-cli-best-practice

codex-cli-best-practice documents Codex as an MCP server; useful for Wren/Claude interop, but it needs a smoke test and explicit boundary before any daemon or persistent route is added

No Tier 2 arXiv fetch this run. The last Tier 2 run was 2026-07-03T06:33:50-04:00; this run started at 2026-07-04T06:30:47-04:00, still inside the 24-hour skip window.

Items reviewed and explicitly declined this cycle, with the reason. Curation discipline matters more than coverage.

Wholesale community skill catalog import - external skills are not trusted without codex-skill-audit --strict, and this setup already has a large curated local skill set.
Auto-wire PreSkillUse/PostSkillUse or other unsupported hook events - no confirmed local support and no existing target script; violates the no-new-hook-without-script Quick Win limit.
Auto-enable native Codex memories - conflicts with the current deliberate features.memories = false posture and omni-mem default workflow.
Auto-upgrade Codex CLI daily - too high-risk for automatic mutation; current config already checks for updates on startup.
Add AGENTS.override.md or edit AGENTS policy as a Quick Win - conflicts with the byte-identical global/project AGENTS contract and the skill's hard policy-doc limit.
Enable plugin_hooks or global MCP parallel calls - adds broad behavior without server-by-server proof.