yfzhou0904/dsh-plugin-gallery--plugins-codex-moot-escalation ↗★ 0
@yfzhou/dsh-codex-moot-escalation
Removes moot sandbox-escalation fields from Codex tool schemas when a DSH session already has danger-full-access.
AI 분석
核心用途是解决 Codex 模型在完全访问权限下因提权字段导致 DSH 报错的问题。适合使用 Codex 且开启全权限的用户。
설치
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:yfzhou0904/dsh-plugin-gallery#0bca5d11a39014e69e5a571bf6d6ca97a491b2bb&path:plugins/codex-moot-escalation@yfzhou/dsh-codex-moot-escalation
A narrowly scoped host plugin for Codex-backed DeepSeek Harness sessions.
Some Codex models emit optional sandbox_permissions and justification arguments whenever those fields appear in a tool schema. In a session already running with danger-full-access, no wider sandbox exists, so DSH rejects such an escalation before running the tool. This plugin removes those two schema fields only when:
- the active provider is
codex; and - the resolved sandbox mode is
danger-full-access.
Other providers and sandbox modes retain DSH's native escalation behavior.
Install
dsh plugin --profile web add @yfzhou/dsh-codex-moot-escalation
The package supplies its own DSH bundle row; do not add a separate cordis.patch.yml entry.
Security scope
This plugin does not grant permissions or change the resolved sandbox mode. It only changes schemas shown to the model after DSH has already resolved unrestricted access.
Tested with DSH 0.1.0-rc.6. The plugin uses the system-prompt/assemble event and optional sandboxPolicy service, which may change in later DSH releases.
License
MIT