{"success":true,"data":{"sites":[{"id":"cursor","name":"Cursor","url":"https://www.cursor.com","icon":"https://www.google.com/s2/favicons?domain=cursor.com&sz=64","description":"The leading AI coding IDE. Combines the editor, codebase understanding, and agent task loops for daily work and multi-file changes.","longDescription":"The leading AI coding IDE. Combines the editor, codebase understanding, and agent task loops for daily work and multi-file changes.","category":["coding"],"capabilities":["Multi-file agent edits","Semantic codebase search","Inline / Tab completion","Terminal and browser tools","Rules and Skills workflows"],"characteristics":["Agent and IDE in one surface","Strong Git-repo context","Clear paid tiers for teams","Extension ecosystem still thinner than VS Code"],"evaluation":"Composite score 94. First-tier capability and ecosystem: Agent mode is best when you state a goal and let it ship a feature. Weak spots are long-running stability and occasional over-editing. Default pick for coding agents.","skills":[{"id":"cursor-agent-scope","title":"Scope the Agent before you run it","description":"Pin @files / @folders. Dumping the whole repo is less reliable.","prompt":"Name the directories and acceptance checks first. Avoid “refactor the entire project” in one shot.","example":"@app/api Add a lang query to GET /sites, keep pagination, add one test.","category":"workflow","rating":5},{"id":"cursor-rules","title":"Encode style in project rules","description":"Write architecture constraints so the agent does not invent a second stack.","prompt":"Rules should say do / do-not. Example: no new dependencies; always go through the catalog layer.","example":"Do not add new data sources. Always load agents from docs/sites via lib/catalog.","category":"advanced","rating":5}],"rank":1,"score":94,"popularity":94,"scoreBreakdown":{"capability":96,"reliability":90,"ecosystem":95,"value":88},"status":"active","lastUpdated":"2026-08-20","tags":["coding","ide","agent"]},{"id":"claude-code","name":"Claude Code","url":"https://claude.com/product/claude-code","icon":"https://www.google.com/s2/favicons?domain=claude.com&sz=64","description":"Anthropic’s CLI coding agent. Strong long-context, tool use, and test-fix loops for repo-scale work.","longDescription":"Anthropic’s CLI coding agent. Strong long-context, tool use, and test-fix loops for repo-scale work.","category":["coding"],"capabilities":["Multi-step coding in the terminal","Repo edits and test runs","Long-session planning","MCP / tool extensions","Tight Claude model integration"],"characteristics":["CLI-first, fits SSH and CI","Steady reasoning and clear diffs","Can handle non-code chores too","Assumes you are fine in a terminal"],"evaluation":"Composite score 93. Capability and reliability sit next to Cursor; ecosystem is CLI/MCP-centric. Best for teams that want the agent to run tests until green. Not a GUI-first product.","skills":[{"id":"claude-code-plan","title":"Plan before it touches code","description":"Ask for a file list and risks, then approve execution.","prompt":"First output a plan: files, verification, risks. Wait for approval before editing.","example":"Plan a pagination bugfix in getSites. Do not edit until I approve the file list.","category":"workflow","rating":5},{"id":"claude-code-verify","title":"Make tests the done condition","description":"Say “tests must pass”. It will iterate instead of dropping a patch note.","prompt":"Done means the relevant tests pass. Keep fixing on failure.","example":"Run npm test and npm run build. A red test means the task is unfinished.","category":"advanced","rating":5}],"rank":2,"score":93,"popularity":93,"scoreBreakdown":{"capability":95,"reliability":92,"ecosystem":88,"value":90},"status":"active","lastUpdated":"2026-08-20","tags":["coding","cli","agent"]},{"id":"chatgpt-agent","name":"ChatGPT Agent","url":"https://chatgpt.com","icon":"https://www.google.com/s2/favicons?domain=chatgpt.com&sz=64","description":"OpenAI’s general agent. Chat, browsing, code interpreter, and task orchestration in one product — knowledge work, not only code.","longDescription":"OpenAI’s general agent. Chat, browsing, code interpreter, and task orchestration in one product — knowledge work, not only code.","category":["general","autonomous"],"capabilities":["Multi-step task orchestration","Browsing and file handling","Code interpreter","GPTs / tool ecosystem","Voice and multimodal input"],"characteristics":["Largest user base, lowest ramp","Widest general-task coverage","Shallower than dedicated coding IDEs","Enterprise controls vary by plan"],"evaluation":"Composite score 89. Still the strongest “ask anything, sometimes finish a task” general agent. Pair it with Cursor / Claude Code for serious software work.","skills":[{"id":"chatgpt-task-brief","title":"Write an acceptance brief","description":"Goal, inputs, output format, and prohibitions keep it on track.","prompt":"Goal / source materials / delivery format / do-nots. Ask if any are missing.","example":"Summarize these 8 PDFs into a 1-page comparison table. Do not invent citations.","category":"basic","rating":5},{"id":"chatgpt-tools","title":"Say when to browse or run code","description":"Live facts need browsing; numbers need the interpreter.","prompt":"For this year’s product changes, browse the official site. Compute figures in code.","example":"Check Cursor’s current pricing page, then compare seats for a 12-person team.","category":"advanced","rating":4}],"rank":3,"score":89,"popularity":89,"scoreBreakdown":{"capability":92,"reliability":87,"ecosystem":94,"value":84},"status":"active","lastUpdated":"2026-08-20","tags":["general","autonomous","agent"]},{"id":"github-copilot","name":"GitHub Copilot","url":"https://github.com/features/copilot","icon":"https://www.google.com/s2/favicons?domain=github.com&sz=64","description":"GitHub / Microsoft coding assistant and agent. Broadest completion coverage, tightest fit with repos, PRs, and Actions.","longDescription":"GitHub / Microsoft coding assistant and agent. Broadest completion coverage, tightest fit with repos, PRs, and Actions.","category":["coding"],"capabilities":["Completions in many IDEs","Copilot Chat / Agent","PR and repo Q&A","Enterprise policy","GitHub workflow integration"],"characteristics":["Runs in almost every mainstream IDE","Mature procurement and permissions","Agent depth trails Cursor / Claude Code","Wins as the company-wide default"],"evaluation":"Composite score 90. Best ecosystem and reliability. Right default for GitHub-native teams. For the sharpest repo-scale agent loop, Cursor or Claude Code still lead.","skills":[{"id":"copilot-pr","title":"Review PRs, do not only write code","description":"Point it at the diff for regressions and missing tests.","prompt":"From this PR diff, list regression risks and missing tests. Do not rewrite unrelated files.","example":"Review this pull request for auth regressions. Suggest tests only.","category":"workflow","rating":4},{"id":"copilot-workspace","title":"Give acceptance criteria from the issue","description":"Clear Issue checks stop “it looks like it runs” PRs.","prompt":"Implement the issue. Keep API contracts, add tests, mention risks in the description.","example":"Implement issue #241. Keep the REST envelope. Add a test for lang=zh.","category":"basic","rating":4}],"rank":4,"score":90,"popularity":90,"scoreBreakdown":{"capability":88,"reliability":91,"ecosystem":97,"value":86},"status":"active","lastUpdated":"2026-08-20","tags":["coding","ide","agent"]},{"id":"devin","name":"Devin","url":"https://devin.ai","icon":"https://www.google.com/s2/favicons?domain=devin.ai&sz=64","description":"Cognition’s autonomous software engineer. Takes an Issue, boots an environment, edits, tests, and opens a PR — a remote teammate, not a plugin.","longDescription":"Cognition’s autonomous software engineer. Takes an Issue, boots an environment, edits, tests, and opens a PR — a remote teammate, not a plugin.","category":["autonomous","coding"],"capabilities":["Issue → PR end to end","Cloud dev environments","Tests and browser checks","Repo-scale planning","Team task queues"],"characteristics":["Among the most autonomous agents","Best on well-bounded backend/fix work","Pricing is steep for individuals","Fuzzy product work still needs a human"],"evaluation":"Composite score 86. High ceiling, reliability tracks how testable the task is. Great for queued bugs and small features with tests. Poor fit for open-ended product exploration.","skills":[{"id":"devin-issue","title":"Put env and test commands in the Issue","description":"Without repro and test commands, you get empty PRs.","prompt":"Include repro, expected behavior, test command, and off-limits directories.","example":"Bug: search ignores skills. Repro: q=API. Test: npm test. Do not touch i18n.","category":"workflow","rating":5},{"id":"devin-scope","title":"One mergeable goal per run","description":"Split epics. Success rate beats “finish the whole module”.","prompt":"Hit this one acceptance check, then stop. No drive-by refactors.","example":"Add relatedSites to the modal only. No visual redesign.","category":"basic","rating":4}],"rank":5,"score":86,"popularity":86,"scoreBreakdown":{"capability":90,"reliability":80,"ecosystem":84,"value":82},"status":"active","lastUpdated":"2026-08-20","tags":["autonomous","coding","agent"]},{"id":"windsurf","name":"Windsurf","url":"https://windsurf.com","icon":"https://www.google.com/s2/favicons?domain=windsurf.com&sz=64","description":"Codeium’s AI IDE (Cascade agent). Cursor-like: in-editor streaming agents with an emphasis on speed.","longDescription":"Codeium’s AI IDE (Cascade agent). Cursor-like: in-editor streaming agents with an emphasis on speed.","category":["coding"],"capabilities":["Cascade multi-file agent","Fast completions","Codebase indexing","In-IDE chat","Rules and workflows"],"characteristics":["Familiar if you know Cursor","Completion speed is the pitch","Ecosystem still trails the leader","Sensible Cursor alternative"],"evaluation":"Composite score 85. Right product shape, solid value. Limited upside if the team is already deep on Cursor. Worth a bake-off if you have not picked an AI IDE yet.","skills":[{"id":"windsurf-cascade","title":"One feature slice per Cascade run","description":"Cut work to what Cascade can see in one pass.","prompt":"Implement list filtering only. Do not touch the detail modal. Summarize the diff.","example":"Add category tabs only. Leave search for a later run.","category":"workflow","rating":4},{"id":"windsurf-index","title":"Wait for the index before architecture questions","description":"Unindexed repos make the agent guess.","prompt":"Confirm lib/catalog is indexed, then explain where site data is loaded.","example":"Where is getSites implemented? Cite the file path.","category":"basic","rating":4}],"rank":6,"score":85,"popularity":85,"scoreBreakdown":{"capability":86,"reliability":84,"ecosystem":83,"value":87},"status":"active","lastUpdated":"2026-08-20","tags":["coding","ide","agent"]},{"id":"manus","name":"Manus","url":"https://manus.im","icon":"https://www.google.com/s2/favicons?domain=manus.im&sz=64","description":"A general autonomous agent for complex jobs. Breaks down goals and drives a browser/tools to finish research, tables, and drafts — a cloud worker.","longDescription":"A general autonomous agent for complex jobs. Breaks down goals and drives a browser/tools to finish research, tables, and drafts — a cloud worker.","category":["autonomous","general"],"capabilities":["Long-task decomposition","Browser operation","Docs and spreadsheet delivery","Multi-tool orchestration","Cloud sessions"],"characteristics":["Strong “just finish it” feel","Popular in Chinese-language workflows","High variance on long jobs","Not a primary IDE"],"evaluation":"Composite score 84. Stands out for non-coding work you want finished. Reliability still swings; humans should accept the deliverable. Keep a dedicated coding agent for the main engineering path.","skills":[{"id":"manus-deliverable","title":"Specify the artifact, not the keystrokes","description":"Autonomous agents optimize for outputs, not micro-steps.","prompt":"Deliver a 12-agent comparison table: capabilities, pricing band, audience. No long essay.","example":"Output a spreadsheet with columns: name, score, best for, main risk.","category":"workflow","rating":5},{"id":"manus-checkpoint","title":"Insert a checkpoint","description":"Pause after research so a bad premise does not run to the end.","prompt":"List sources first. Wait for confirmation before conclusions.","example":"Stop after the source list. Do not draft the final memo yet.","category":"advanced","rating":4}],"rank":7,"score":84,"popularity":84,"scoreBreakdown":{"capability":88,"reliability":78,"ecosystem":80,"value":85},"status":"active","lastUpdated":"2026-08-20","tags":["general","autonomous","agent"]},{"id":"gemini","name":"Gemini","url":"https://gemini.google.com","icon":"https://www.google.com/s2/favicons?domain=gemini.google.com&sz=64","description":"Google’s general model and agent surface. Search, Workspace, long context, and multimodal first — with Gemini CLI / Jules catching up on coding.","longDescription":"Google’s general model and agent surface. Search, Workspace, long context, and multimodal first — with Gemini CLI / Jules catching up on coding.","category":["general","research"],"capabilities":["Search-grounded answers","Very long context","Multimodal understanding","Workspace integration","CLI / coding-agent extras"],"characteristics":["Deep Google account / Docs / Drive fit","Excellent research retrieval","Coding-agent UX still chasing specialists","Generous free-tier surface"],"evaluation":"Composite score 83. Ecosystem score is high: Workspace-native teams get a lot for free. For repo-scale edits, still pair with a coding agent.","skills":[{"id":"gemini-ground","title":"Demand sources on research tasks","description":"Use the search stack. Ban unsourced summaries.","prompt":"Attach a source link to each key fact. Mark uncertainty explicitly.","example":"Compare Gemini CLI and Claude Code with citations from official docs only.","category":"basic","rating":5},{"id":"gemini-workspace","title":"Treat Docs / Sheets as the canvas","description":"Editing the doc beats pasting back into chat.","prompt":"Update the Agent scores section in the named Google Doc. Keep heading structure.","example":"Update the Scores section in the shared doc. Do not change other headings.","category":"workflow","rating":4}],"rank":8,"score":83,"popularity":83,"scoreBreakdown":{"capability":85,"reliability":84,"ecosystem":92,"value":86},"status":"active","lastUpdated":"2026-08-20","tags":["general","research","agent"]},{"id":"cline","name":"Cline","url":"https://cline.bot","icon":"https://www.google.com/s2/favicons?domain=cline.bot&sz=64","description":"Open-source VS Code coding agent (and forks like Roo). Bring your own model, MCP, browser, and terminal — fully self-hostable.","longDescription":"Open-source VS Code coding agent (and forks like Roo). Bring your own model, MCP, browser, and terminal — fully self-hostable.","category":["opensource","coding"],"capabilities":["VS Code agent loop","Bring-your-own model","MCP tools","Terminal and browser","Open-source and auditable"],"characteristics":["Best-in-class value","No single-vendor lock-in","You operate keys and spend","UX is rougher; docs are community-led"],"evaluation":"Composite score 82. Open source and value are the win. Fits teams that can manage keys and want an auditable agent. Out-of-box polish trails Cursor.","skills":[{"id":"cline-model","title":"Swap models by task","description":"Plan on a frontier model; do mechanical edits on a cheap one.","prompt":"Architecture on a flagship model; bulk renames on a small one. State which model you are using.","example":"Plan with a frontier model, then apply the mechanical edits on a cheap one.","category":"advanced","rating":5},{"id":"cline-permissions","title":"Tighten auto-approve first","description":"Wide default permissions can delete files.","prompt":"Do not auto-run dangerous commands. rm, force-push, and .env edits need a human.","example":"Never auto-approve shell commands that delete files or change git history.","category":"basic","rating":5}],"rank":9,"score":82,"popularity":82,"scoreBreakdown":{"capability":84,"reliability":80,"ecosystem":86,"value":92},"status":"active","lastUpdated":"2026-08-20","tags":["coding","opensource","agent"]},{"id":"trae","name":"Trae","url":"https://www.trae.ai","icon":"https://www.google.com/s2/favicons?domain=trae.ai&sz=64","description":"ByteDance’s AI IDE. Fast adoption among Chinese developers, with Builder / Agent aimed at idea-to-running-app speed.","longDescription":"ByteDance’s AI IDE. Fast adoption among Chinese developers, with Builder / Agent aimed at idea-to-running-app speed.","category":["coding"],"capabilities":["In-IDE agent","Chinese-language product fit","Project scaffolding","Completions and refactors","Generous free tier"],"characteristics":["Strong Chinese UX and network fit","Fast 0-to-1","Shallower on huge existing repos","Global ecosystem still growing"],"evaluation":"Composite score 81. Outstanding value for Chinese-speaking teams and solos. For large TypeScript monorepos, Cursor / Claude Code stay more reliable. Trae shines at bootstrapping and small-to-mid projects.","skills":[{"id":"trae-bootstrap","title":"Scaffold fast, then freeze structure","description":"Speed is the feature. Write directory rules immediately.","prompt":"After scaffolding Next.js App Router, encode folder rules. No parallel trees.","example":"Scaffold the app, then freeze routes under app/ and data under docs/sites.","category":"workflow","rating":4},{"id":"trae-zh","title":"Keep UI language and identifiers consistent","description":"Mixed Chinese requirements and English identifiers cause bad renames.","prompt":"Chinese for UI copy; English for code identifiers. No pinyin function names.","example":"Keep function names in English. Put Chinese only in i18n strings.","category":"basic","rating":4}],"rank":10,"score":81,"popularity":81,"scoreBreakdown":{"capability":82,"reliability":80,"ecosystem":78,"value":90},"status":"active","lastUpdated":"2026-08-20","tags":["coding","ide","agent"]},{"id":"perplexity","name":"Perplexity","url":"https://www.perplexity.ai","icon":"https://www.google.com/s2/favicons?domain=perplexity.ai&sz=64","description":"A citation-first research agent / search. Built for verifiable roundups and tracking — not for editing your repo.","longDescription":"A citation-first research agent / search. Built for verifiable roundups and tracking — not for editing your repo.","category":["research","general"],"capabilities":["Cited Q&A","Multi-source comparison","Web and academic search","Collections and monitoring","Pro deep research"],"characteristics":["Citations are the product","Fits research and competitive scans","Almost no repo-scale coding","Deep research trades speed for quality"],"evaluation":"Composite score 80. More trustworthy than general chat for “look this up before we decide”. Do not use it as a coding agent. Use it as the research layer, then hand conclusions to a coding agent.","skills":[{"id":"perplexity-compare","title":"Lock sources on comparison tasks","description":"Named source types beat blog-to-blog citation loops.","prompt":"Official docs plus one major review only. List citations.","example":"Compare Devin vs Claude Code using official docs plus one reputable review.","category":"basic","rating":5},{"id":"perplexity-focus","title":"Turn on Focus before a narrow question","description":"Academic / finance / coding Focus cuts noise.","prompt":"Academic focus: Agent evals after 2025. Exclude marketing pages.","example":"Academic focus: papers on autonomous SWE agents since 2025.","category":"advanced","rating":4}],"rank":11,"score":80,"popularity":80,"scoreBreakdown":{"capability":84,"reliability":86,"ecosystem":79,"value":83},"status":"active","lastUpdated":"2026-08-20","tags":["research","general","agent"]},{"id":"replit-agent","name":"Replit Agent","url":"https://replit.com","icon":"https://www.google.com/s2/favicons?domain=replit.com&sz=64","description":"Replit’s cloud agent. Natural language to a running app, with hosting and a database — prototypes and teaching, not complex brownfield systems.","longDescription":"Replit’s cloud agent. Natural language to a running app, with hosting and a database — prototypes and teaching, not complex brownfield systems.","category":["autonomous","coding"],"capabilities":["NL to full-stack apps","Cloud run and hosting","Built-in DB and secrets","Share / preview links","Iterative requirement changes"],"characteristics":["Fastest 0-to-clickable-URL","No local toolchain","Painful to import huge existing repos","Limited control of production architecture"],"evaluation":"Composite score 78. Excellent for prototypes, hackathons, and internal demos. Do not lift a serious codebase into it. Incubate there, then export to Git.","skills":[{"id":"replit-iterate","title":"Patch the preview, do not restate the product","description":"One change against the current preview beats a regenerate.","prompt":"Keep the data. Move the score badge to the top-right of each card.","example":"Do not regenerate the app. Move the score badge to the top-right of each card.","category":"workflow","rating":4},{"id":"replit-mvp","title":"Ask for a demoable MVP first","description":"Auth + multi-tenant on pass one produces a mess.","prompt":"Single-user list + detail, running. Auth in a later pass.","example":"Build a ranked agent list with a detail modal. No auth yet.","category":"basic","rating":4}],"rank":12,"score":78,"popularity":78,"scoreBreakdown":{"capability":80,"reliability":76,"ecosystem":82,"value":84},"status":"active","lastUpdated":"2026-08-20","tags":["autonomous","coding","agent"]}],"pagination":{"total":12,"page":1,"pageSize":12,"hasMore":false}},"mcp":{"version":"1.0","endpoint":"/api/mcp/sites"},"timestamp":"2026-08-20T09:13:34.720Z"}