Skip to main content
Glama

Ask Antigravity (new conversation)

antigravity_ask

Ask Gemini via Antigravity CLI a question or short task in a new conversation, with an optional workspace for repo context and follow-ups through antigravity_continue.

Instructions

Best for fast, cheap tool-calling and short tasks on Gemini (agy CLI), in a NEW conversation. Pass workspace=; without it the sub-agent sees no repo. Follow up with antigravity_continue. Uses your AI Pro login; pick a bigger model for heavier reasoning. No real sandbox (plan=True is agent-enforced read-only). Autonomous agent: 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": {
      "prompt": "<prompt>"
    },
    "name": "antigravity_ask"
  }
}

Input Schema

NameRequiredDescriptionDefault
planNoIf true, run agy in PLAN mode (agy 1.1.12+): it reads and answers, and diverts file writes and shell commands into a plan document instead (verified even when the prompt insists). Use it to point Antigravity at a repo you don't want edited. Agent-enforced, not an OS sandbox (codex has the real one). A prompt whose first token is a slash command is rejected in plan mode, and an agy older than 1.1.12 raises rather than running unrestricted. Default false.
modelNoOptional model slug (agy's --model), e.g. "gemini-4-argon-high" or "gemini-3.8-flash-low", or one without its effort suffix ("gemini-4-argon"). Omit for agy's settings.json default (Gemini 4 Argon (Medium) as of agy 1.5.1). Checked against `agy models`: an unknown slug is rejected up front, since agy renames and drops models between releases. Quota is per model family; a quota failure names a family that still has some.
watchNoIf true, open a live "watch" view in your browser that streams agy's steps (narration + the real commands it runs) as it works. agy still runs headless; the same final text is returned. Best- effort and cross-platform — if the browser can't open, the run completes normally. Default false.
promptYesQuestion or instruction for Antigravity.
schemaNoOptional JSON Schema (an object, or its JSON text). Returns agy's VALIDATED object (`structured_output`) as JSON text instead of prose; json.loads it. Raises if agy produces none. Needs agy 1.1.8+. IMPORTANT: write the prompt so the ANSWER is worked out in the turn, and let the schema only shape it. agy fills the schema in a final pass that does not re-reason, so a judgment left to the schema gets guessed (measured: flipping an enum's order flipped the verdict). Extraction is reliable; delegated judgment is not.
timeout_sNoMax seconds to wait for agy to complete. Default 180.
workspaceNoWorking directory for the conversation. Defaults to cwd. Choose an existing project dir for context-aware responses.

Output Schema

NameRequiredDescriptionDefault
resultYes

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed3 schema fields changedv0.37.1
    • changedInput schema / properties / model / description
      Previous value: -"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"gemini-4-argon-high\" or \"gemini-3.8-flash-low\". Omit to use\n   the model set in agy's settings.json, or agy's own default when\n   none is set (Gemini 4 Argon (Medium) as of agy 1.4.12). Must be one\n   of `agy models` — an unknown slug is rejected up front. agy renames\n   and drops models between releases, often with no changelog entry,\n   so a slug from older docs may be gone; see antigravity_status /\n   `agy models` for the valid ones. Quota is counted per model family:\n   if a call fails on quota, the error lists each family's quota and a\n   model from a family that still has some."New value: +"Optional model slug (agy's --model), e.g. \"gemini-4-argon-high\" or\n   \"gemini-3.8-flash-low\", or one without its effort suffix\n   (\"gemini-4-argon\"). Omit for agy's settings.json default (Gemini 4\n   Argon (Medium) as of agy 1.5.1). Checked against `agy models`: an\n   unknown slug is rejected up front, since agy renames and drops models\n   between releases. Quota is per model family; a quota failure names a\n   family that still has some."
    • changedInput schema / properties / plan / description
      Previous value: -"If true, run agy in PLAN mode (agy 1.1.12+): it investigates and\n  writes an implementation plan instead of touching anything. Verified\n  on 1.1.20 that a file write and a shell command are both refused and\n  diverted into a plan document under agy's own directory — even when\n  the prompt insists, and even though the bridge still passes\n  --dangerously-skip-permissions — while file READS answer normally.\n  Use it to point Antigravity at a repo you don't want it editing.\n  Two caveats. It is agent-enforced, not an OS sandbox: it constrains\n  agy's agent loop, so treat it as a strong default rather than a\n  boundary you'd rely on against a hostile prompt (Codex has the real\n  one — see codex_ask's sandbox, and its Windows caveat: as of codex\n  0.162.1 a sandboxed run there refuses every command and answers\n  anyway). And it is exclusive with the bridge's\n  slash-command shield, because agy silently disables plan mode when\n  that shield is on; a prompt whose first token is a slash command is\n  therefore rejected up front rather than run. Raises on agy older than\n  1.1.12, which ignores --mode in print mode, rather than silently\n  running your prompt unrestricted. Default false."New value: +"If true, run agy in PLAN mode (agy 1.1.12+): it reads and answers, and\n  diverts file writes and shell commands into a plan document instead\n  (verified even when the prompt insists). Use it to point Antigravity\n  at a repo you don't want edited. Agent-enforced, not an OS sandbox\n  (codex has the real one). A prompt whose first token is a slash\n  command is rejected in plan mode, and an agy older than 1.1.12 raises\n  rather than running unrestricted. Default false."
    • changedInput schema / properties / schema / description
      Previous value: -"Optional JSON Schema (an object, or its JSON text). When given, agy\n    is asked to produce output matching it (agy 1.1.8's --json-schema)\n    and this tool returns the VALIDATED OBJECT as JSON text instead of\n    prose — json.loads it. What comes back is agy's own\n    `structured_output`, which carries exactly the declared fields;\n    agy's prose `response` on the same run also picks up its internal\n    toolAction/toolSummary keys and can be prefixed with a sentence, so\n    the two are NOT interchangeable. If agy produces no structured\n    output the call RAISES rather than handing back prose you would\n    have to parse anyway. Needs agy 1.1.8+.\n\n    IMPORTANT — write the prompt so the ANSWER is in the turn, and let\n    the schema only shape it. agy fills the schema in a finishing pass\n    that does not re-reason about the content, so a field the turn never\n    established gets guessed from the schema itself. Measured on 1.1.20\n    with \"this broke my build and wasted my whole afternoon\": with\n    enum [\"positive\",\"negative\"] it answered \"positive\" 3 times out of 4,\n    and simply REVERSING the enum to [\"negative\",\"positive\"] flipped it\n    to \"negative\" 2 out of 2 — it was following field order, not the\n    sentence. Adding a `reason` field did not help; the reason came back\n    \"Completed sentiment classification task.\" Asking the prompt to state\n    the verdict and why, and keeping the same biased enum, was correct\n    3 out of 3. So: extraction of what the model has already worked out\n    is reliable; a judgment delegated to the schema is not."New value: +"Optional JSON Schema (an object, or its JSON text). Returns agy's\n    VALIDATED object (`structured_output`) as JSON text instead of prose;\n    json.loads it. Raises if agy produces none. Needs agy 1.1.8+.\n    IMPORTANT: write the prompt so the ANSWER is worked out in the turn,\n    and let the schema only shape it. agy fills the schema in a final\n    pass that does not re-reason, so a judgment left to the schema gets\n    guessed (measured: flipping an enum's order flipped the verdict).\n    Extraction is reliable; delegated judgment is not."
  2. Changed1 schema field changedv0.34.0
    • changedInput schema / properties / model / description
      Previous value: -"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"claude-sonnet-5-5-high\" or \"gemini-3.8-flash-low\". Omit to use\n   the model set in agy's settings.json (gemini-3.8-flash-high as of\n   agy 1.3.5). Must be one of `agy models` — an unknown slug is\n   rejected up front. agy renames and drops models between releases,\n   often with no changelog entry, so a slug from older docs may be\n   gone; see antigravity_status / `agy models` for the valid ones."New value: +"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"gemini-4-argon-high\" or \"gemini-3.8-flash-low\". Omit to use\n   the model set in agy's settings.json, or agy's own default when\n   none is set (Gemini 4 Argon (Medium) as of agy 1.4.12). Must be one\n   of `agy models` — an unknown slug is rejected up front. agy renames\n   and drops models between releases, often with no changelog entry,\n   so a slug from older docs may be gone; see antigravity_status /\n   `agy models` for the valid ones. Quota is counted per model family:\n   if a call fails on quota, the error lists each family's quota and a\n   model from a family that still has some."
  3. Changed2 schema fields changedv0.33.0
    • changedInput schema / properties / model / description
      Previous value: -"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"gemini-3.1-pro-high\" or \"claude-sonnet-4-6\". Omit to use the\n   model set in agy's settings.json (gemini-3.8-flash-high as of agy\n   1.1.25). Must be one of `agy models` — an unknown slug is\n   rejected up front (agy would otherwise silently ignore it and fall\n   back to the default). agy 1.1.5 replaced the old human labels\n   (\"Gemini 3.1 Pro (High)\") with these slugs, and the default has\n   since moved to the gemini-3.8-flash family; the old form is no\n   longer accepted. Note 1.1.25 also DROPPED the gemini-3.5-flash\n   family with no changelog entry, so a 3.5 slug you saw in older\n   docs is now rejected. See antigravity_status / `agy models` for\n   the valid slugs."New value: +"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"claude-sonnet-5-5-high\" or \"gemini-3.8-flash-low\". Omit to use\n   the model set in agy's settings.json (gemini-3.8-flash-high as of\n   agy 1.3.5). Must be one of `agy models` — an unknown slug is\n   rejected up front. agy renames and drops models between releases,\n   often with no changelog entry, so a slug from older docs may be\n   gone; see antigravity_status / `agy models` for the valid ones."
    • changedInput schema / properties / plan / description
      Previous value: -"If true, run agy in PLAN mode (agy 1.1.12+): it investigates and\n  writes an implementation plan instead of touching anything. Verified\n  on 1.1.20 that a file write and a shell command are both refused and\n  diverted into a plan document under agy's own directory — even when\n  the prompt insists, and even though the bridge still passes\n  --dangerously-skip-permissions — while file READS answer normally.\n  Use it to point Antigravity at a repo you don't want it editing.\n  Two caveats. It is agent-enforced, not an OS sandbox: it constrains\n  agy's agent loop, so treat it as a strong default rather than a\n  boundary you'd rely on against a hostile prompt (Codex has the real\n  one — see codex_ask's sandbox, and its Windows caveat: as of codex\n  0.149.1 a sandboxed run there refuses every command and answers\n  anyway). And it is exclusive with the bridge's\n  slash-command shield, because agy silently disables plan mode when\n  that shield is on; a prompt whose first token is a slash command is\n  therefore rejected up front rather than run. Raises on agy older than\n  1.1.12, which ignores --mode in print mode, rather than silently\n  running your prompt unrestricted. Default false."New value: +"If true, run agy in PLAN mode (agy 1.1.12+): it investigates and\n  writes an implementation plan instead of touching anything. Verified\n  on 1.1.20 that a file write and a shell command are both refused and\n  diverted into a plan document under agy's own directory — even when\n  the prompt insists, and even though the bridge still passes\n  --dangerously-skip-permissions — while file READS answer normally.\n  Use it to point Antigravity at a repo you don't want it editing.\n  Two caveats. It is agent-enforced, not an OS sandbox: it constrains\n  agy's agent loop, so treat it as a strong default rather than a\n  boundary you'd rely on against a hostile prompt (Codex has the real\n  one — see codex_ask's sandbox, and its Windows caveat: as of codex\n  0.162.1 a sandboxed run there refuses every command and answers\n  anyway). And it is exclusive with the bridge's\n  slash-command shield, because agy silently disables plan mode when\n  that shield is on; a prompt whose first token is a slash command is\n  therefore rejected up front rather than run. Raises on agy older than\n  1.1.12, which ignores --mode in print mode, rather than silently\n  running your prompt unrestricted. Default false."
  4. Changed1 schema field changedv0.29.1
    • changedInput schema / properties / model / description
      Previous value: -"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"gemini-3.1-pro-high\" or \"claude-sonnet-4-6\". Omit to use the\n   model set in agy's settings.json (gemini-3.7-flash-high by\n   default). Must be one of `agy models` — an unknown slug is\n   rejected up front (agy would otherwise silently ignore it and fall\n   back to the default). agy 1.1.5 replaced the old human labels\n   (\"Gemini 3.1 Pro (High)\") with these slugs, and the default has\n   since moved to the gemini-3.7-flash family; the old form is no\n   longer accepted. See antigravity_status / `agy models` for the\n   valid slugs."New value: +"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"gemini-3.1-pro-high\" or \"claude-sonnet-4-6\". Omit to use the\n   model set in agy's settings.json (gemini-3.8-flash-high as of agy\n   1.1.25). Must be one of `agy models` — an unknown slug is\n   rejected up front (agy would otherwise silently ignore it and fall\n   back to the default). agy 1.1.5 replaced the old human labels\n   (\"Gemini 3.1 Pro (High)\") with these slugs, and the default has\n   since moved to the gemini-3.8-flash family; the old form is no\n   longer accepted. Note 1.1.25 also DROPPED the gemini-3.5-flash\n   family with no changelog entry, so a 3.5 slug you saw in older\n   docs is now rejected. See antigravity_status / `agy models` for\n   the valid slugs."
  5. Changed3 schema fields changedv0.28.0
    • changedInput schema / properties / model / description
      Previous value: -"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"gemini-3.1-pro-high\" or \"claude-sonnet-4-6\". Omit to use the\n   model set in agy's settings.json (gemini-3.6-flash-high by\n   default). Must be one of `agy models` — an unknown slug is\n   rejected up front (agy would otherwise silently ignore it and fall\n   back to the default). agy 1.1.5 replaced the old human labels\n   (\"Gemini 3.1 Pro (High)\") with these slugs and 1.1.6 added the\n   gemini-3.6-flash family; the old form is no longer accepted. See\n   antigravity_status / `agy models` for the valid slugs."New value: +"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"gemini-3.1-pro-high\" or \"claude-sonnet-4-6\". Omit to use the\n   model set in agy's settings.json (gemini-3.7-flash-high by\n   default). Must be one of `agy models` — an unknown slug is\n   rejected up front (agy would otherwise silently ignore it and fall\n   back to the default). agy 1.1.5 replaced the old human labels\n   (\"Gemini 3.1 Pro (High)\") with these slugs, and the default has\n   since moved to the gemini-3.7-flash family; the old form is no\n   longer accepted. See antigravity_status / `agy models` for the\n   valid slugs."
    • addedInput schema / properties / plan
      Added value: +{
      +  "default": false,
      +  "description": "If true, run agy in PLAN mode (agy 1.1.12+): it investigates and\n  writes an implementation plan instead of touching anything. Verified\n  on 1.1.20 that a file write and a shell command are both refused and\n  diverted into a plan document under agy's own directory — even when\n  the prompt insists, and even though the bridge still passes\n  --dangerously-skip-permissions — while file READS answer normally.\n  Use it to point Antigravity at a repo you don't want it editing.\n  Two caveats. It is agent-enforced, not an OS sandbox: it constrains\n  agy's agent loop, so treat it as a strong default rather than a\n  boundary you'd rely on against a hostile prompt (Codex has the real\n  one — see codex_ask's sandbox, and its Windows caveat: as of codex\n  0.149.1 a sandboxed run there refuses every command and answers\n  anyway). And it is exclusive with the bridge's\n  slash-command shield, because agy silently disables plan mode when\n  that shield is on; a prompt whose first token is a slash command is\n  therefore rejected up front rather than run. Raises on agy older than\n  1.1.12, which ignores --mode in print mode, rather than silently\n  running your prompt unrestricted. Default false.",
      +  "type": "boolean"
      +}
    • addedInput schema / properties / schema
      Added value: +{
      +  "anyOf": [
      +    {
      +      "additionalProperties": true,
      +      "type": "object"
      +    },
      +    {
      +      "type": "string"
      +    },
      +    {
      +      "type": "null"
      +    }
      +  ],
      +  "default": null,
      +  "description": "Optional JSON Schema (an object, or its JSON text). When given, agy\n    is asked to produce output matching it (agy 1.1.8's --json-schema)\n    and this tool returns the VALIDATED OBJECT as JSON text instead of\n    prose — json.loads it. What comes back is agy's own\n    `structured_output`, which carries exactly the declared fields;\n    agy's prose `response` on the same run also picks up its internal\n    toolAction/toolSummary keys and can be prefixed with a sentence, so\n    the two are NOT interchangeable. If agy produces no structured\n    output the call RAISES rather than handing back prose you would\n    have to parse anyway. Needs agy 1.1.8+.\n\n    IMPORTANT — write the prompt so the ANSWER is in the turn, and let\n    the schema only shape it. agy fills the schema in a finishing pass\n    that does not re-reason about the content, so a field the turn never\n    established gets guessed from the schema itself. Measured on 1.1.20\n    with \"this broke my build and wasted my whole afternoon\": with\n    enum [\"positive\",\"negative\"] it answered \"positive\" 3 times out of 4,\n    and simply REVERSING the enum to [\"negative\",\"positive\"] flipped it\n    to \"negative\" 2 out of 2 — it was following field order, not the\n    sentence. Adding a `reason` field did not help; the reason came back\n    \"Completed sentiment classification task.\" Asking the prompt to state\n    the verdict and why, and keeping the same biased enum, was correct\n    3 out of 3. So: extraction of what the model has already worked out\n    is reliable; a judgment delegated to the schema is not."
      +}
  6. Changed1 schema field changedv0.21.4
    • changedInput schema / properties / model / description
      Previous value: -"Optional model label to run this conversation on (agy's --model),\n   e.g. \"Gemini 3.1 Pro (High)\" or \"Claude Sonnet 4.6 (Thinking)\".\n   Omit to use the model set in agy's settings.json (Gemini 3.5 Flash\n   (High) by default). Must be one of `agy models` — an unknown label\n   is rejected up front (agy would otherwise silently ignore it and\n   fall back to the default). See antigravity_status / `agy models`\n   for the valid labels."New value: +"Optional model slug to run this conversation on (agy's --model),\n   e.g. \"gemini-3.1-pro-high\" or \"claude-sonnet-4-6\". Omit to use the\n   model set in agy's settings.json (gemini-3.6-flash-high by\n   default). Must be one of `agy models` — an unknown slug is\n   rejected up front (agy would otherwise silently ignore it and fall\n   back to the default). agy 1.1.5 replaced the old human labels\n   (\"Gemini 3.1 Pro (High)\") with these slugs and 1.1.6 added the\n   gemini-3.6-flash family; the old form is no longer accepted. See\n   antigravity_status / `agy models` for the valid slugs."
  7. Changed1 schema field changedv0.15.4
    • addedInput schema / properties / model
      Added value: +{
      +  "anyOf": [
      +    {
      +      "type": "string"
      +    },
      +    {
      +      "type": "null"
      +    }
      +  ],
      +  "default": null,
      +  "description": "Optional model label to run this conversation on (agy's --model),\n   e.g. \"Gemini 3.1 Pro (High)\" or \"Claude Sonnet 4.6 (Thinking)\".\n   Omit to use the model set in agy's settings.json (Gemini 3.5 Flash\n   (High) by default). Must be one of `agy models` — an unknown label\n   is rejected up front (agy would otherwise silently ignore it and\n   fall back to the default). See antigravity_status / `agy models`\n   for the valid labels."
      +}
  8. First observedv0.11.0

TDQS

A4.5/5.0
Behavior5/5

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

Annotations only declare readOnly=false, openWorld=true, non-idempotent; the description adds the material facts: no real sandbox, plan=True is agent-enforced and not an OS sandbox, and it runs under the user's AI Pro login as an autonomous agent. These are exactly the caveats an agent needs before firing off a mutating, autonomous call.

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-loads what it is best for, then tacks on operational caveats in short clauses. Dense and telegraphic, and every clause carries information, though the fragment style makes it slightly harder to scan than flowing prose.

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

Completeness4/5

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

For a 7-parameter, non-idempotent autonomous call with an output schema, the description covers the risk profile, the required workspace context, and the continuation path. Return-value shape is left to the output schema, which is appropriate; only the timeout parameter is unaddressed in prose.

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 baseline is 3; the description still earns extra credit by explaining the consequence of omitting workspace (no repo context) and the model choice rationale (bigger model for heavier reasoning). It leaves the remaining parameters to the schema, which is fine given how rich those descriptions are.

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 capability (fast, cheap tool-calling and short tasks on Gemini via the agy CLI) and pins the scope as a NEW conversation, distinguishing it from antigravity_continue by name. An agent can pick between the two without opening either schema.

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?

Explicit routing guidance: use antigravity_continue for follow-ups, pass workspace or the sub-agent sees no repo, pick a bigger model for heavier reasoning. It also sets a safety precondition (trusted prompts only). It stops short of saying when to prefer a different model family's ask tool.

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