DeepSeek Harness PluginManifest Valid

dsh-deliberation

Primary-controlled deliberation and opt-in reasoning-masked review plugin bundle for DeepSeek Harness.

UI & ProductivityTerminal & TUIDeveloper ToolsSecurity & PolicyRemote Execution

0GitHub Stars0ForksUpdated2026-08-31

Install

No clear install command was found in the current repository evidence.

Plugin Overview & Capabilities

AI-assisted organization based on the public repository snapshot. The content must be grounded in source evidence and does not replace compatibility or security verification.

source-grounded

dsh-deliberation is an installable plugin bundle for DeepSeek Harness that provides bounded, Primary-controlled deliberation and opt-in reasoning-masked review. It adds a model-facing `deliberate` tool to launch bounded sub-agent branches (independent-alternative, trajectory-audit, or masked-review) as well as an optional `agent/turn-stopping` automatic masked review hook, returning structured JSON support packets without polluting Primary context.

Key Capabilities

Registers a `deliberate` tool for the Primary agent to spawn up to 3 bounded alternative, audit, or masked-review child sessions
Provides reasoning-masked review projections that re-evaluate observable actions without recorded reasoning history
Supports an opt-in automatic turn-stopping review hook with configurable publication modes (updates-only, all, observe-only)
Delivers compact, structured JSON result packets (observations, assumptions, errors, suggestions) rather than full transcripts
Configures subagent provider rows via Cordis patch definitions for mid-fork and deliberation tool handling

Useful For

Auditing and falsifying recent trajectories before continuing multi-step agent actions
Conducting blinded/masked reviews of tool-bearing turns to detect potential hallucinations or execution errors
Exploring alternative mechanisms or strategies through independent fresh sub-agent runs

Who It Fits

Developers building robust multi-step agent workflows on DeepSeek Harness
Researchers exploring test-time compute, reasoning verification, and trajectory auditing in LLM agents

Documented Limitations

Experimental preview status; paired benchmark data does not yet establish verified correctness or cost benefits
Subagents share the in-process workspace; tool filtering does not provide OS-level filesystem or network sandboxing
Cannot roll back file modifications or side effects caused by previous actions
Automatic review increases token and LLM invocation costs for each evaluated turn

DSH Compatibility

Version-specific runtime evidence collected by DSH Plugin. A missing result means we have not tested that combination yet.

Not tested yetNo runtime compatibility tests have been published yet.

Security Signals

Objective signals discovered from package metadata and source inspection. These are not a guarantee that a plugin is safe.

Dependency Counts

package.json declares 1 runtime, 13 development, 9 peer, and 0 optional dependencies.

info

Dsh Bundle Declared

package.json declares DSH bundle metadata.

info

License Declared

GitHub reports the repository license as MIT.

info

Package Manifest Available

A root package.json was captured and can be inspected by the registry.

info

Prepare Script Detected

package.json declares a prepare lifecycle script that may run during relevant package installation workflows.

info

Source Available

Public GitHub source metadata is available for this registry snapshot.

info

Source & Registry Notes

Public provenance, Registry classification, and the latest source check for this entry, kept separate from runtime verification.

Source repository: fly1989/dsh-deliberation
Registry source: GitHub · dsh-plugin topic
Registry classification: Plugin Bundle
Source checked: 17c70ff · 2026-08-31

This project is independently indexed from public source information. DSH Plugin is not affiliated with DeepSeek or the plugin author. Always check the author repository before installation.