dsh-autogate
A DeepSeek Harness permission-approval plugin that layers deterministic rules and optional LLM review over the workspace-write sandbox.
Install
$ dsh plugin --profile web add github:wangxing-git/dsh-autogatePlugin 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.
Key Capabilities
- Adds auto-ask and auto approval presets with workspace-write sandbox configuration.
- Applies deterministic allow/deny rules for defined tools, shell operations, paths, and high-risk actions.
- Can send ambiguous or semantically risky operations to an LLM classifier when preflight is enabled.
- Uses fail-closed handling for classifier errors, timeouts, missing routing, and malformed output.
- Reviews sandbox-escalation retry requests and can route outcomes to LLM decision or human approval depending on mode.
- Provides a Web UI settings card and approval-trail overlay with recent decision details.
- Supports configuration through the autogate section of $DSH_HOME/settings.yaml.
Useful For
- Reducing repetitive approval prompts in DeepSeek Harness Auto sessions.
- Using a semi-automatic approval flow that can fall back to human escalation prompts.
- Applying LLM review to ambiguous shell, network, database, Git-state, sensitive-path, or dynamic-target operations.
- Reviewing recent approval decisions through the Web approval-trail panel.
Who It Fits
- DeepSeek Harness users working with workspace-write sandbox sessions.
- Developers who want configurable approval automation while retaining DSH sandbox enforcement.
- Teams that need visibility into automated approval decisions.
Documented Limitations
- It is a decision layer rather than a security boundary; execution boundaries remain DSH sandboxing and escalation approval.
- The auto mode makes LLM decisions final and does not show a human approval prompt after denial or classifier failure.
- Preflight is disabled by default, so ordinary tool calls rely on the workspace-write sandbox unless enabled.
- LLM classification can misclassify operations; prompt-injection defenses are described as soft defenses.
- Static path checks retain a TOCTOU window if a symlink is repointed after checking and before writing.
- Credential-exfiltration detection uses shallow text patterns and may not detect encoded or split credentials.
- Approval trails are process-memory only and are cleared when DSH restarts.
DSH Compatibility
Version-specific runtime evidence collected by DSH Plugin. A missing result means we have not tested that combination yet.
Security Signals
Objective signals discovered from package metadata and source inspection. These are not a guarantee that a plugin is safe.
Source & Registry Notes
Public provenance, Registry classification, and the latest source check for this entry, kept separate from runtime verification.
- Source repository
- wangxing-git/dsh-autogate
- Registry source
- GitHub · dsh-plugin topic
- Registry classification
- Plugin
- Source checked
- 82e1c8f · 2026-09-17
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.
Repository Activity
- GitHub stars
- 3GitHub stars
- Forks
- 0
- Open issues
- 0
- Last commit
- 2026-09-03
- Last release
- No release detected
Related DSH Plugins
Ranked by overlapping capabilities, use cases, plugin type, categories, and DSH profile.
Second-model AI auto-review for DeepSeek Harness approval requests with fail-closed safety and session audit.
View plugin →Persistent multi-model Agent teams and bounded DAG workflows with GUI controls for DeepSeek Harness.
View plugin →Native SQLite-backed local project taskboard plugin and UI for DeepSeek Harness with Agent claim and review flows.
View plugin →Sandbox-first automatic permission policy and intelligent risk classification for DeepSeek Harness.
View plugin →