[
 {
  "slug": "claude-code-token-savers",
  "kind": "shootout",
  "status": "testing",
  "tier": "deep",
  "title": "Claude Code token-saver shootout",
  "category": "coding-agents/skills-plugins",
  "repos": [
   {
    "repo": "thedotmack/claude-mem",
    "stars": 101376,
    "license": "Apache-2.0"
   },
   {
    "repo": "mksglu/context-mode",
    "stars": 30474,
    "license": "NOASSERTION"
   },
   {
    "repo": "JuliusBrussee/caveman",
    "stars": 113243,
    "license": "Apache-2.0"
   },
   {
    "repo": "xerj-org/xerj",
    "stars": 7622,
    "license": "Apache-2.0"
   }
  ],
  "claims": [
   {
    "claim": "These repos stop Claude Code from burning through tokens and forgetting context",
    "source": "https://x.com/RoundtableSpace/status/2111483624679969197"
   },
   {
    "claim": "They actually reduce Claude Code token waste",
    "source": "https://x.com/RoundtableSpace/status/2110955662544982061"
   }
  ],
  "plan": "Run the same coding task 3 times with each tool and once with no tool, using the same model and budget. Measure tokens used, task success and wall time.",
  "who_its_for": "People who use Claude Code every day and pay for tokens.",
  "notes_so_far": [
   "gstack and Orca appear in some versions of this list but are parallel-agent tools, so they're tested in the parallel-agents shootout instead.",
   "One listed repo, anthropics/claude-code-mcp, returned 404 on 2026-10-17."
  ],
  "recycle_group": "\"Claude Code token-saver\" repo list",
  "topic": "coding-agents",
  "explainer": "Four add-ons that try to make AI coding agents like Claude Code waste fewer tokens. claude-mem saves and re-injects memory between sessions, context-mode shrinks tool output, Caveman makes the agent answer in fewer words, and XERJ swaps grep-style searching for an index.",
  "measure": [
   "Tokens used on the same coding task, with each tool and with no tool",
   "Whether the task still gets done correctly",
   "Wall-clock time per run",
   "How hard each one is to install"
  ],
  "sneak_peek": [],
  "gallery": []
 },
 {
  "slug": "lyogavin-airllm",
  "kind": "review",
  "status": "testing",
  "tier": "claim",
  "title": "AirLLM",
  "category": "local-models/runtimes",
  "repos": [
   {
    "repo": "lyogavin/airllm",
    "stars": 39145,
    "license": "Apache-2.0"
   }
  ],
  "claims": [
   {
    "claim": "Kimi K3, 2.8 trillion parameters, in 3.72 GB … 70B models on a single 4 GB card",
    "source": "https://x.com/RoundtableSpace/status/2111045739430269341"
   }
  ],
  "plan": "Run a 70B model with GPU memory capped at 4 GB. Measure whether it loads, memory used, and tokens per second.",
  "who_its_for": "Tinkerers who want to run big models on small GPUs.",
  "notes_so_far": [
   "The post doesn't mention speed, so Rocky will measure it."
  ],
  "topic": "ai-models",
  "explainer": "A Python library for running very large language models on a small GPU. Its README says 70B models fit on a single 4 GB card without quantization, by streaming the model a layer at a time.",
  "measure": [
   "Does a 70B model load with GPU memory capped at 4 GB?",
   "Peak GPU and system memory used",
   "Tokens per second (the post doesn't mention speed)"
  ],
  "sneak_peek": [],
  "gallery": []
 },
 {
  "slug": "morluto-rea",
  "kind": "review",
  "status": "testing",
  "tier": "deep",
  "title": "REA (reverse engineer anything)",
  "category": "coding-agents/skills-plugins",
  "repos": [
   {
    "repo": "morluto/rea",
    "stars": 304631,
    "license": "MIT"
   }
  ],
  "claims": [
   {
    "claim": "REA gives Claude Code or Codex the power to pull any app apart … 25K+ stars on GitHub in one day",
    "source": "https://x.com/humzaakhalid/status/2110305451565031610"
   }
  ],
  "plan": "Point it at open-source apps only and check whether its answers come with proof. Then try the rebuild claim on a small open-source app with an existing test suite.",
  "who_its_for": "Developers learning unfamiliar codebases.",
  "notes_so_far": [
   "Rocky only tests on open-source apps whose licenses allow it."
  ],
  "topic": "security",
  "explainer": "An MCP server that lets an AI agent take apart an app it doesn't have the source for (native binaries, Electron and JavaScript apps, .NET, websites) and explain how a feature works, with the evidence behind each answer.",
  "measure": [
   "Do its explanations of an open-source app match the real source code?",
   "Does every answer point to proof?",
   "Can it rebuild a small feature that passes the app's existing tests?"
  ],
  "sneak_peek": [],
  "gallery": []
 },
 {
  "slug": "parallel-coding-agents",
  "kind": "shootout",
  "status": "testing",
  "tier": "deep",
  "title": "Parallel AI coding agents shootout",
  "category": "agent-frameworks/orchestrators",
  "repos": [
   {
    "repo": "stablyai/orca",
    "stars": 95173,
    "license": "MIT"
   },
   {
    "repo": "ruvnet/ruflo",
    "stars": 75950,
    "license": "MIT"
   },
   {
    "repo": "OpenHands/OpenHands",
    "stars": null,
    "license": null
   },
   {
    "repo": "NousResearch/hermes-agent",
    "stars": 254625,
    "license": "MIT"
   },
   {
    "repo": "kdlbs/kandev",
    "stars": 963,
    "license": "AGPL-3.0"
   }
  ],
  "claims": [
   {
    "claim": "5 GitHub repos that let you run multiple AI coding agents in parallel",
    "source": "https://x.com/RoundtableSpace/status/2111181634821062671"
   },
   {
    "claim": "Run many AI coding tasks at once, then review all changes in one place",
    "source": "https://x.com/GithubProjects/status/2109708440235057612"
   }
  ],
  "plan": "Give every tool the same 3 independent issues on a sample repo, with the same model and budget. Measure correct merged changes, conflicts, cost and wall time.",
  "who_its_for": "Developers juggling several coding tasks at once.",
  "notes_so_far": [],
  "topic": "robotics",
  "explainer": "Tools that run several AI coding agents at the same time, each on its own task or git worktree, and then let you review all the changes in one place. Orca, Kandev and OpenHands are workspaces for this; Ruflo is a multi-agent harness; Hermes Agent is a self-improving agent.",
  "measure": [
   "Correct, mergeable changes on the same 3 independent issues",
   "Merge conflicts between agents",
   "Cost and wall-clock time",
   "Setup effort"
  ],
  "sneak_peek": [],
  "gallery": []
 },
 {
  "slug": "bherbruck-solidcraft",
  "kind": "review",
  "status": "testing",
  "tier": "claim",
  "title": "SolidCraft",
  "category": "cad-3d/cad-apps",
  "repos": [
   {
    "repo": "bherbruck/solidcraft",
    "stars": null,
    "license": "Apache-2.0"
   }
  ],
  "claims": [
   {
    "claim": "A free SolidWorks alternative that runs in your browser, and Claude Code can drive it … demo builds a full Raspberry Pi case",
    "source": "https://x.com/RoundtableSpace/status/2111053289127333933"
   }
  ],
  "plan": "Reproduce the Raspberry Pi case demo through its MCP server. Check that the STEP file is valid and the dimensions are right, then add it to the CAD shootout.",
  "who_its_for": "Hobby and maker CAD users who want agent-driven modeling.",
  "notes_so_far": [
   "The repo URL redirected on 2026-10-17 (moved or renamed). Rocky will test the canonical repo."
  ],
  "topic": "cad",
  "explainer": "An open-source parametric 3D CAD app written in Rust that runs on the desktop or in the browser. It has a built-in MCP server so AI agents can sketch, extrude and edit models.",
  "measure": [
   "Can an agent rebuild the README's Raspberry Pi 4 case through the MCP server?",
   "Is the exported STEP file valid?",
   "Are the key dimensions right (58 × 49 mm hole pattern)?"
  ],
  "sneak_peek": [],
  "gallery": []
 }
]