Skip to main content
Glama

Agent swarm (mixed Antigravity + Codex + Copilot + Cursor + …, parallel)

agent_swarm

Run many independent sub-tasks in parallel across multiple AI backends in one call, mixing model families and returning each labelled answer plus any failed worker in place.

Instructions

Best for many independent sub-tasks: runs them IN PARALLEL across backends in one call. Each task names its backend (Antigravity, Codex, Copilot, Cursor, Claude Code, Grok, opencode, Muse), so one swarm can mix model families; every answer comes back in one labelled block, and a failed worker is reported in place. Give each task workspace=; without it that worker sees no repo. Workers are one-shot (no continue). N autonomous agents at once is N times the prompt-injection surface: trusted prompts only.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "tasks": "<tasks>"
    },
    "name": "agent_swarm"
  }
}

Input Schema

NameRequiredDescriptionDefault
tasksYesOne object per worker. backend (required): "antigravity" ("agy"), "codex", "copilot", "cursor", "claude", "opencode", "grok" or "muse". prompt (required). workspace: the project dir (default: the server cwd, so no repo). sandbox: "read-only" (default), "workspace-write" or "danger-full-access", each with that backend's own meaning (see its *_ask). ANTIGRAVITY differs: "read-only" is plan mode, "workspace-write" is refused, and an omitted sandbox leaves the worker UNRESTRICTED, so set one. model: optional, in the backend's own format (opencode: "provider/model").
watchNoIf true, open the live "Agent Swarm" dashboard window (one card per worker, with its backend's logo; click a card for its full step log).
timeout_sNoPer-worker timeout in seconds. Default 180. An opencode worker is given at least 300s regardless, as a ceiling: its free models answer in seconds today but have been queue-scheduled to minutes before. The budget is only ever raised, never lowered.
max_concurrencyNoMax workers running at once (default 4). Higher = faster but more quota/rate-limit pressure and more agents at once.

Output Schema

NameRequiredDescriptionDefault
resultYes

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed2 schema fields changedv0.37.1
    • changedInput schema / properties / tasks / description
      Previous value: -"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", \"opencode\"\n              (alias \"oc\" — the one backend that needs no\n              subscription; see opencode_ask), \"grok\" (alias\n              \"xai\"; EXPERIMENTAL — see grok_ask), or \"muse\" (alias\n              \"meta\"; EXPERIMENTAL — see muse_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: \"read-only\" (default), \"workspace-write\", or\n              \"danger-full-access\". Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's, Cursor's and opencode's are agent/tool-level,\n              not OS boundaries; Muse's read-only switches its write,\n              shell and web tools off — see copilot_ask / cursor_ask /\n              grok_ask / opencode_ask / muse_ask.\n              ANTIGRAVITY is the odd one: \"read-only\" maps to agy's\n              plan mode (it investigates and writes a plan instead of\n              editing files or running commands — see antigravity_ask's\n              `plan`, and note it is agent-enforced, and needs agy\n              1.1.12+), \"danger-full-access\" states plainly that the\n              worker is unrestricted, and \"workspace-write\" is REFUSED\n              because agy has no write scoping to offer. Omitting it\n              leaves an Antigravity worker unrestricted — that is the\n              long-standing default, unlike every other backend here,\n              so fence it explicitly if you want it fenced.\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's/Muse's `--model`, Grok's/opencode's `-m`\n              (opencode wants \"provider/model\"), or Antigravity's\n              `--model` (an agy slug like \"claude-sonnet-5-5-high\";\n              validated against each backend's model list). Omit for\n              each backend's default."New value: +"One object per worker. backend (required): \"antigravity\" (\"agy\"),\n   \"codex\", \"copilot\", \"cursor\", \"claude\", \"opencode\", \"grok\" or\n   \"muse\". prompt (required). workspace: the project dir (default: the\n   server cwd, so no repo). sandbox: \"read-only\" (default),\n   \"workspace-write\" or \"danger-full-access\", each with that backend's\n   own meaning (see its *_ask). ANTIGRAVITY differs: \"read-only\" is plan\n   mode, \"workspace-write\" is refused, and an omitted sandbox leaves the\n   worker UNRESTRICTED, so set one. model: optional, in the backend's\n   own format (opencode: \"provider/model\")."
    • changedInput schema / properties / timeout_s / description
      Previous value: -"Per-worker timeout in seconds. Default 180. An opencode\n       worker is given at least 300s regardless — its free models are\n       queue-scheduled and were measured at 152-428s, so the shared\n       default would kill about half of them mid-answer and report a\n       slow worker as a broken one. The budget is only ever raised,\n       never lowered."New value: +"Per-worker timeout in seconds. Default 180. An opencode\n       worker is given at least 300s regardless, as a ceiling: its free\n       models answer in seconds today but have been queue-scheduled to\n       minutes before. The budget is only ever raised, never lowered."
  2. Changed1 schema field changedv0.33.0
    • changedInput schema / properties / tasks / description
      Previous value: -"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", \"opencode\"\n              (alias \"oc\" — the one backend that needs no\n              subscription; see opencode_ask), \"grok\" (alias\n              \"xai\"; EXPERIMENTAL — see grok_ask), or \"muse\" (alias\n              \"meta\"; EXPERIMENTAL — see muse_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: \"read-only\" (default), \"workspace-write\", or\n              \"danger-full-access\". Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's, Cursor's and opencode's are agent/tool-level,\n              not OS boundaries; Muse's read-only switches its write,\n              shell and web tools off — see copilot_ask / cursor_ask /\n              grok_ask / opencode_ask / muse_ask.\n              ANTIGRAVITY is the odd one: \"read-only\" maps to agy's\n              plan mode (it investigates and writes a plan instead of\n              editing files or running commands — see antigravity_ask's\n              `plan`, and note it is agent-enforced, and needs agy\n              1.1.12+), \"danger-full-access\" states plainly that the\n              worker is unrestricted, and \"workspace-write\" is REFUSED\n              because agy has no write scoping to offer. Omitting it\n              leaves an Antigravity worker unrestricted — that is the\n              long-standing default, unlike every other backend here,\n              so fence it explicitly if you want it fenced.\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's/Muse's `--model`, Grok's/opencode's `-m`\n              (opencode wants \"provider/model\"), or Antigravity's\n              `--model` (an agy slug like \"claude-sonnet-4-6\";\n              validated against each backend's model list). Omit for\n              each backend's default."New value: +"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", \"opencode\"\n              (alias \"oc\" — the one backend that needs no\n              subscription; see opencode_ask), \"grok\" (alias\n              \"xai\"; EXPERIMENTAL — see grok_ask), or \"muse\" (alias\n              \"meta\"; EXPERIMENTAL — see muse_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: \"read-only\" (default), \"workspace-write\", or\n              \"danger-full-access\". Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's, Cursor's and opencode's are agent/tool-level,\n              not OS boundaries; Muse's read-only switches its write,\n              shell and web tools off — see copilot_ask / cursor_ask /\n              grok_ask / opencode_ask / muse_ask.\n              ANTIGRAVITY is the odd one: \"read-only\" maps to agy's\n              plan mode (it investigates and writes a plan instead of\n              editing files or running commands — see antigravity_ask's\n              `plan`, and note it is agent-enforced, and needs agy\n              1.1.12+), \"danger-full-access\" states plainly that the\n              worker is unrestricted, and \"workspace-write\" is REFUSED\n              because agy has no write scoping to offer. Omitting it\n              leaves an Antigravity worker unrestricted — that is the\n              long-standing default, unlike every other backend here,\n              so fence it explicitly if you want it fenced.\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's/Muse's `--model`, Grok's/opencode's `-m`\n              (opencode wants \"provider/model\"), or Antigravity's\n              `--model` (an agy slug like \"claude-sonnet-5-5-high\";\n              validated against each backend's model list). Omit for\n              each backend's default."
  3. Changed2 schema fields changedv0.32.1
    • changedInput schema / properties / tasks / description
      Previous value: -"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", \"opencode\"\n              (alias \"oc\" — the one backend that needs no\n              subscription; see opencode_ask), or \"grok\" (alias\n              \"xai\"; EXPERIMENTAL — see grok_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: \"read-only\" (default), \"workspace-write\", or\n              \"danger-full-access\". Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's, Cursor's and opencode's are agent/tool-level,\n              not OS boundaries — see copilot_ask / cursor_ask /\n              grok_ask / opencode_ask.\n              ANTIGRAVITY is the odd one: \"read-only\" maps to agy's\n              plan mode (it investigates and writes a plan instead of\n              editing files or running commands — see antigravity_ask's\n              `plan`, and note it is agent-enforced, and needs agy\n              1.1.12+), \"danger-full-access\" states plainly that the\n              worker is unrestricted, and \"workspace-write\" is REFUSED\n              because agy has no write scoping to offer. Omitting it\n              leaves an Antigravity worker unrestricted — that is the\n              long-standing default, unlike every other backend here,\n              so fence it explicitly if you want it fenced.\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, Grok's/opencode's `-m`\n              (opencode wants \"provider/model\"), or Antigravity's\n              `--model` (an agy slug like \"claude-sonnet-4-6\";\n              validated against each backend's model list). Omit for\n              each backend's default."New value: +"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", \"opencode\"\n              (alias \"oc\" — the one backend that needs no\n              subscription; see opencode_ask), \"grok\" (alias\n              \"xai\"; EXPERIMENTAL — see grok_ask), or \"muse\" (alias\n              \"meta\"; EXPERIMENTAL — see muse_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: \"read-only\" (default), \"workspace-write\", or\n              \"danger-full-access\". Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's, Cursor's and opencode's are agent/tool-level,\n              not OS boundaries; Muse's read-only switches its write,\n              shell and web tools off — see copilot_ask / cursor_ask /\n              grok_ask / opencode_ask / muse_ask.\n              ANTIGRAVITY is the odd one: \"read-only\" maps to agy's\n              plan mode (it investigates and writes a plan instead of\n              editing files or running commands — see antigravity_ask's\n              `plan`, and note it is agent-enforced, and needs agy\n              1.1.12+), \"danger-full-access\" states plainly that the\n              worker is unrestricted, and \"workspace-write\" is REFUSED\n              because agy has no write scoping to offer. Omitting it\n              leaves an Antigravity worker unrestricted — that is the\n              long-standing default, unlike every other backend here,\n              so fence it explicitly if you want it fenced.\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's/Muse's `--model`, Grok's/opencode's `-m`\n              (opencode wants \"provider/model\"), or Antigravity's\n              `--model` (an agy slug like \"claude-sonnet-4-6\";\n              validated against each backend's model list). Omit for\n              each backend's default."
    • changedInput schema / properties / watch / description
      Previous value: -"If true, open the live \"Agent Swarm\" dashboard window (one row per\n   worker, with a backend badge; click a row for its full step log)."New value: +"If true, open the live \"Agent Swarm\" dashboard window (one card per\n   worker, with its backend's logo; click a card for its full step log)."
  4. Changed2 schema fields changedv0.30.0
    • changedInput schema / properties / tasks / description
      Previous value: -"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", or \"grok\"\n              (alias \"xai\"; EXPERIMENTAL — see grok_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: \"read-only\" (default), \"workspace-write\", or\n              \"danger-full-access\". Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's and Cursor's are agent/tool-level, not OS\n              boundaries — see copilot_ask / cursor_ask / grok_ask.\n              ANTIGRAVITY is the odd one: \"read-only\" maps to agy's\n              plan mode (it investigates and writes a plan instead of\n              editing files or running commands — see antigravity_ask's\n              `plan`, and note it is agent-enforced, and needs agy\n              1.1.12+), \"danger-full-access\" states plainly that the\n              worker is unrestricted, and \"workspace-write\" is REFUSED\n              because agy has no write scoping to offer. Omitting it\n              leaves an Antigravity worker unrestricted — that is the\n              long-standing default, unlike every other backend here,\n              so fence it explicitly if you want it fenced.\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, Grok's `-m`, or\n              Antigravity's `--model` (an agy slug like\n              \"claude-sonnet-4-6\"; validated against each backend's\n              model list). Omit for each backend's default."New value: +"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", \"opencode\"\n              (alias \"oc\" — the one backend that needs no\n              subscription; see opencode_ask), or \"grok\" (alias\n              \"xai\"; EXPERIMENTAL — see grok_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: \"read-only\" (default), \"workspace-write\", or\n              \"danger-full-access\". Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's, Cursor's and opencode's are agent/tool-level,\n              not OS boundaries — see copilot_ask / cursor_ask /\n              grok_ask / opencode_ask.\n              ANTIGRAVITY is the odd one: \"read-only\" maps to agy's\n              plan mode (it investigates and writes a plan instead of\n              editing files or running commands — see antigravity_ask's\n              `plan`, and note it is agent-enforced, and needs agy\n              1.1.12+), \"danger-full-access\" states plainly that the\n              worker is unrestricted, and \"workspace-write\" is REFUSED\n              because agy has no write scoping to offer. Omitting it\n              leaves an Antigravity worker unrestricted — that is the\n              long-standing default, unlike every other backend here,\n              so fence it explicitly if you want it fenced.\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, Grok's/opencode's `-m`\n              (opencode wants \"provider/model\"), or Antigravity's\n              `--model` (an agy slug like \"claude-sonnet-4-6\";\n              validated against each backend's model list). Omit for\n              each backend's default."
    • changedInput schema / properties / timeout_s / description
      Previous value: -"Per-worker timeout in seconds. Default 180."New value: +"Per-worker timeout in seconds. Default 180. An opencode\n       worker is given at least 300s regardless — its free models are\n       queue-scheduled and were measured at 152-428s, so the shared\n       default would kill about half of them mid-answer and report a\n       slow worker as a broken one. The budget is only ever raised,\n       never lowered."
  5. Changed1 schema field changedv0.28.0
    • changedInput schema / properties / tasks / description
      Previous value: -"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", or \"grok\"\n              (alias \"xai\"; EXPERIMENTAL — see grok_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: Codex/Copilot/Cursor/Grok only — \"read-only\" (default),\n              \"workspace-write\", or \"danger-full-access\". Ignored for\n              Antigravity. (Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's and Cursor's are agent/tool-level, not OS\n              boundaries — see copilot_ask / cursor_ask / grok_ask.)\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, Grok's `-m`, or\n              Antigravity's `--model` (an agy slug like\n              \"claude-sonnet-4-6\"; validated against each backend's\n              model list). Omit for each backend's default."New value: +"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", or \"grok\"\n              (alias \"xai\"; EXPERIMENTAL — see grok_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: \"read-only\" (default), \"workspace-write\", or\n              \"danger-full-access\". Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's and Cursor's are agent/tool-level, not OS\n              boundaries — see copilot_ask / cursor_ask / grok_ask.\n              ANTIGRAVITY is the odd one: \"read-only\" maps to agy's\n              plan mode (it investigates and writes a plan instead of\n              editing files or running commands — see antigravity_ask's\n              `plan`, and note it is agent-enforced, and needs agy\n              1.1.12+), \"danger-full-access\" states plainly that the\n              worker is unrestricted, and \"workspace-write\" is REFUSED\n              because agy has no write scoping to offer. Omitting it\n              leaves an Antigravity worker unrestricted — that is the\n              long-standing default, unlike every other backend here,\n              so fence it explicitly if you want it fenced.\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, Grok's `-m`, or\n              Antigravity's `--model` (an agy slug like\n              \"claude-sonnet-4-6\"; validated against each backend's\n              model list). Omit for each backend's default."
  6. Changed1 schema field changedv0.26.0
    • changedInput schema / properties / tasks / description
      Previous value: -"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), or \"cursor\" (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: Codex/Copilot/Cursor only — \"read-only\" (default),\n              \"workspace-write\", or \"danger-full-access\". Ignored for\n              Antigravity. (Codex's is an enforced OS sandbox; Copilot's\n              and Cursor's are agent/tool-level, not OS boundaries — see\n              copilot_ask / cursor_ask.)\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, or Antigravity's `--model`\n              (an agy slug like \"claude-sonnet-4-6\"; validated\n              against each backend's model list). Omit for each\n              backend's default."New value: +"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), \"cursor\", or \"grok\"\n              (alias \"xai\"; EXPERIMENTAL — see grok_ask) (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: Codex/Copilot/Cursor/Grok only — \"read-only\" (default),\n              \"workspace-write\", or \"danger-full-access\". Ignored for\n              Antigravity. (Codex's is an enforced OS sandbox\n              everywhere; Grok's is enforced on Linux/macOS only;\n              Copilot's and Cursor's are agent/tool-level, not OS\n              boundaries — see copilot_ask / cursor_ask / grok_ask.)\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, Grok's `-m`, or\n              Antigravity's `--model` (an agy slug like\n              \"claude-sonnet-4-6\"; validated against each backend's\n              model list). Omit for each backend's default."
  7. Changed1 schema field changedv0.21.4
    • changedInput schema / properties / tasks / description
      Previous value: -"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), or \"cursor\" (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: Codex/Copilot/Cursor only — \"read-only\" (default),\n              \"workspace-write\", or \"danger-full-access\". Ignored for\n              Antigravity. (Codex's is an enforced OS sandbox; Copilot's\n              and Cursor's are agent/tool-level, not OS boundaries — see\n              copilot_ask / cursor_ask.)\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, or Antigravity's `--model`\n              (an agy label like \"Claude Sonnet 4.6 (Thinking)\";\n              validated against each backend's model list). Omit for\n              each backend's default."New value: +"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), or \"cursor\" (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: Codex/Copilot/Cursor only — \"read-only\" (default),\n              \"workspace-write\", or \"danger-full-access\". Ignored for\n              Antigravity. (Codex's is an enforced OS sandbox; Copilot's\n              and Cursor's are agent/tool-level, not OS boundaries — see\n              copilot_ask / cursor_ask.)\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, or Antigravity's `--model`\n              (an agy slug like \"claude-sonnet-4-6\"; validated\n              against each backend's model list). Omit for each\n              backend's default."
  8. Changed1 schema field changedv0.21.0
    • changedInput schema / properties / tasks / description
      Previous value: -"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\", or\n              \"copilot\" (alias \"gh\"/\"github\") (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: Codex/Copilot only — \"read-only\" (default),\n              \"workspace-write\", or \"danger-full-access\". Ignored for\n              Antigravity. (Codex's is an enforced OS sandbox; Copilot's\n              is best-effort tool/path permissions — see copilot_ask.)\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's `--model`, or Antigravity's `--model` (an agy\n              label like \"Claude Sonnet 4.6 (Thinking)\"; validated\n              against `agy models`). Omit for each backend's default."New value: +"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\",\n              \"copilot\" (alias \"gh\"/\"github\"), or \"cursor\" (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: Codex/Copilot/Cursor only — \"read-only\" (default),\n              \"workspace-write\", or \"danger-full-access\". Ignored for\n              Antigravity. (Codex's is an enforced OS sandbox; Copilot's\n              and Cursor's are agent/tool-level, not OS boundaries — see\n              copilot_ask / cursor_ask.)\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's/Cursor's `--model`, or Antigravity's `--model`\n              (an agy label like \"Claude Sonnet 4.6 (Thinking)\";\n              validated against each backend's model list). Omit for\n              each backend's default."
  9. Changed1 schema field changedv0.15.4
    • changedInput schema / properties / tasks / description
      Previous value: -"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\") or \"codex\" (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: Codex only — \"read-only\" (default), \"workspace-write\",\n              or \"danger-full-access\". Ignored for Antigravity.\n   - model:   Codex only — model override (`-m`). Ignored for Antigravity."New value: +"One object per parallel worker:\n   - backend: \"antigravity\" (alias \"agy\"/\"gemini\"), \"codex\", or\n              \"copilot\" (alias \"gh\"/\"github\") (required)\n   - prompt:  the question or instruction (required)\n   - workspace: working dir for that worker (default: server cwd)\n   - sandbox: Codex/Copilot only — \"read-only\" (default),\n              \"workspace-write\", or \"danger-full-access\". Ignored for\n              Antigravity. (Codex's is an enforced OS sandbox; Copilot's\n              is best-effort tool/path permissions — see copilot_ask.)\n   - model:   optional model override for ANY backend — Codex's `-m`,\n              Copilot's `--model`, or Antigravity's `--model` (an agy\n              label like \"Claude Sonnet 4.6 (Thinking)\"; validated\n              against `agy models`). Omit for each backend's default."
  10. Addedv0.12.2

TDQS

A4.6/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations only say readOnlyHint=false, openWorldHint=true, idempotentHint=false; the description adds far more — results returned in one labelled block, a failed worker reported in place rather than aborting the swarm, workers being one-shot, and a prompt-injection risk warning that scales with N. The workspace default ('no repo') and the Antigravity sandbox caveat are exactly the non-obvious behaviors an agent would otherwise get wrong.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with the core use case, then constraints in descending priority; nearly every clause carries information (backend list, one-shot, workspace, injection risk). The single dense paragraph and mid-sentence parenthetical enumeration make it slightly harder to scan than a short bulleted form.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Covers the decision-relevant surface for a parallel multi-backend fan-out tool: when to use it, failure isolation, workspace requirement, sandbox semantics, and safety exposure. Return values are covered by the output schema, so nothing essential is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the baseline is 3, but the description adds operational meaning: omitting workspace means that worker sees no repo, and N agents means N times the injection surface, framing the concurrency/sandbox choices as risk decisions rather than plain fields. Timeout and watch behavior are left entirely to the schema, which is acceptable.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and scope: 'runs them IN PARALLEL across backends in one call', with each task naming its own backend so a single swarm mixes model families. This clearly separates it from the single-backend *_ask siblings and from preset_swarm without needing either schema opened.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

'Best for many independent sub-tasks' gives an explicit selection condition and implicitly excludes dependent/sequential work. 'Workers are one-shot (no continue)' steers follow-up work toward the *_continue siblings, though no sibling is named outright, which keeps it just short of a 5.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.