Keep agents under control
Guardrails, audits, and kill switches for AI agents running in production.
shortleash
Circuit breaker for AI agent runs. Budget, spawn, token, and timeout guards with per-run receipts. Now includes planok's human approval gate.
mcp-audit
npm audit for MCP servers. Scans for prompt injection, credential exfiltration, and misconfigurations. A–F security scoring.
mcp-jail
Runtime permission firewall for MCP servers. Policy enforcement and audit logging through a stdio proxy.
@decnutz/warden
Enterprise agent flight recorder. Runtime supervision for AI agent fleets: audit everything, live fleet view, one-tap kill switch.
Trust what ships
Deterministic checks that catch what reviews miss — before it reaches production.
pr-verify
Deterministic trust for agent-authored PRs. Blame provenance, blast radius analysis, and a 0–100 risk score on every pull request.
promptregress
pytest for prompts. Regression testing for LLM prompts, local-first — catch prompt drift before your users do.
adr-drift
Deterministic architecture tests from arch.yaml and ADR constraints. Catches architectural drift before it ships.
Move faster
Tools that remove the friction from everyday development work.
standupguy
Local-first standup notes from git ground truth. No more "what did I do yesterday."
sessionsteward
Versioned snapshots for agent working state. Undo what compaction dropped.
stuntkey
Your AI agent uses your API keys without seeing them. Keys stay in your computer’s keychain, the agent gets a stand-in, every key gets a spending cap, and a leaked key is fixed in one command. Formerly OathKey and SafeEasyAPI.
onememory-mcp
One shared memory layer for all your AI agents. Namespaces, scoped tokens, MCP-native.
@decnutz/contextrouter
Cross-repo context routing for AI agents. Answers "which repos" — not "where in this repo."
calas-gatewayComing soon
Multi-provider LLM gateway (Anthropic, OpenAI). Zero runtime dependencies, smart batching, cost tracking.