agy-delegation
Skill0 starsChanged 4 months agoMITView on GitHub
Full protocol for delegating work to the Antigravity CLI (agy / Gemini) — commands, tier choice, exit codes, stdin pitfalls, prompt template, verification gate. Load when actually delegating to agy; routing policy lives in the model-routing skill.
- Reads credentials
3 other projects use a version of this.
Adopted by
Most recently taken up by
- Haruhiko-Joe/skillstheir own version, first seen Nov 15, 2026
- PyModel/claude-agy-mcptheir own version, first seen Nov 15, 2026
- sshahzaiib/agy-bridgetheir own version, first seen Jun 21, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- sshahzaiib/agy-bridge★ 61same name
- PyModel/claude-agy-mcp★ 8same name
- Haruhiko-Joe/skills★ 6same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository3 other files it gives its agents.
- go-ceremonySkill
- model-routingSkill
- quality-gateSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 0 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
0
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
All delegation goes through the plugin wrappers agy-delegate / agy-job (or the /antigravity:delegate skill / antigravity-delegate subagent).
- Very specific: line 29
Read/analysis tasks use --digest and return a summary only.
- Very specific: line 96
Never send agy: API keys, auth tokens, .env, personal data, secrets, or full build logs (extract error/warning lines ±3-5 lines only).
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
agy Delegation Protocol
Prerequisites: this skill assumes the antigravity-for-claude-code plugin is installed, which provides the
agy-delegate/agy-jobwrapper commands and the/antigravity:*skills referenced below. Without that plugin, this skill does not apply — fall back to the mid-tier subagent routing described inmodel-routing.
All delegation goes through the plugin wrappers agy-delegate / agy-job (or the /antigravity:delegate skill / antigravity-delegate subagent). Never call agy -p directly.
Commands
# Read/analysis tasks (~5 min): synchronous + digest
agy-delegate --tier flash --dir . --digest "task"
# Implementation (writes files): --yolo required, on a work branch
agy-delegate --tier pro --dir . --yolo "task"
# Long jobs (>2 min expected, e.g. implementation): background job
ID=$(agy-job start --tier pro --dir . --yolo "task")
# progress: /antigravity:status collect: /antigravity:result <ID>Rules:
- Always pass
--dir .so agy reads the project's own agent-instructions file and real files itself. Never paste file contents into the prompt. --yolois mandatory for any writing task — without it agy only describes edits and reports completion without writing (known plugin issue).- Read/analysis tasks use
--digestand return a summary only. If the wrapper warns "raw dump" on stderr, discard and re-run with--digest. - Tier: investigation / log analysis / boilerplate =
flash; hard implementation / hard reasoning =pro.
Long prompts & record keeping
agy-delegate accepts stdin (-). Keep prompt/response records in the project's own scratch/output directory (e.g. .agents/out/):
cat > .agents/out/agy-<task>-prompt.md << 'EOF'
(prompt)
EOF
agy-delegate --tier pro --dir . --yolo - \
< .agents/out/agy-<task>-prompt.md \
> .agents/out/agy-<task>-response.mdBackground job output lands in ~/.antigravity-jobs/<ID>/; copy the digest into .agents/out/agy-<task>-response.md after collecting.
Known pitfall:
agy-job start ... - < filefails with rc=1 ("no prompt given") — the internalnohupdetachment drops stdin. Synchronousagy-delegatewith stdin is fine; only backgrounding viaagy-job startbreaks. Pass the prompt as a string argument instead:agy-job start --tier pro --dir . --yolo "$(cat .agents/out/agy-<task>-prompt.md)"
Prompt template (implementation tasks)
## Role
You are the implementation agent. Follow the plan from the main thread (the decision engine) and implement only the given scope.
## Plan
(paste the finalized plan)
## Allowed to change
- (explicit files/dirs)
## Must not touch
- (explicit files/modules)
## Completion report must include
- changed file list
- Preview / smoke-check result (if applicable)
- git diff summary
(Do not run the full build — it tends to time out for agy; build verification is the main thread's job.)Pre-flight checks
pwd
git status
git branch --show-current
command -v agy-delegate # if missing: /antigravity:setupExit codes — abort delegation and report to user
- 2 agy execution failed
- 3 empty response
- 10 quota exceeded (user action needed — stop)
- 11 auth error (user action needed — stop)
- 12 timeout (may switch to
agy-job startbackground mode) - 13 agy command not found
Verification gate (non-negotiable)
- Never trust agy's self-report ("Build Succeeded" / "GREEN" / "done"). After the completion report, run the build/tests yourself. agy has been observed patching its own environment to force a pass.
- Review with
git diffonly; do not re-read whole files agy touched (context preservation). - Never send agy: API keys, auth tokens,
.env, personal data, secrets, or full build logs (extract error/warning lines ±3-5 lines only).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

