delegate

A command for sending a clearly defined coding or research subtask to Gemini through the Antigravity command-line tool. It can give Gemini access to a repository and, when authorized, let it write files or use tools.

In plain words
What is it for?
Use it for scoped tasks such as code changes, repository analysis, web research, or terminal work. Choose the Flash tier for routine jobs, Pro for harder reasoning, and provide a repository directory when the task needs project files.
Why use it?
It helps hand off bounded work while keeping the main agent responsible for the overall task and checking the result. It also makes the required permissions and model tier explicit.

Command

Part of the antigravity-for-claude-code plugin — 2 skills, 10 commands, 1 agent, 2 hooks shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/yuting0624/antigravity-for-claude-code/delegate
Clone the repo
git clone --depth 1 https://github.com/yuting0624/antigravity-for-claude-code

Or install antigravity-for-claude-code, the plugin that ships this one along with the rest of its 2 skills, 10 commands, 1 agent, 2 hooks.

Per session 24 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 898 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00024 $0.00898
Opus 5 $0.00012 $0.00449
Sonnet 5 $0.00005 $0.00180
Haiku 4.5 $0.00002 $0.00090

Measured 3d ago against content hash 103bf629c50c, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

delegate scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

commands/delegate.md · 53 lines

What it actually says

Delegate the following task to Antigravity (agy / Gemini) via the plugin wrapper, following the antigravity skill's Cost discipline and Verification gates.

Task: $ARGUMENTS

Do this:

  1. Pick a tier (flash default; pro for hard reasoning). If the task needs the repo, add --dir <repo-root> so agy reads the real files (don't paste them into context). If the task WRITES files or uses tools (web / Vertex AI Search / terminal), it needs a grant. For a plain file write the narrower one is a write_file(<dir>) entry under permissions.allow in ~/.gemini/antigravity-cli/settings.json (recursive beneath <dir>, no flag needed — substitute a real path for <dir>; if a rule is already there and the write is still denied, agy-doctor checks whether agy can parse it). Otherwise pass --yolo, which auto-approves all tools and is what web / Vertex AI Search / terminal need. Without a grant, headless agy leaves your workspace untouched, and only the newest versions admit it (it describes / scratch-diverts / soft-denies / fails outright depending on version; issue #10). --mode accept-edits is not a grant either: measured on agy 1.1.13, where the flag is applied at all, the write is denied exactly like one without it. Run write tasks on a dedicated branch — --sandbox is not containment, it was measured doing nothing under --yolo — and verify files actually changed with git status. Claude Code may prompt for or block --dangerously-skip-permissions — approve it or pre-allow it; non-interactive (claude -p) without that permission can't write/use-tools via agy. (If the wrapper returns exit 15, that's exactly this: agy denied the write. Both shapes land here — the soft deny on agy 1.1.3 and the hard error on 1.1.13 — and both take the same fix: a permissions.allow rule covering the target, or --yolo.)
  2. Run synchronously (you may be headless — do not background-and-wait): agy-delegate --tier <tier> [--dir .] [--yolo] [--digest] "<task>" For read/analysis tasks, add --digest — it appends a digest-only output contract so agy returns compact bullets instead of raw content.
  3. Ingest only the result/digest — do NOT re-read the files agy already handled (keeps your context lean; that's where the cost savings come from). If the wrapper prints a "looks like a raw dump" note on stderr, do NOT ingest the raw output — re-run with --digest or ask agy to summarize it first.
  4. Verify: actually run/check the output; never trust a self-reported "done". Report what you delegated and how you verified it.

Remember the break-even: only delegate if the offloaded volume clearly exceeds the spec + round-trip + verification overhead. Tiny tasks are cheaper to just do yourself.

Long task, interactive session? A sync delegation can also hit Claude Code's ~2-min Bash-tool limit — start it in the background and keep working (this also keeps the prompt cache warm and frees you to do other turns): ID=$(agy-job start --tier pro --dir . "<task>") then check /antigravity:status and collect with /antigravity:result <id>. (Don't do this when YOU are headless claude -p — one-shot, no later turn to collect; delegate synchronously there.)

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 53 lines · 24 tokens per session scan A 103bf629c50c

Subscribe to this mod's changes

delegate is a command published in the GitHub repository yuting0624/antigravity-for-claude-code (296 stars, last pushed 7d ago), licensed MIT. It adds 24 tokens to every session and 898 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.