v2.1.5 v2.9.4

high risk

msitarzewski/agency-agentsAug 12, 2026 → Aug 13, 20260 days between releases

What's New

Adds security hardening for MCP requests and prompt-injection detection.

Breaking Changes

  • HTTP MCP clients can no longer scan local filesystem paths or provide local YARA-rule directories.

New Features

  • Detects whitespace padding used to hide prompt-injection instructions.
  • Adds HIGH SC8 finding for skills shipping Python bytecode or __pycache__ content.
  • Adds the Skill Inspector companion skill guide.
  • Treats allowed-tools as valid least-privilege permission guidance in remediations.

Improvements

  • Harden HTTP MCP transport against local-path exploitation.
  • Improve minimum risk scoring for high-impact findings.
  • Refresh project documentation with OpenSSF Scorecard badge.

Bug Fixes

  • Reject symlinked skill content during discovery and disable Git symlink materialization.
  • Ensure high-impact findings receive an appropriate minimum risk score.

Upgrade Notes

HTTP MCP clients can no longer scan local filesystems. Update configurations to use stdio transport or move YARA rule directories to remote locations to maintain scanning capabilities.

The Two Releases

v2.1.5

Spice v2.1.5 (August 12, 2026)

Aug 12, 2026· 2 months ago

A patch release focused on dependable cached data and predictable operation under load. Cache invalidation is more reliable, Cayenne memory is better managed, and dashboards provide a complete view.

Full release notes
v2.9.4

SkillSpector v2.9.4

Aug 13, 2026· 2 months ago

Strengthens safe handling of MCP requests and untrusted skill content. Adds prompt-injection and supply-chain detection coverage.

Full release notes