Some of the skills in this repository are intentionally user-invoked and already set `disable-model-invocation: true` for Claude Code and `allow_implicit_invocation: false` for Codex. OpenCode, one of the widely used OSS AI harness, ignores those fields in v2 and advertises the skills to the model unless they include: ``` metadata: opencode/autoinvoke: "false" ``` OpenCode reference: https://opencode.ai/v2/docs/skills#frontmatter Would you be open to adding this to the existing user-invoked skills and documenting it in .agents/invocation.md? It preserves explicit slash invocation, requires no installer changes and `metadata` is part of the Agent Skills specification. **Standardization research** There is already an active proposal: [agentskills/agentskills#105](https://github.com/agentskills/agentskills/issues/105). It discusses standardizing invocation controls, including disable-model-invocation, user-invocable, or a neutral invocation: model | user | both enum. The additional parameter can later be replaced if [#105](https://github.com/agentskills/agentskills/issues/105) produces a portable standard.
