Codex ignores Claude Code's disable-model-invocation flag — use allow_implicit_invocation instead
SeanOliver · x · 2026-09-25
SeanOliver shares a cross-tool gotcha: when sharing skills between Claude Code and Codex, disable-model-invocation: true only works in Claude Code — Codex ignores it and still offers the skill in every session. The Codex equivalent is allowimplicitinvocation: false under policy: in agents/openai.yaml, placed next to SKILL.md. If you want request-only skills across both tools, you need to maintain separate config for each.
More from coding & agent
- Hill Sampling: Condition on the Best Verified Solution, Generate 512 Parallel Edits, Repeat — LChoshen · 2026-09-25
- Nat Friedman fires back at Verge: those 'leaked secrets' are just files on the user's VM, intended behavior — dtrinh · 2026-09-25
- ServingStudio: simulate LLM serving configs before burning expensive GPU time — bariskasikci · 2026-09-25
- issue-graph prioritizes your backlog to coordinate coding agents, now on GitHub — Chamoen · 2026-09-25
- Claude now spots bugs, gathers logs, and asks users to submit packaged reports — ColleenMBrady · 2026-09-25
- SkillRL (NeurIPS 2026): 7B model beats GPT-4o by 41% via recursive skill evolution — cihangxie · 2026-09-25