Skip to content

Add aperion-shield under Agent Runtime Security & Sandboxing#1

Open
scottbitbot2502 wants to merge 1 commit into
mainfrom
add-aperion-shield
Open

Add aperion-shield under Agent Runtime Security & Sandboxing#1
scottbitbot2502 wants to merge 1 commit into
mainfrom
add-aperion-shield

Conversation

@scottbitbot2502

@scottbitbot2502 scottbitbot2502 commented May 16, 2026

Copy link
Copy Markdown
Member

Adds aperion-shield to the Agent Runtime Security & Sandboxing section.
What it is. Apache-2.0 Rust binary that wraps any MCP server and blocks
destructive agent operations (DROP DATABASE, rm -rf, force-push, credential
exfiltration, dangerous SQL UPDATEs) before they reach the upstream tool.
Released v0.5.0 on 2026-05-09.
Why this section. Neighbours in the list — claude-code-safety-net,
leash, vibekit, pipelock — wrap a single coding agent or sandbox a
single process. Shield generalises the same idea at the MCP transport
layer
, so any MCP-speaking agent (Cursor, Claude Desktop, custom) is
governed via stdio without an SDK integration into the agent.
Defense properties.

  • Adaptive composite scoring across five signals (rule severity, workspace
    context probe, decision memory, burst detector, composite score)
  • Validated 98.4% pass-through against 13,000 real Cursor agent commands —
    reproducible via the bundled tests/corpus/ and aperion-shield --check
  • Pre-execution blocking (denies/prompts before tool dispatch)
  • Identity-gated tool calls via Ed25519 + OIDC for high-severity ops (v0.5+)
  • Tamper-evident audit chain (SHA-256 hash chain; no signing yet — that's
    on the v0.7 roadmap and we acknowledge asqav / AgentMint are ahead
    there)
    Repo: https://github.com/AperionAI/shield
    Docs: https://docs.aperion.ai/aperion-shield.html
    License: Apache-2.0
    Maintainer: AperionAI (active — v0.6 ships 2026-05-18)

Apache-2.0 Rust MCP gateway, v0.5.0 released 2026-05-09.
Fits neighbor entries (claude-code-safety-net, leash, vibekit) but
operates at the MCP transport layer instead of wrapping a single
coding agent. Validated false-positive rate published in the README
and reproducible via the bundled corpus.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant