dsh-sandbox-escalation-fix
Run the following command in DeepSeek Harness:
dsh plugin install JUSTMONIKA2022/dsh-sandbox-escalation-fix
Paste the following prompt into your AI chat to install this plugin:
Install the dsh-sandbox-escalation-fix plugin in DeepSeek Harness from the full source repository at https://github.com/JUSTMONIKA2022/dsh-sandbox-escalation-fix.
About this plugin
dsh-sandbox-escalation-fix is a zero-configuration compatibility plugin for DeepSeek Harness sessions where unavailable sandbox escalation fields are still exposed to the model. In All Access sessions or when the approval policy is set to never, models may continue to see and send parameters such as sandbox_permissions and justification, causing tools like bash, pwsh, write, and edit to fail before execution and potentially enter repeated retry loops.
Instead of merely cleaning up failed calls, the plugin projects session-accurate tool schemas based on the live Sandbox Mode and Approval Policy, hiding escalation options that cannot succeed. It also applies a narrow execution-time fallback for redundant same-mode requests and removes contradictory escalation hints from descriptions and results. Genuine escalation attempts, downgrade requests, invalid requests, and missing justifications are still handed to DSH validation, so the plugin does not bypass security checks, auto-approve requests, or grant extra permissions.
It is best suited for users on affected DSH releases who see third-party models failing with incorrect escalation parameters, blank justifications, or retry loops, and who want a fix without changing model settings, sandbox modes, or approval policies. Because newer DSH releases partially address the issue, it is especially appropriate after reproducing the same failure behavior in the built-in system.
Screenshots
Use Cases
- Models still send unavailable sandbox escalation parameters in All Access sessions, causing tool calls to fail.
- With approval=never, escalation requests are always rejected and models may enter retry loops.
- Tools such as bash, pwsh, write, and edit fail because of blank or invalid justification values.
Best For
- Administrators running third-party models in DSH who encounter tool-call failures.
- Developers who want to keep sandbox and approval policies intact without widening security boundaries.
- Users who need a zero-config fix for escalation parameters and retry loops.
Related Plugins
Adds an Auto approval mode on the official workspace-write sandbox, classifying semantic risks, asking when ambiguous, and denying destructive operations.
Undo/rollback system for DSH: config & plugin snapshots, one-click undo/redo/restore, message-level revert, secret masking, safe mode, and offline WebUI/GUI/CLI rescue tools for when DSH won't start.
Packages all 87 SKILL.md files from upstream reverse-skill as a DeepSeek Harness plugin that auto-registers them for authorized reverse engineering, penetration testing, and security research.
Dockyard DSH is a native DeepSeek Harness plugin that unifies official OAuth/client sessions for Codex, Antigravity, Grok, Claude, and Cursor, providing account pool, model catalog, and quota status.

