jnMetaCode/shellward

v0.6.0 Security

This release includes 1 security fix for security teams reviewing exposed deployments.

Published 1mo MCP Security & Auth

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

This release patches 1 known CVE

Topics

agent-security ai-agent ai-firewall ai-safety ai-security claude-code

+14 more

cursor data-exfiltration dlp guardrails hermes-agent langchain llm-security mcp mcp-security openclaw pii-detection prompt-injection security shellward

ReleasePort's take

Moderate signal

editorial:auto 1mo

Release v0.6.0 lowers the default injectionThreshold from 60 to 40 after benchmark calibration and fixes two catastrophic ReDoS vulnerabilities causing self‑DoS.

Why it matters: The change reduces false positives by setting injectionThreshold to 40; critical because it also patches two severe ReDoS flaws that can cause service denial.

Summary

AI summary

Default injectionThreshold lowered from 60 to 40 based on benchmark calibration.

Changes in this release

Type	Severity	Summary	CVE
Security	Critical	Fixes two catastrophic ReDoS vulnerabilities causing self‑DoS Fixes two catastrophic ReDoS vulnerabilities causing self‑DoS Source: llm_adapter@2026-06-05 Confidence: high	—
Feature
Feature	Medium	Adds MCP tool-poisoning scanner with rug‑pull detection via /scan-mcp (stdio and remote HTTP) Adds MCP tool-poisoning scanner with rug‑pull detection via /scan-mcp (stdio and remote HTTP) Source: llm_adapter@2026-06-05 Confidence: high	—
Feature	Medium	Adds support for custom rules (`customRules`) to extend detection without forking Adds support for custom rules (`customRules`) to extend detection without forking Source: llm_adapter@2026-06-05 Confidence: high	—
Feature	Low	Adds detection benchmark command `npm run bench` with 100% precision/recall on labeled corpus Adds detection benchmark command `npm run bench` with 100% precision/recall on labeled corpus Source: llm_adapter@2026-06-05 Confidence: high	—
Bugfix	Medium	Reduces default injectionThreshold from 60 to 40 based on benchmark calibration (revertible via config) Reduces default injectionThreshold from 60 to 40 based on benchmark calibration (revertible via config) Source: llm_adapter@2026-06-05 Confidence: high	—
Bugfix	Low	Fixes PII false‑positive issues for carrier‑validated mobile numbers and UnionPay‑only bank cards Fixes PII false‑positive issues for carrier‑validated mobile numbers and UnionPay‑only bank cards Source: llm_adapter@2026-06-05 Confidence: high	—

Full changelog

Highlights

MCP tool-poisoning scanner + rug-pull detection + /scan-mcp (stdio + remote HTTP)
Custom rules (customRules) for extending detection without forking
Detection benchmark (npm run bench) — 100% precision/recall on labeled corpus, CI-gated
ReDoS audit — fixed 2 catastrophic-backtracking self-DoS vulns
Default injectionThreshold 60→40 (benchmark-calibrated; revertible via config)
PII false-positive fixes (carrier-validated mobile, UnionPay-only bank card)

See CHANGELOG. Published to npm: [email protected].

Note: npm provenance via CI activates once the repo NPM_TOKEN secret is set; 0.6.0 was published from the maintainer's authenticated local environment.

Security Fixes

Fixed two catastrophic backtracking ReDoS vulnerabilities causing self‑DoS

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track jnMetaCode/shellward

Get notified when new releases ship.

About jnMetaCode/shellward

AI Agent Security Middleware & MCP Server with 8-layer defense including prompt injection detection, DLP data flow tracking, command blocking, and PII detection. 7 MCP tools, zero dependencies.

All releases →