AgentMagnify

About the Project

AgentMagnify is the live project-visibility and status-tracking platform we built for developers, tech leads, and agencies who run AI coding agents. Your agents say "done" — in AgentMagnify, the evidence gets a vote: the panel keeps what the agent claims visibly separate from what the tests, builds, and screenshots actually support. A busy terminal proves nothing. In the moments you are not watching it, there is no answer to "what is the agent doing right now, and what state is my project actually in?" AgentMagnify solves this by tracking three numbers and refusing to blend them: Reported (what the agents claim), Roadmap (weighted arithmetic over the declared roadmap), and Verified (what the evidence supports). The distance between those three numbers is the most useful thing a monitoring tool can tell you. A "done" without evidence stays visibly a claim; an independent observer layer announces a contradiction as the headline of the project screen, never as a footnote: "The agent said done, the tests fail." Setup does not add another workflow: no CLI daemon, no IDE plugin. Any agent environment that can run bash can carry the skill — a native plugin in Claude Code, a single command elsewhere, a scoped token in CI and containers, a single-use pair code in cloud agent sessions. Reporting never blocks the real work: offline, events queue locally and deliver in order once the connection returns. What makes AgentMagnify stand out: • Reported / Roadmap / Verified: three progress numbers that are never blended • Claims come with receipts: screenshots and test reports attach to the claim they prove; e2e runs render as step-by-step visual stories • Live Board, Attention Queue, and Replay: every project on one screen, only what genuinely waits on a human in the attention queue, and the whole project history scrubbed back like a tape • Decisions route themselves: a sub-agent's question goes to its orchestrator; only questions that truly need a human reach you • Metadata, never code: events carry task names, counts, and commit hashes; source code, secrets, and terminal output never leave the machine • Self-host option: the whole platform runs from a docker compose file on your own infrastructure; the agent skill is MIT-licensed and open source Stop watching the terminal — trust the panel, and let the evidence speak.

Key Features