dsh-advisor
Run the following command in DeepSeek Harness:
dsh plugin install slhssb/dsh-advisor
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install slhssb/dsh-advisor in DeepSeek Harness to install; the project is open-sourced at https://github.com/slhssb/dsh-advisor.
About this plugin
DeepSeek Harness ships without a built-in oversight mechanism, so a multi-step agent can silently execute destructive operations, violate schema contracts, or drift from the user's explicit request with no one watching. dsh-advisor fills that gap with a lightweight, always-on safety layer.
After every step that contains tool results, an independent reviewer model audits the agent's most recent operations. When it spots a genuine concern, a short [advisor] guidance message is spliced into the very next model call so the agent can self-correct on the fly; when the work is sound, nothing is injected and the only cost is a cheap reviewer call. Beyond the LLM pass, deterministic regex rules let you match tool names and raw arguments for zero-cost warnings or hard blocks, and a misconfigured rule can never block the main agent flow.
It is built for developers running tool-heavy, multi-step workflows in DeepSeek Harness, any scenario that demands a trust-but-verify posture. The plugin ships pre-configured for the DeepSeek official API with no override required, and the entire review pipeline degrades silently, ensuring your agent's forward progress is never held hostage by a reviewer hiccup.
Use Cases
- Catch destructive or irreversible tool calls before they cause damage in multi-step workflows
- Detect steps where the agent deviates from the user's explicit request or violates schema contracts
- Enforce zero-cost regex rules on specific tool parameters for hard blocks
Best For
- Developers running tool-heavy multi-step agent workflows
- Engineering teams adding an audit layer to production agents
- Product engineers building external-facing services on DeepSeek Harness
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.