sashankh/dsh-taintguard0

dsh-taintguard

Indirect prompt-injection guard for DeepSeek Harness: taints tool output by origin and gates privileged tool calls that follow untrusted content

包名
dsh-taintguard
版本
0.1.0
许可证
MIT
最近更新
2026年8月16日

安装

$npx -p @deepseek-ai/dsh dsh plugin --profile web add github:sashankh/dsh-taintguard

Configuration

KeyDefaultMeaning
modeaskask routes a gated call to the operator, deny refuses it, observe only annotates.
untrustedSourcesweb_fetch, web_search, read, read_image, mcp__*Tools whose results carry externally controlled content.
privilegedSinksbash, pwsh, run_code, write, edit, str_replace_editor, cordis_*, schedule_create, subagent, send_messageTools gated once the agent is tainted.
egressSinksweb_fetch, bash, pwsh, run_codeTools whose arguments are scanned for credentials on every call.
canaries[]Marker strings that must never leave; any occurrence in an egress argument is refused.
mediumQuorum2Medium-confidence signals required to taint when no high-confidence signal matched.
maxScanChars20000Characters scanned per result and per argument set.
annotateUntrustedResultstrueWhether to re-frame a tainted result as quoted data.

Names are *-wildcard patterns over tool names. A pattern that matches nothing currently registered is legal, so mcp__* can be declared before any MCP server is configured.