Codenotary’s Apache-licensed tool watches for unfinished work, flags problems needing human attention, and helps developers spot wasted tokens.
One of the big problems with coding agents is their value rapidly diminishes when you start to figure in the added attention they require as their cost of doing business — not to count the large number of tokens they can use when left to their own devices.
You already know the drill. Although AI coding agents are performing increasingly complex development tasks — seemingly all on their own — developers often waste a lot of time supervising their execution, reviewing lengthy conversations, and repeating instructions. Worse: agents may report completion when tests have failed, verification can predate the latest code changes, or critical requested steps can remain unfinished on a “completed” job.
Codenotary, a cybersecurity company that specializes in software supply-chain security and AI-agent monitoring (and a FOSS Force Platinum Sponsor), has announced something that might help. On Monday it released AgentMon Live, software that’s designed to improve AI coding productivity. Specifically, it claims to help developers using Claude Code to verify completed work and identify tasks requiring attention, while reducing token waste as well as development time.
The best thing is that the software is free — as in speech and beer — licensed under the Apache 2.0 license. In other words, open source the way we like it, with source code you can inspect, modify (no “do not bend, staple, mutilate, or fold” notice here), extend, and even contribute upstream.
“AI coding productivity is measured by how much useful, verified work gets completed,” Dennis Zimmer, Codenotary’s chief technology officer and co-founder, said in a statement. “Developers should not have to spend their days supervising agents, checking whether they really finished, or guessing where their tokens went. We built AgentMon Live to address these problems, and we’re releasing it completely free and open source because we believe the entire agentic development community should benefit.”
AgentMon Live in a Nutshell
It’s one of those single pane of glass things, that promises much more. For the former, it provides a live dashboard monitoring agent activity, tool use, file changes, and verification results. It also watches your back, with its Done Gate feature that checks recorded session activity for evidence of unfinished work — like failed or outdated tests, incomplete tasks, and subagent failures — before the agent finishes.
For the latter, the “much more” part, it brings into play three core capabilities, says Codenotary:
- Verify completed work: The AgentMon “Done Gate” examines recorded session activity for unresolved failures and missing or outdated verification, prompting Claude Code to finish the work or clearly explain what remains. An optional second-model advisor can separately assess whether the agent’s response addresses the developer’s original request.
- Focus developer attention: The AgentMon Live real-time dashboard displays active tools, subagents, file modifications, pending approvals and verification results. Developers can identify stalled subagents, repeated failures, and other situations requiring intervention without continuously reading agent conversations.
- Identify wasted tokens and time: AgentMon Live provides visibility into token consumption, context growth, cache efficiency and tool execution time. It identifies potential inefficiencies, including repeated file reads, excessively large tool outputs and unused Model Context Protocol (MCP) servers. A built-in benchmarking command lets developers compare sessions with and without AgentMon Live.
On-the-Fly Testing & Availability
According to Codenotary, it’s already been seeing positive anecdotal results from an early evaluation conducted in the wild.
In one user’s initial two-day evaluation across nine projects, Codenotary says the Done Gate flagged potentially unfinished work and returned it to Claude Code 59 times. In 12 of those instances, the agent subsequently performed a substantial amount of additional work, which included completing an unfinished production deployment, rebuilding software after further source-code changes, and performing end-to-end verification of applications previously reported as complete although testing was not done.
AgentMon Live’s dashboard and verification checks based on recorded session activity run locally. The customizable interface supports six languages, and developers can choose whether to enable optional model-based advisory checks. It’s available now at no charge for Claude Code 2.1.308 and later. Source code, installation instructions and contribution information are available on the project’s GitHub page.
Developers can install AgentMon Live from a Claude Code terminal session using /plugin install agentmon-live --marketplace codenotary/agentmon-live.
Christine Hall has been a journalist since 1971. In 2001, she began writing a weekly consumer computer column and started covering Linux and FOSS in 2002 after making the switch to GNU/Linux. Follow her on Twitter: @BrideOfLinux
Be First to Comment