[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-vercel-labs-ai-python-durable-execution":3,"mdc--yzx58f-key":33,"related-org-vercel-labs-ai-python-durable-execution":634,"related-repo-vercel-labs-ai-python-durable-execution":806},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":22,"repoUrl":23,"updatedAt":24,"license":25,"forks":26,"topics":27,"repo":28,"sourceUrl":31,"mdContent":32},"ai-python-durable-execution","add durable execution to AI agents","Use when adding durable execution to AI SDK for Python, building durable agent loops, or serializing messages across workflow steps.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},"vercel-labs","Vercel Labs","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fvercel-labs.png",[12,16,19],{"name":13,"slug":14,"type":15},"Python","python","tag",{"name":17,"slug":18,"type":15},"Agents","agents",{"name":20,"slug":21,"type":15},"AI SDK","ai-sdk",92,"https:\u002F\u002Fgithub.com\u002Fvercel-labs\u002Fai-python","2026-07-29T05:40:42.780847",null,11,[],{"repoUrl":23,"stars":22,"forks":26,"topics":29,"description":30},[],"AI SDK for Python","https:\u002F\u002Fgithub.com\u002Fvercel-labs\u002Fai-python\u002Ftree\u002FHEAD\u002Fskills\u002Fai-python-durable-execution","---\nname: ai-python-durable-execution\ndescription: Use when adding durable execution to AI SDK for Python, building durable agent loops, or serializing messages across workflow steps.\nmetadata:\n  sdk-version: \"0.4.0\"\n---\n\n# ai-python-durable-execution\n\nUse durable execution when an agent run must survive restarts, worker moves, or\nlong waits.\n\nThe SDK does not provide durability by itself. Build a custom `Agent.loop`, and\nput side effects inside durable steps:\n\n- model calls\n- tool I\u002FO\n- approval or resume boundaries\n\nKeep the workflow replayable. Durable steps should take JSON inputs and return\nJSON outputs.\n\nSerialize messages like this:\n\n```python\ndata = message.model_dump(mode=\"json\")\nmessage = ai.messages.Message.model_validate(data)\n```\n\n## Model Step\n\nA durable model step should drain `ai.stream(...)` inside the step and return one\ncomplete assistant `Message`.\n\n```python\n@workflow.step\nasync def llm_step(\n    model_data: dict[str, object],\n    messages_data: list[dict[str, object]],\n    tools_data: list[dict[str, object]],\n) -> dict[str, object]:\n    model = ai.Model.model_validate(model_data)\n    messages = [\n        ai.messages.Message.model_validate(message)\n        for message in messages_data\n    ]\n    tools = [ai.Tool.model_validate(tool) for tool in tools_data]\n\n    async with ai.stream(model, messages, tools=tools) as stream:\n        async for _event in stream:\n            pass\n\n        if stream.message is None:\n            raise RuntimeError(\"LLM stream ended without a message\")\n\n        return stream.message.model_dump(mode=\"json\")\n```\n\n## Durable Tools\n\nPrefer wrapping the tool body in the durable step:\n\n```python\n@ai.tool\n@workflow.step\nasync def ask_mothership(question: str) -> str:\n    response = await mothership_client.ask(question)\n    return response.summary\n```\n\nIf the workflow system needs separate activity dispatch, schedule a zero-arg\ncallable that returns `ai.tool_result(...)`. Do not call `tool.fn` directly.\n\n## Agent Loop\n\nUse the model step result as a complete message. Do not wrap it in `ai.Stream`,\n`ai.events.replay_message_events`, or `ai.util.merge`, those utilities are\nused for fluent dispatch in non-durable applications, which is impossible\nin a workflow setting since streams are considered side-effects.\n\n```python\nclass DurableAgent(ai.Agent):\n    async def loop(self, context: ai.Context):\n        while context.keep_running():\n            result = await llm_step(\n                context.model.model_dump(mode=\"json\"),\n                [m.model_dump(mode=\"json\") for m in context.messages],\n                [t.model_dump(mode=\"json\") for t in context.tools],\n            )\n\n            assistant_message = ai.messages.Message.model_validate(result)\n            context.add(assistant_message)\n\n            async with ai.ToolRunner() as runner:\n                for tool_call in assistant_message.tool_calls:\n                    runner.schedule(context.resolve(tool_call))\n\n                async for event in runner.events():\n                    yield event\n\n                context.add(runner.get_tool_message())\n```\n\nThis pattern does not stream model tokens to the caller. That is usually the\nright tradeoff for durable workflows, because many durable systems do not support\nasync generators. You can build a queue-based side channel for streaming; however,\nthat kind of stream can't be used to dispatch tools and affect control flow directly.\n",{"data":34,"body":37},{"name":4,"description":6,"metadata":35},{"sdk-version":36},"0.4.0",{"type":38,"children":39},"root",[40,47,53,67,87,92,97,127,134,155,347,353,358,404,425,431,460,623,628],{"type":41,"tag":42,"props":43,"children":44},"element","h1",{"id":4},[45],{"type":46,"value":4},"text",{"type":41,"tag":48,"props":49,"children":50},"p",{},[51],{"type":46,"value":52},"Use durable execution when an agent run must survive restarts, worker moves, or\nlong waits.",{"type":41,"tag":48,"props":54,"children":55},{},[56,58,65],{"type":46,"value":57},"The SDK does not provide durability by itself. Build a custom ",{"type":41,"tag":59,"props":60,"children":62},"code",{"className":61},[],[63],{"type":46,"value":64},"Agent.loop",{"type":46,"value":66},", and\nput side effects inside durable steps:",{"type":41,"tag":68,"props":69,"children":70},"ul",{},[71,77,82],{"type":41,"tag":72,"props":73,"children":74},"li",{},[75],{"type":46,"value":76},"model calls",{"type":41,"tag":72,"props":78,"children":79},{},[80],{"type":46,"value":81},"tool I\u002FO",{"type":41,"tag":72,"props":83,"children":84},{},[85],{"type":46,"value":86},"approval or resume boundaries",{"type":41,"tag":48,"props":88,"children":89},{},[90],{"type":46,"value":91},"Keep the workflow replayable. Durable steps should take JSON inputs and return\nJSON outputs.",{"type":41,"tag":48,"props":93,"children":94},{},[95],{"type":46,"value":96},"Serialize messages like this:",{"type":41,"tag":98,"props":99,"children":103},"pre",{"className":100,"code":101,"language":14,"meta":102,"style":102},"language-python shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","data = message.model_dump(mode=\"json\")\nmessage = ai.messages.Message.model_validate(data)\n","",[104],{"type":41,"tag":59,"props":105,"children":106},{"__ignoreMap":102},[107,118],{"type":41,"tag":108,"props":109,"children":112},"span",{"class":110,"line":111},"line",1,[113],{"type":41,"tag":108,"props":114,"children":115},{},[116],{"type":46,"value":117},"data = message.model_dump(mode=\"json\")\n",{"type":41,"tag":108,"props":119,"children":121},{"class":110,"line":120},2,[122],{"type":41,"tag":108,"props":123,"children":124},{},[125],{"type":46,"value":126},"message = ai.messages.Message.model_validate(data)\n",{"type":41,"tag":128,"props":129,"children":131},"h2",{"id":130},"model-step",[132],{"type":46,"value":133},"Model Step",{"type":41,"tag":48,"props":135,"children":136},{},[137,139,145,147,153],{"type":46,"value":138},"A durable model step should drain ",{"type":41,"tag":59,"props":140,"children":142},{"className":141},[],[143],{"type":46,"value":144},"ai.stream(...)",{"type":46,"value":146}," inside the step and return one\ncomplete assistant ",{"type":41,"tag":59,"props":148,"children":150},{"className":149},[],[151],{"type":46,"value":152},"Message",{"type":46,"value":154},".",{"type":41,"tag":98,"props":156,"children":158},{"className":100,"code":157,"language":14,"meta":102,"style":102},"@workflow.step\nasync def llm_step(\n    model_data: dict[str, object],\n    messages_data: list[dict[str, object]],\n    tools_data: list[dict[str, object]],\n) -> dict[str, object]:\n    model = ai.Model.model_validate(model_data)\n    messages = [\n        ai.messages.Message.model_validate(message)\n        for message in messages_data\n    ]\n    tools = [ai.Tool.model_validate(tool) for tool in tools_data]\n\n    async with ai.stream(model, messages, tools=tools) as stream:\n        async for _event in stream:\n            pass\n\n        if stream.message is None:\n            raise RuntimeError(\"LLM stream ended without a message\")\n\n        return stream.message.model_dump(mode=\"json\")\n",[159],{"type":41,"tag":59,"props":160,"children":161},{"__ignoreMap":102},[162,170,178,187,196,205,214,223,232,241,250,258,267,277,286,295,304,312,321,330,338],{"type":41,"tag":108,"props":163,"children":164},{"class":110,"line":111},[165],{"type":41,"tag":108,"props":166,"children":167},{},[168],{"type":46,"value":169},"@workflow.step\n",{"type":41,"tag":108,"props":171,"children":172},{"class":110,"line":120},[173],{"type":41,"tag":108,"props":174,"children":175},{},[176],{"type":46,"value":177},"async def llm_step(\n",{"type":41,"tag":108,"props":179,"children":181},{"class":110,"line":180},3,[182],{"type":41,"tag":108,"props":183,"children":184},{},[185],{"type":46,"value":186},"    model_data: dict[str, object],\n",{"type":41,"tag":108,"props":188,"children":190},{"class":110,"line":189},4,[191],{"type":41,"tag":108,"props":192,"children":193},{},[194],{"type":46,"value":195},"    messages_data: list[dict[str, object]],\n",{"type":41,"tag":108,"props":197,"children":199},{"class":110,"line":198},5,[200],{"type":41,"tag":108,"props":201,"children":202},{},[203],{"type":46,"value":204},"    tools_data: list[dict[str, object]],\n",{"type":41,"tag":108,"props":206,"children":208},{"class":110,"line":207},6,[209],{"type":41,"tag":108,"props":210,"children":211},{},[212],{"type":46,"value":213},") -> dict[str, object]:\n",{"type":41,"tag":108,"props":215,"children":217},{"class":110,"line":216},7,[218],{"type":41,"tag":108,"props":219,"children":220},{},[221],{"type":46,"value":222},"    model = ai.Model.model_validate(model_data)\n",{"type":41,"tag":108,"props":224,"children":226},{"class":110,"line":225},8,[227],{"type":41,"tag":108,"props":228,"children":229},{},[230],{"type":46,"value":231},"    messages = [\n",{"type":41,"tag":108,"props":233,"children":235},{"class":110,"line":234},9,[236],{"type":41,"tag":108,"props":237,"children":238},{},[239],{"type":46,"value":240},"        ai.messages.Message.model_validate(message)\n",{"type":41,"tag":108,"props":242,"children":244},{"class":110,"line":243},10,[245],{"type":41,"tag":108,"props":246,"children":247},{},[248],{"type":46,"value":249},"        for message in messages_data\n",{"type":41,"tag":108,"props":251,"children":252},{"class":110,"line":26},[253],{"type":41,"tag":108,"props":254,"children":255},{},[256],{"type":46,"value":257},"    ]\n",{"type":41,"tag":108,"props":259,"children":261},{"class":110,"line":260},12,[262],{"type":41,"tag":108,"props":263,"children":264},{},[265],{"type":46,"value":266},"    tools = [ai.Tool.model_validate(tool) for tool in tools_data]\n",{"type":41,"tag":108,"props":268,"children":270},{"class":110,"line":269},13,[271],{"type":41,"tag":108,"props":272,"children":274},{"emptyLinePlaceholder":273},true,[275],{"type":46,"value":276},"\n",{"type":41,"tag":108,"props":278,"children":280},{"class":110,"line":279},14,[281],{"type":41,"tag":108,"props":282,"children":283},{},[284],{"type":46,"value":285},"    async with ai.stream(model, messages, tools=tools) as stream:\n",{"type":41,"tag":108,"props":287,"children":289},{"class":110,"line":288},15,[290],{"type":41,"tag":108,"props":291,"children":292},{},[293],{"type":46,"value":294},"        async for _event in stream:\n",{"type":41,"tag":108,"props":296,"children":298},{"class":110,"line":297},16,[299],{"type":41,"tag":108,"props":300,"children":301},{},[302],{"type":46,"value":303},"            pass\n",{"type":41,"tag":108,"props":305,"children":307},{"class":110,"line":306},17,[308],{"type":41,"tag":108,"props":309,"children":310},{"emptyLinePlaceholder":273},[311],{"type":46,"value":276},{"type":41,"tag":108,"props":313,"children":315},{"class":110,"line":314},18,[316],{"type":41,"tag":108,"props":317,"children":318},{},[319],{"type":46,"value":320},"        if stream.message is None:\n",{"type":41,"tag":108,"props":322,"children":324},{"class":110,"line":323},19,[325],{"type":41,"tag":108,"props":326,"children":327},{},[328],{"type":46,"value":329},"            raise RuntimeError(\"LLM stream ended without a message\")\n",{"type":41,"tag":108,"props":331,"children":333},{"class":110,"line":332},20,[334],{"type":41,"tag":108,"props":335,"children":336},{"emptyLinePlaceholder":273},[337],{"type":46,"value":276},{"type":41,"tag":108,"props":339,"children":341},{"class":110,"line":340},21,[342],{"type":41,"tag":108,"props":343,"children":344},{},[345],{"type":46,"value":346},"        return stream.message.model_dump(mode=\"json\")\n",{"type":41,"tag":128,"props":348,"children":350},{"id":349},"durable-tools",[351],{"type":46,"value":352},"Durable Tools",{"type":41,"tag":48,"props":354,"children":355},{},[356],{"type":46,"value":357},"Prefer wrapping the tool body in the durable step:",{"type":41,"tag":98,"props":359,"children":361},{"className":100,"code":360,"language":14,"meta":102,"style":102},"@ai.tool\n@workflow.step\nasync def ask_mothership(question: str) -> str:\n    response = await mothership_client.ask(question)\n    return response.summary\n",[362],{"type":41,"tag":59,"props":363,"children":364},{"__ignoreMap":102},[365,373,380,388,396],{"type":41,"tag":108,"props":366,"children":367},{"class":110,"line":111},[368],{"type":41,"tag":108,"props":369,"children":370},{},[371],{"type":46,"value":372},"@ai.tool\n",{"type":41,"tag":108,"props":374,"children":375},{"class":110,"line":120},[376],{"type":41,"tag":108,"props":377,"children":378},{},[379],{"type":46,"value":169},{"type":41,"tag":108,"props":381,"children":382},{"class":110,"line":180},[383],{"type":41,"tag":108,"props":384,"children":385},{},[386],{"type":46,"value":387},"async def ask_mothership(question: str) -> str:\n",{"type":41,"tag":108,"props":389,"children":390},{"class":110,"line":189},[391],{"type":41,"tag":108,"props":392,"children":393},{},[394],{"type":46,"value":395},"    response = await mothership_client.ask(question)\n",{"type":41,"tag":108,"props":397,"children":398},{"class":110,"line":198},[399],{"type":41,"tag":108,"props":400,"children":401},{},[402],{"type":46,"value":403},"    return response.summary\n",{"type":41,"tag":48,"props":405,"children":406},{},[407,409,415,417,423],{"type":46,"value":408},"If the workflow system needs separate activity dispatch, schedule a zero-arg\ncallable that returns ",{"type":41,"tag":59,"props":410,"children":412},{"className":411},[],[413],{"type":46,"value":414},"ai.tool_result(...)",{"type":46,"value":416},". Do not call ",{"type":41,"tag":59,"props":418,"children":420},{"className":419},[],[421],{"type":46,"value":422},"tool.fn",{"type":46,"value":424}," directly.",{"type":41,"tag":128,"props":426,"children":428},{"id":427},"agent-loop",[429],{"type":46,"value":430},"Agent Loop",{"type":41,"tag":48,"props":432,"children":433},{},[434,436,442,444,450,452,458],{"type":46,"value":435},"Use the model step result as a complete message. Do not wrap it in ",{"type":41,"tag":59,"props":437,"children":439},{"className":438},[],[440],{"type":46,"value":441},"ai.Stream",{"type":46,"value":443},",\n",{"type":41,"tag":59,"props":445,"children":447},{"className":446},[],[448],{"type":46,"value":449},"ai.events.replay_message_events",{"type":46,"value":451},", or ",{"type":41,"tag":59,"props":453,"children":455},{"className":454},[],[456],{"type":46,"value":457},"ai.util.merge",{"type":46,"value":459},", those utilities are\nused for fluent dispatch in non-durable applications, which is impossible\nin a workflow setting since streams are considered side-effects.",{"type":41,"tag":98,"props":461,"children":463},{"className":100,"code":462,"language":14,"meta":102,"style":102},"class DurableAgent(ai.Agent):\n    async def loop(self, context: ai.Context):\n        while context.keep_running():\n            result = await llm_step(\n                context.model.model_dump(mode=\"json\"),\n                [m.model_dump(mode=\"json\") for m in context.messages],\n                [t.model_dump(mode=\"json\") for t in context.tools],\n            )\n\n            assistant_message = ai.messages.Message.model_validate(result)\n            context.add(assistant_message)\n\n            async with ai.ToolRunner() as runner:\n                for tool_call in assistant_message.tool_calls:\n                    runner.schedule(context.resolve(tool_call))\n\n                async for event in runner.events():\n                    yield event\n\n                context.add(runner.get_tool_message())\n",[464],{"type":41,"tag":59,"props":465,"children":466},{"__ignoreMap":102},[467,475,483,491,499,507,515,523,531,538,546,554,561,569,577,585,592,600,608,615],{"type":41,"tag":108,"props":468,"children":469},{"class":110,"line":111},[470],{"type":41,"tag":108,"props":471,"children":472},{},[473],{"type":46,"value":474},"class DurableAgent(ai.Agent):\n",{"type":41,"tag":108,"props":476,"children":477},{"class":110,"line":120},[478],{"type":41,"tag":108,"props":479,"children":480},{},[481],{"type":46,"value":482},"    async def loop(self, context: ai.Context):\n",{"type":41,"tag":108,"props":484,"children":485},{"class":110,"line":180},[486],{"type":41,"tag":108,"props":487,"children":488},{},[489],{"type":46,"value":490},"        while context.keep_running():\n",{"type":41,"tag":108,"props":492,"children":493},{"class":110,"line":189},[494],{"type":41,"tag":108,"props":495,"children":496},{},[497],{"type":46,"value":498},"            result = await llm_step(\n",{"type":41,"tag":108,"props":500,"children":501},{"class":110,"line":198},[502],{"type":41,"tag":108,"props":503,"children":504},{},[505],{"type":46,"value":506},"                context.model.model_dump(mode=\"json\"),\n",{"type":41,"tag":108,"props":508,"children":509},{"class":110,"line":207},[510],{"type":41,"tag":108,"props":511,"children":512},{},[513],{"type":46,"value":514},"                [m.model_dump(mode=\"json\") for m in context.messages],\n",{"type":41,"tag":108,"props":516,"children":517},{"class":110,"line":216},[518],{"type":41,"tag":108,"props":519,"children":520},{},[521],{"type":46,"value":522},"                [t.model_dump(mode=\"json\") for t in context.tools],\n",{"type":41,"tag":108,"props":524,"children":525},{"class":110,"line":225},[526],{"type":41,"tag":108,"props":527,"children":528},{},[529],{"type":46,"value":530},"            )\n",{"type":41,"tag":108,"props":532,"children":533},{"class":110,"line":234},[534],{"type":41,"tag":108,"props":535,"children":536},{"emptyLinePlaceholder":273},[537],{"type":46,"value":276},{"type":41,"tag":108,"props":539,"children":540},{"class":110,"line":243},[541],{"type":41,"tag":108,"props":542,"children":543},{},[544],{"type":46,"value":545},"            assistant_message = ai.messages.Message.model_validate(result)\n",{"type":41,"tag":108,"props":547,"children":548},{"class":110,"line":26},[549],{"type":41,"tag":108,"props":550,"children":551},{},[552],{"type":46,"value":553},"            context.add(assistant_message)\n",{"type":41,"tag":108,"props":555,"children":556},{"class":110,"line":260},[557],{"type":41,"tag":108,"props":558,"children":559},{"emptyLinePlaceholder":273},[560],{"type":46,"value":276},{"type":41,"tag":108,"props":562,"children":563},{"class":110,"line":269},[564],{"type":41,"tag":108,"props":565,"children":566},{},[567],{"type":46,"value":568},"            async with ai.ToolRunner() as runner:\n",{"type":41,"tag":108,"props":570,"children":571},{"class":110,"line":279},[572],{"type":41,"tag":108,"props":573,"children":574},{},[575],{"type":46,"value":576},"                for tool_call in assistant_message.tool_calls:\n",{"type":41,"tag":108,"props":578,"children":579},{"class":110,"line":288},[580],{"type":41,"tag":108,"props":581,"children":582},{},[583],{"type":46,"value":584},"                    runner.schedule(context.resolve(tool_call))\n",{"type":41,"tag":108,"props":586,"children":587},{"class":110,"line":297},[588],{"type":41,"tag":108,"props":589,"children":590},{"emptyLinePlaceholder":273},[591],{"type":46,"value":276},{"type":41,"tag":108,"props":593,"children":594},{"class":110,"line":306},[595],{"type":41,"tag":108,"props":596,"children":597},{},[598],{"type":46,"value":599},"                async for event in runner.events():\n",{"type":41,"tag":108,"props":601,"children":602},{"class":110,"line":314},[603],{"type":41,"tag":108,"props":604,"children":605},{},[606],{"type":46,"value":607},"                    yield event\n",{"type":41,"tag":108,"props":609,"children":610},{"class":110,"line":323},[611],{"type":41,"tag":108,"props":612,"children":613},{"emptyLinePlaceholder":273},[614],{"type":46,"value":276},{"type":41,"tag":108,"props":616,"children":617},{"class":110,"line":332},[618],{"type":41,"tag":108,"props":619,"children":620},{},[621],{"type":46,"value":622},"                context.add(runner.get_tool_message())\n",{"type":41,"tag":48,"props":624,"children":625},{},[626],{"type":46,"value":627},"This pattern does not stream model tokens to the caller. That is usually the\nright tradeoff for durable workflows, because many durable systems do not support\nasync generators. You can build a queue-based side channel for streaming; however,\nthat kind of stream can't be used to dispatch tools and affect control flow directly.",{"type":41,"tag":629,"props":630,"children":631},"style",{},[632],{"type":46,"value":633},"html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"items":635,"total":805},[636,652,664,676,691,708,720,733,746,759,771,790],{"slug":637,"name":637,"fn":638,"description":639,"org":640,"tags":641,"stars":649,"repoUrl":650,"updatedAt":651},"agent-browser","automate browser interactions for AI agents","Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to \"open a website\", \"fill out a form\", \"click a button\", \"take a screenshot\", \"scrape data from a page\", \"test this web app\", \"login to a site\", \"automate browser actions\", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[642,643,646],{"name":17,"slug":18,"type":15},{"name":644,"slug":645,"type":15},"Automation","automation",{"name":647,"slug":648,"type":15},"Browser Automation","browser-automation",38346,"https:\u002F\u002Fgithub.com\u002Fvercel-labs\u002Fagent-browser","2026-07-20T05:55:17.314329",{"slug":653,"name":653,"fn":654,"description":655,"org":656,"tags":657,"stars":649,"repoUrl":650,"updatedAt":663},"agentcore","run browser automation on AWS Bedrock","Run agent-browser on AWS Bedrock AgentCore cloud browsers. Use when the user wants to use AgentCore, run browser automation on AWS, use a cloud browser with AWS credentials, or needs a managed browser session backed by AWS infrastructure. Triggers include \"use agentcore\", \"run on AWS\", \"cloud browser with AWS\", \"bedrock browser\", \"agentcore session\", or any task requiring AWS-hosted browser automation.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[658,659,662],{"name":644,"slug":645,"type":15},{"name":660,"slug":661,"type":15},"AWS","aws",{"name":647,"slug":648,"type":15},"2026-07-17T06:08:33.665276",{"slug":665,"name":665,"fn":666,"description":667,"org":668,"tags":669,"stars":649,"repoUrl":650,"updatedAt":675},"core","navigate and interact with web pages","Core agent-browser usage guide. Read this before running any agent-browser commands. Covers the snapshot-and-ref workflow, navigating pages, interacting with elements (click, fill, type, select), extracting text and data, taking screenshots, managing tabs, handling forms and auth, waiting for content, running multiple browser sessions in parallel, and troubleshooting common failures. Use when the user asks to interact with a website, fill a form, click something, extract data, take a screenshot, log into a site, test a web app, or automate any browser task.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[670,671,672],{"name":17,"slug":18,"type":15},{"name":647,"slug":648,"type":15},{"name":673,"slug":674,"type":15},"Navigation","navigation","2026-07-26T05:47:42.378419",{"slug":677,"name":677,"fn":678,"description":679,"org":680,"tags":681,"stars":649,"repoUrl":650,"updatedAt":690},"derive-client","reverse engineer internal APIs from browser traffic","Reverse-engineer a website's internal API by recording browser traffic into a HAR file, then generate a standalone client or CLI that calls the endpoints directly, with no browser needed after the first recording. Use when asked to \"derive a client\", \"build a CLI for \u003Csite>\", \"reverse engineer this site's API\", \"record network requests\", \"turn this site into an API\", or when the same site will be automated repeatedly and direct HTTP calls would beat driving the browser every time.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[682,685,686,687],{"name":683,"slug":684,"type":15},"API Development","api-development",{"name":644,"slug":645,"type":15},{"name":647,"slug":648,"type":15},{"name":688,"slug":689,"type":15},"Web Scraping","web-scraping","2026-07-20T06:24:11.928835",{"slug":692,"name":692,"fn":693,"description":694,"org":695,"tags":696,"stars":649,"repoUrl":650,"updatedAt":707},"dogfood","perform exploratory testing on web applications","Systematically explore and test a web application to find bugs, UX issues, and other problems. Use when asked to \"dogfood\", \"QA\", \"exploratory test\", \"find issues\", \"bug hunt\", \"test this app\u002Fsite\u002Fplatform\", or review the quality of a web application. Produces a structured report with full reproduction evidence -- step-by-step screenshots, repro videos, and detailed repro steps for every issue -- so findings can be handed directly to the responsible teams.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[697,698,701,704],{"name":647,"slug":648,"type":15},{"name":699,"slug":700,"type":15},"Debugging","debugging",{"name":702,"slug":703,"type":15},"QA","qa",{"name":705,"slug":706,"type":15},"Testing","testing","2026-07-17T06:07:41.421482",{"slug":709,"name":709,"fn":710,"description":711,"org":712,"tags":713,"stars":649,"repoUrl":650,"updatedAt":719},"electron","automate Electron desktop applications","Automate Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify, etc.) using agent-browser via Chrome DevTools Protocol. Use when the user needs to interact with an Electron app, automate a desktop app, connect to a running app, control a native app, or test an Electron application. Triggers include \"automate Slack app\", \"control VS Code\", \"interact with Discord app\", \"test this Electron app\", \"connect to desktop app\", or any task requiring automation of a native Electron application.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[714,715,716],{"name":17,"slug":18,"type":15},{"name":647,"slug":648,"type":15},{"name":717,"slug":718,"type":15},"Desktop","desktop","2026-07-17T06:08:28.007783",{"slug":721,"name":721,"fn":722,"description":723,"org":724,"tags":725,"stars":649,"repoUrl":650,"updatedAt":732},"slack","interact with Slack workspaces","Interact with Slack workspaces using browser automation. Use when the user needs to check unread channels, navigate Slack, send messages, extract data, find information, search conversations, or automate any Slack task. Triggers include \"check my Slack\", \"what channels have unreads\", \"send a message to\", \"search Slack for\", \"extract from Slack\", \"find who said\", or any task requiring programmatic Slack interaction.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[726,727,730],{"name":647,"slug":648,"type":15},{"name":728,"slug":729,"type":15},"Messaging","messaging",{"name":731,"slug":721,"type":15},"Slack","2026-07-17T06:08:27.679015",{"slug":734,"name":734,"fn":735,"description":736,"org":737,"tags":738,"stars":649,"repoUrl":650,"updatedAt":745},"vercel-sandbox","run browser automation in Vercel Sandbox","Run agent-browser + Chrome inside Vercel Sandbox microVMs for browser automation from any Vercel-deployed app. Use when the user needs browser automation in a Vercel app (Next.js, SvelteKit, Nuxt, Remix, Astro, etc.), wants to run headless Chrome without binary size limits, needs persistent browser sessions across commands, or wants ephemeral isolated browser environments. Triggers include \"Vercel Sandbox browser\", \"microVM Chrome\", \"agent-browser in sandbox\", \"browser automation on Vercel\", or any task requiring Chrome in a Vercel Sandbox.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[739,740,741,742],{"name":644,"slug":645,"type":15},{"name":647,"slug":648,"type":15},{"name":705,"slug":706,"type":15},{"name":743,"slug":744,"type":15},"Vercel","vercel","2026-07-17T06:08:28.349899",{"slug":747,"name":747,"fn":748,"description":749,"org":750,"tags":751,"stars":756,"repoUrl":757,"updatedAt":758},"deploy-to-vercel","deploy applications to Vercel","Deploy applications and websites to Vercel. Use when the user requests deployment actions like \"deploy my app\", \"deploy and give me the link\", \"push this live\", or \"create a preview deployment\".",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[752,755],{"name":753,"slug":754,"type":15},"Deployment","deployment",{"name":743,"slug":744,"type":15},28993,"https:\u002F\u002Fgithub.com\u002Fvercel-labs\u002Fagent-skills","2026-07-17T06:08:41.18374",{"slug":760,"name":760,"fn":761,"description":762,"org":763,"tags":764,"stars":756,"repoUrl":757,"updatedAt":770},"vercel-cli-with-tokens","manage Vercel projects via CLI","Deploy and manage projects on Vercel using token-based authentication. Use when working with Vercel CLI using access tokens rather than interactive login — e.g. \"deploy to vercel\", \"set up vercel\", \"add environment variables to vercel\".",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[765,768,769],{"name":766,"slug":767,"type":15},"CLI","cli",{"name":753,"slug":754,"type":15},{"name":743,"slug":744,"type":15},"2026-07-17T06:08:41.84179",{"slug":772,"name":772,"fn":773,"description":774,"org":775,"tags":776,"stars":756,"repoUrl":757,"updatedAt":789},"vercel-composition-patterns","implement scalable React composition patterns","React composition patterns that scale. Use when refactoring components with boolean prop proliferation, building flexible component libraries, or designing reusable APIs. Triggers on tasks involving compound components, render props, context providers, or component architecture. Includes React 19 API changes.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[777,780,783,786],{"name":778,"slug":779,"type":15},"Best Practices","best-practices",{"name":781,"slug":782,"type":15},"Frontend","frontend",{"name":784,"slug":785,"type":15},"React","react",{"name":787,"slug":788,"type":15},"UI Components","ui-components","2026-07-17T06:05:40.576913",{"slug":791,"name":791,"fn":792,"description":793,"org":794,"tags":795,"stars":756,"repoUrl":757,"updatedAt":804},"vercel-optimize","optimize Vercel project performance and costs","Use for Vercel cost and performance optimization on deployed projects, especially Next.js, SvelteKit, Nuxt, and limited Astro apps. Collect Vercel metrics, usage, project config, and code scan results first; investigate only metric-backed candidates; produce ranked recommendations grounded in verified files and version-aware Vercel\u002Fframework docs. Trigger for Vercel bill reduction, slow or expensive routes, caching opportunities, Function Invocations, Build Minutes, Fast Data Transfer, Core Web Vitals, Bot Management, Fluid compute, or cost breakdown requests.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[796,799,800,803],{"name":797,"slug":798,"type":15},"Cost Optimization","cost-optimization",{"name":753,"slug":754,"type":15},{"name":801,"slug":802,"type":15},"Performance","performance",{"name":743,"slug":744,"type":15},"2026-07-17T06:04:08.327515",100,{"items":807,"total":234},[808,821,831,840,846,858,868],{"slug":809,"name":809,"fn":810,"description":811,"org":812,"tags":813,"stars":22,"repoUrl":23,"updatedAt":820},"ai-python-basics","build AI agents with Python SDK","Use for AI SDK for Python basics. Configure a model, make messages, stream, declare tools, build a basic agent.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[814,815,816,819],{"name":17,"slug":18,"type":15},{"name":20,"slug":21,"type":15},{"name":817,"slug":818,"type":15},"LLM","llm",{"name":13,"slug":14,"type":15},"2026-07-29T05:40:38.753254",{"slug":822,"name":822,"fn":823,"description":824,"org":825,"tags":826,"stars":22,"repoUrl":23,"updatedAt":830},"ai-python-custom-loop","build custom agent loops in Python","Use when building custom agent loops. Modify tool dispatch, history management, hooks, control flow.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[827,828,829],{"name":17,"slug":18,"type":15},{"name":20,"slug":21,"type":15},{"name":13,"slug":14,"type":15},"2026-07-29T05:40:40.763184",{"slug":832,"name":832,"fn":833,"description":834,"org":835,"tags":836,"stars":22,"repoUrl":23,"updatedAt":839},"ai-python-custom-provider","implement custom AI SDK providers","Use for implementing custom providers in AI SDK for Python.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[837,838],{"name":20,"slug":21,"type":15},{"name":13,"slug":14,"type":15},"2026-07-29T05:40:39.785476",{"slug":4,"name":4,"fn":5,"description":6,"org":841,"tags":842,"stars":22,"repoUrl":23,"updatedAt":24},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[843,844,845],{"name":17,"slug":18,"type":15},{"name":20,"slug":21,"type":15},{"name":13,"slug":14,"type":15},{"slug":847,"name":847,"fn":848,"description":849,"org":850,"tags":851,"stars":22,"repoUrl":23,"updatedAt":857},"ai-python-serverless-execution","build serverless AI SDK endpoints","Use when building serverless AI SDK for Python endpoints, handling hook approvals, deferring hooks, or resuming runs across requests.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[852,853,854],{"name":20,"slug":21,"type":15},{"name":13,"slug":14,"type":15},{"name":855,"slug":856,"type":15},"Serverless","serverless","2026-07-29T05:40:37.750244",{"slug":859,"name":859,"fn":860,"description":861,"org":862,"tags":863,"stars":22,"repoUrl":23,"updatedAt":867},"ai-python-streaming-tools","stream AI SDK Python tool outputs","Use for AI SDK for Python async-generator tools, streaming tool output, subagent tools, PartialToolCallResult events, and custom tool aggregation.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[864,865,866],{"name":17,"slug":18,"type":15},{"name":20,"slug":21,"type":15},{"name":13,"slug":14,"type":15},"2026-07-29T05:40:36.755057",{"slug":869,"name":869,"fn":870,"description":871,"org":872,"tags":873,"stars":22,"repoUrl":23,"updatedAt":879},"ai-python-subagents","implement subagent patterns in Python","Use for the subagent-as-a-tool pattern.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[874,875,878],{"name":17,"slug":18,"type":15},{"name":876,"slug":877,"type":15},"Multi-Agent","multi-agent",{"name":13,"slug":14,"type":15},"2026-07-29T05:40:35.804764"]