{
  "capabilities": [
    "tool:duckduckgo_search",
    "skill:research"
  ]
}

{
  "ok": true,
  "equipped": [
    {
      "capability": "tool:duckduckgo_search",
      "instruction": "Search the public internet with the `ddgs` library and return compact,\nstructured evidence.\n\nInput contract:\n{\n  \"query\": \"search query to execute\",\n  \"topic\": \"general|news\",\n  \"region\": \"us-en\",\n  \"safesearch\": \"on|moderate|off\",\n  \"timelimit\": \"d|w|m|y\",\n  \"max_results\": 5,\n  \"page\": 1,\n  \"backend\": \"auto\",\n  \"include_domains\": [\"example.com\"],\n  \"exclude_domains\": [\"example.com\"],\n  \"content_chars\": 700,\n  \"max_total_content_chars\": 3200\n}\n\nExecution guidance:\n- Use this tool for current web discovery, news lookup, quick verification, and\n  public source gathering.\n- Keep the goal search-only: find candidate URLs, titles, snippets, domains,\n  dates, and source evidence.\n- Do not use this tool to open pages interactively, click, inspect live UI,\n  fill forms, verify booking widgets, or capture screenshots.\n- If the task also needs live page inspection or a screenshot, return candidate\n  URLs and state that the next step should use a browser automation tool.\n- Prefer one focused search over many broad searches.\n- Use `topic: \"news\"` for current events and recent reporting.\n- Keep `backend: \"auto\"` unless you have a concrete reason to change it.\n- Use `include_domains` or `exclude_domains` when source boundaries matter; the\n  tool translates them into query operators.\n- Keep result budgets conservative unless wider coverage is genuinely needed.\n- Pair this tool with the `time` sense when the request depends on recency.\n\nOutput contract:\n{\n  \"ok\": true,\n  \"tool\": \"duckduckgo_search\",\n  \"query\": \"...\",\n  \"effective_query\": \"...\",\n  \"topic\": \"general\",\n  \"backend\": \"auto\",\n  \"results\": [\n    {\n      \"title\": \"...\",\n      \"url\": \"https://...\",\n      \"content\": \"...\",\n      \"published_date\": \"2026-05-06\",\n      \"source\": \"Example Source\"\n    }\n  ],\n  \"sources\": [\"https://...\"],\n  \"source_details\": [\n    {\n      \"title\": \"...\",\n      \"url\": \"https://...\",\n      \"published_date\": \"2026-05-06\",\n      \"source\": \"Example Source\"\n    }\n  ],\n  \"assessment\": {\n    \"coverage\": \"strong|partial|limited\",\n    \"source_count\": 3,\n    \"distinct_domains\": 2,\n    \"has_dates\": true,\n    \"evidence_note\": \"...\"\n  },\n  \"truthfulness_note\": \"...\"\n}"
    },
    {
      "capability": "skill:research",
      "instruction": "# Research Skill\n\nUse this skill when the task requires finding facts, comparing sources, checking recency, or producing source-grounded answers.\n\nResearch framework:\n\n1. Define the question, scope, and freshness requirements.\n2. Pick search or browsing tools based on whether broad discovery or page inspection is needed.\n3. Prefer primary or authoritative sources when accuracy matters.\n4. Track durable candidate findings, source quality, dates, and contradictions with memory updates when they should help future chats.\n5. Verify important claims with more than one source when feasible.\n6. Separate observed facts from inferences.\n7. Finish with a concise answer and links or file refs when available.\n\nFor Project Zuza:\n\n- Do not claim a search or page observation happened without a tool result.\n- Avoid sending screenshots to the model unless the visual content is necessary.\n- If a tool returns thin or stale evidence, change the query/source strategy before retrying.\n- If the needed research capability is not loaded, use Equip to load it before continuing.\n- If no existing tool can collect the needed evidence, plan a new local tool package under `tools/<tool_id>/` relative to the active runtime workspace, then use it only after discovery loads its contract."
    }
  ]
}

{
  "set": [
    {
      "id": "search",
      "status": "completed"
    },
    {
      "id": "compose",
      "status": "in_progress"
    }
  ]
}

{
  "ok": true,
  "state_changed": true,
  "changed_step_ids": [
    "search",
    "compose"
  ],
  "steps": [
    {
      "id": "search",
      "status": "completed"
    },
    {
      "id": "compose",
      "status": "in_progress"
    },
    {
      "id": "verify",
      "status": "planned"
    }
  ]
}