thinkrun.ai
Record it. Your agent does the rest. Screen, voice, clicks, console, and network—captured in one pass. Your AI agent watches, understands, and proposes the fix. thinkrun.ai
👤 Joined on 20 July 2026
Followers
Following
Total Posts
@thinkrun.bsky.social@thinkrun.bsky.social → @thinkrun.aiCoding agents inherit their own earlier changes. Against developer patches, that history cut later-task success by 5.6 points, 14.0 when earlier tests also had to pass: the cost is inherited failing tests. https://arxiv.org/abs/2610.14768
Show PostIndeed..although often I find the MCP server leaves out some functionality and the agent has to use the API anyway…that’s what happened with #Nozbe when working with tags.
Show PostWe hit a similar premise failure in our AI-operated E2E: missing CSS exposed a hidden button, so a bad test passed. Asserting the starting state caught it. Four runs, same URL: github.com/aichance/bus...
Show Postdeterministic field mapping beats browser automation until you hit XFA forms, pdf-lib family just drops that layer silently on save. anything touching current IRS or USCIS PDFs needs to handle that explicitly or the fill looks fine until someone opens it in the wrong viewer
Show PostA proposed agents directory would hold skills and other resources for coding assistants, starting with one that finds the commit behind a bug: Linux May Get a New Home for AI Agent Skills buff.ly/46qHsVQ
Show Post