{
  "context": "The agent was given a Go microservice repository (Makefile, go.mod, README.md, CI config, INTERNAL_NOTES.md) and asked to create an AGENTS.md so AI assistants can help engineers effectively. The skill recommends four specific sections: 'Scope & routing', 'Non-discoverable commands', 'Landmines / do-not-touch areas', and 'Task-specific constraints'. This scenario tests whether the agent adopts that structure rather than producing a generic documentation file.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Non-discoverable commands section present",
      "description": "The output contains a section explicitly named 'Non-discoverable commands', 'Non-obvious commands', 'Hidden commands', or a closely equivalent heading that signals commands not obvious from the Makefile alone.",
      "max_score": 12
    },
    {
      "name": "Landmines / do-not-touch section present",
      "description": "The output contains a section named 'Landmines', 'Do-not-touch areas', 'Danger zones', or a closely equivalent heading that signals areas requiring special caution.",
      "max_score": 12
    },
    {
      "name": "Scope or routing or task-specific constraints section present",
      "description": "The output contains at least one of: a 'Scope & routing' section, a 'Task-specific constraints' section, or an equivalent section that scopes what the agent should or should not do in this repo.",
      "max_score": 8
    },
    {
      "name": "No generic documentation sections",
      "description": "The output does NOT contain sections named 'Overview', 'Architecture', 'Tech Stack', 'Directory Structure', 'Introduction', 'Getting Started', or other generic README-style headings that add no agent-specific value.",
      "max_score": 8
    },
    {
      "name": "make db-reset caveat captured",
      "description": "The commands section (or equivalent) explicitly warns that 'make db-reset' drops ALL data in the local dev database and reseeds — the dangerous side effect is called out, not just the command name listed.",
      "max_score": 12
    },
    {
      "name": "SKIP_AUTH=1 flag captured",
      "description": "The commands section (or equivalent) documents that 'make run' requires the SKIP_AUTH=1 environment variable in local dev, and explains why (auth middleware blocks all requests without it).",
      "max_score": 10
    },
    {
      "name": "billing/ package landmine captured",
      "description": "The landmines section (or equivalent) explicitly warns that the billing/ package is maintained by a separate team, that PRs touching billing/ must go through that team's review, and must NOT be merged by the main team.",
      "max_score": 12
    },
    {
      "name": "config/local.yaml auth URL landmine captured",
      "description": "The landmines section (or equivalent) warns that the internal auth service URL in config/local.yaml must not be changed to staging URLs in local dev.",
      "max_score": 6
    },
    {
      "name": "No generic development advice",
      "description": "The output does NOT contain generic advice such as 'write tests for your code', 'follow Go best practices', 'keep commits small', 'run linters before committing', or other boilerplate that would appear in any project's documentation.",
      "max_score": 8
    },
    {
      "name": "Sections are short and high-signal",
      "description": "Each section is concise — bullet points or short statements rather than multi-paragraph prose explanations. The file as a whole should be scannable in under 60 seconds and contain no filler sentences.",
      "max_score": 6
    },
    {
      "name": "Content is specific enough to execute",
      "description": "A developer or AI agent reading the file gains immediately actionable information: exact command invocations, exact package paths, exact file names. Vague references like 'some commands have side effects' or 'certain packages need care' do not satisfy this criterion.",
      "max_score": 6
    }
  ]
}