DashClaw
Approval gates and an audit ledger for coding agents left running unattended
Emerged finds new products worth a look, daily — what’s new today
CTRLRun
An open-source safety layer that checks every AI agent action before it runs
It sits between an agent deciding to act and the call that acts, so a refund of €5,000 typed instead of €500 waits for a human and can never execute twice.
CTRLRun is an Apache-2.0 Python library for execution safety. Every action is checked against YAML policy before it runs: allowed actions go through, sensitive ones wait for a person, forbidden ones are blocked. Approvals bind to the exact action so a mutated payload is refused, effects fire at most once across processes, and an unknown outcome is marked AMBIGUOUS rather than failed, which stops blind retries. It runs on SQLite or Postgres with a decorator, LangGraph and OpenAI Agents adapters, and an MCP gateway for agents you cannot modify. 652 commits, by Arpan Ghoshal and a friend.
More like this
Shield KYA
A local ledger of every tool call your coding agents make
no_human
Open-source agent that takes a ticket to a reviewed pull request, with proof
OpenBot
Self-hosted AI coworkers, each with its own computer and a policy gateway
Taster
A guard layer that checks agent tool calls and screens what tools send back
AgentRun
Open-source DSL that freezes repeated agent work into typed, inspectable steps