JayYuen666/dsh-danger-guard

A pre-execution guard that physically refuses dangerous shell commands and credential edits before dispatch, and hands the model the reason plus a way out.

The decision lives in the synchronous predicate on ctx.tools.guard — returning a string rejects the call, and the host turns that into the call's error result, so the model reads the refusal rather than being told about it later. Covered ground includes git hook bypasses, catastrophic rm targets, download-and-run, long-running dev servers, the git-destruction family and infrastructure-destroying commands. Each refusal carries a structured fix that names the offending flags or targets, and an ask-level hit runs one read-only blast-radius rehearsal whose output is attached to the approval request.

Tools & Capabilities ★ 0
View on GitHub ↗

Install

dsh plugin --profile web add @jayyuen66/dsh-danger-guard

The npm line from the README (@jayyuen66/dsh-danger-guard 0.1.7, MIT). Note the scope @jayyuen66 against the GitHub owner JayYuen666 — the registry repository field resolves back to this repo. The README documents a remove command for uninstalling and a separate 'verify it is live' section; a package-manager exit alone is not treated as proof it is enforcing.

Compatibility

Requires Node ≥ 22.12 and host dsh at the 0.2.1-rc.2 line per the README badges. Three defences: dangerous bash/pwsh command rules, a credential-path deny list matched by argument key shape rather than tool name (tool names drift with the provider), and a first-edit fact gate that refuses an edit whose target has not been read. Permission presets decide what each mode blocks; the README lists known boundaries and what it does not provide.

Details

Alternatives

tancheng33/dsh-egress-guard · Q1hangL/dsh-ask-guard · CaiZongyuan/dsh-ag-ui

More plugins in Tools & Capabilities

Browse more in Tools & Capabilities

Guides for Tools & Capabilities plugins