dsh-managed-approval
Run the following command in DeepSeek Harness:
dsh plugin install jalllychun/dsh-managed-approval
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install jalllychun/dsh-managed-approval in the DeepSeek Harness terminal to install; the source repository is https://github.com/jalllychun/dsh-managed-approval . After installation, restart the Web Host and select Approve for me from the permission menu in any conversation.
About this plugin
Working with DeepSeek Harness (DSH) day to day, the permission menu offers only three presets—Read Only, Workspace Write, and Full Access—and the constant approval prompts fragment focus. Existing auto-approve plugins either blanket-permit everything or limit themselves to Shell command allowlists. dsh-managed-approval adds a fourth session-level preset, "Approve for me," where a tool-free, one-shot model reviewer evaluates each pending operation against its risk and your trusted authorization context, auto-granting or denying it so approval fatigue drops to a minimum while DSH's sandbox stays in place as the safety floor.
The decision semantics are Codex-inspired: read-only search, list, and fetch-style MCP calls continue without an extra prompt, while side-effecting operations—create, update, delete, write, send, publish, upload—or unknown MCP actions are promoted into the approval path. The reviewer issues a one-time grant per operation, never a persistent permission. On denial it not only blocks the call but instructs the agent to find a materially safer alternative. If the reviewer times out, returns malformed output, or hits any other infrastructure failure, the request falls back to DSH's native human-approval prompt; three consecutive denials in a single turn trip a circuit breaker. You can point the reviewer at a separate model and provider via the profile patch, and layer glob-based toolRules for fine-grained allow/review/deny routing—though no rule can ever weaken an ask or deny returned by another DSH policy layer.
This plugin is for DSH users who want to cut approval interruptions dramatically without resorting to blanket auto-approve, especially in workflows heavy on MCP tool calls. It is not a full Codex clone—there is no exact-action /approve retry or rolling 10-of-50 rejection breaker yet—but within the hooks DSH rc.6 exposes it provides a conservative, configurable, risk-aware approval layer. If the session experience you want is "grant when safe, deny with a safer-path hint, hand off to a human on failure," dsh-managed-approval is one of the closest community plugins to Codex-style approval semantics on DSH today.
Use Cases
- Reducing approval interruptions during heavy MCP tool-call workflows
- Auto-granting low-risk operations while keeping the DSH sandbox intact
- Fine-grained allow, review, or deny routing via glob-based tool rules
Best For
- DSH users who want less approval fatigue without blanket auto-approve
- Developers whose daily workflow is heavy on MCP tool calls
- Users seeking Codex-style approval semantics with configurable risk routing
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.