Agent Signal

The screen is two columns and one matrix. The left column is what agents are choosing: posts where an agent picked your product, picked a named competitor instead, or published a workaround that names you. The right column is where they get stuck: documentation an agent could not follow, an API it used after it was deprecated, and questions nobody has answered. Below both sits the agent to agent matrix, where one agent wrote the change and another agent reviewed it. Every cell links to the original post or pull request.

Live today: agent-authored GitHub activity. The networks listed as planned on the screen are the roadmap, shown so you can see what is coming rather than guess.

• GitHub agent activity. Pull requests, reviews, review comments, issue comments and pushes in the repositories you already listen to, kept when the actor is a known agent or the text carries a phrase the tool writes into its own output.
• Moltbook. The public agent feed, read directly. Posts are kept when they name your product, a package of yours or a competitor.
• MoltOverflow, Mozilla cq, Clawstr. Read as public pages, behind the same relevance threshold as everything else.
• Stack Overflow for Agents. Ready to read as soon as an agent key is set for the network. Until then the row stays planned and says so.
• Trace corpora. Published agent traces arrive as dataset drops rather than a feed, so there is nothing to poll yet. When a corpus is loaded it is labeled as a sample, never as the market.

Each kept post is read by Jev, which answers three fixed questions: what the agent decided about your product, which Insights track the post belongs to, and whether the post states a concrete outcome rather than a passing mention. The six decisions are chose you, chose a competitor, docs failed, deprecated usage, fix shared, and no decision.

Every reading stores the confidence it came back with and the exact line that triggered it, so you can check the call against the post instead of taking it on faith. A competitor is only named when that competitor actually appears in the quoted line. Attribution is strict on purpose: an agent login on the roster or a signature phrase in the body, never a branch name or a hunch.

Nothing to configure beyond what you already did. Agent Signal reads the repositories in your connected GitHub sources and matches against your product name, your competitors, the package names and terms in your listening sources, and the hosts of your connected documentation. Open Agent Signal, choose a window, and run a pull. The screen shows how many posts were kept and how many Jev read.