rldyour-antigravity-cli
rldyour-antigravity-cli is the Antigravity CLI-native configuration adapter for the rldyour AI CLI control plane. It provides a public, source-validated Antigravity CLI configuration using native surfaces: GEMINI.md, .gemini/antigravity-cli/settings.json, .gemini/antigravity-cli/mcp_config.json, gemini-extension.json, TOML commands, skills, subagents, hooks, policies, MCP servers, browser-provider routing, Serena memory, and release validation.
Current Baseline
| Surface | Value |
|---|---|
| Adapter version | 1.7.31 |
| Runtime baseline | Antigravity CLI 1.1.1 |
| Runtime channel | bootstrap-owned generation-pinned artifact |
| GitHub release tag | 1.7.31 |
Antigravity CLI freshness uses agy --version as the primary
source of truth, with the GitHub release tag as release provenance.
The pinned commit and product version are governed by the root control plane
config/repositories.json.
What This Repository Provides
This repository is a configuration package for Antigravity CLI, not a fork or
copy of the upstream runtime. It wires together
the native Antigravity CLI configuration surfaces - context files, project
settings, MCP config, extension manifest, TOML commands, Agent Skills,
subagents, hooks, policies, and MCP servers - into a coherent,
version-controlled, owner-validated setup. The upstream Antigravity CLI binary
is installed separately via the official installer and is not modified here.
All changes in this repository are configuration and adapter logic; runtime
behavior is governed by the installed agy binary version.
Native Boundaries
Antigravity runtime configuration is represented through Antigravity native
files only; .gemini/, GEMINI.md, and gemini-extension.json are native
namespace filenames retained for Antigravity compatibility.
Claude Code slash-command files, Codex plugin manifests, and OpenCode command
JSON are not Antigravity runtime surfaces and are not present here.
Active native surfaces:
GEMINI.md- repository context injected at session start..gemini/settings.json- project settings, hook wiring, and MCP server declarations.gemini-extension.json- extension metadata and extension MCP wiring..gemini/commands/**/*.toml- custom commands (13 TOML files across browser, flow, release, ry, security, and serena groups)..gemini/skills/*/SKILL.md- Agent Skills (14 skills)..gemini/agents/*.md- Antigravity subagents (7 agents: browser-reviewer, codebase-investigator, design-reviewer, antigravity-cli-specialist, release-reviewer, security-reviewer, serena-curator)..gemini/hooks/- native hook scripts for SessionStart, BeforeTool, AfterAgent, and SessionEnd events..gemini/policies/*.toml- policy extension data.
Hooks use the native event-keyed schema: read JSON from stdin, write exactly
one JSON object to stdout, and send diagnostics to stderr. Antigravity subagents
are internal Antigravity CLI delegation and are not cmux worker terminals.
Headless agy -p is allowed for smoke, doctor, and CI checks only.
Install / Update / ry-repair
Install the runtime through the parent bootstrap's verified artifact channel:
bash scripts/bootstrap.sh --platform <macos|ubuntu> --profile <desktop|server> --apply
agy --version
The bootstrap verifies the platform artifact digest, verifies the macOS signer
where applicable, disables self-update, and publishes a tamper-evident
~/.local/bin/agy launcher. This adapter never executes a streamed remote
installer.
Clone or update this adapter:
# First time
git clone https://github.com/NDDev-it-com/rldyour-antigravity-cli.git
cd rldyour-antigravity-cli
# Update to latest
git pull origin main
Run /ry-repair convergence from within a Antigravity CLI session to normalize
adapter configuration, validate surface adoption, and sync Serena memories.
For offline or check-only mode, run the static validation lane (see
## Validation). Runtime doctor:
agy --version
agy -p 'Return exactly READY.' --output-format json
The installed-runtime smoke checks require a deliberately configured owner machine with valid authentication.
Active Catalog
| Surface | Count |
|---|---|
Agent Skills (.gemini/skills/*/SKILL.md) |
14 |
TOML commands (.gemini/commands/**/*.toml) |
13 |
Antigravity subagents (.gemini/agents/*.md) |
7 |
Hook scripts (.gemini/hooks/) |
4 |
MCP servers (.gemini/settings.json) |
11 |
Active MCP inventory: serena, chrome-devtools, sequential-thinking,
shadcn, dart-flutter, context7, github, deepwiki, grep, figma,
and openai-docs.
Commands are grouped into: browser, flow, release, ry, security, and
serena. Hooks cover SessionStart, BeforeTool, AfterAgent, and
SessionEnd events.
Browser / Design / DevTools Routing
Browser work has exactly two execution providers behind the bootstrap-owned
CloakBrowser boundary. Every browser action first runs exact
$HOME/.local/bin/cloakbrowser-cdp-health; missing or nonzero health stops as
NOT_PROVEN, with no fallback:
- Managed Playwright CLI (
$HOME/.local/bin/playwright-cli, runtime0.1.17) - flows, screenshots, snapshots, traces, responsive matrices, visual evidence, and long-horizon stepwise workflows.run-codeand--filenameare forbidden. - Managed Chrome DevTools MCP (
chrome-devtools, runtime1.5.0) - console, network, runtime, DOM/layout, performance, Lighthouse, and memory through the exact configured wrapper.
webwright-task is compatibility intent routed to browser:validate; it never
authorizes Webwright runtime execution. Stock/raw/in-app Browser,
browser_agent, node_repl, computer-use, Playwright MCP, raw Playwright,
direct browser packages, alternate CDP/executable/config paths, and every
fallback are forbidden.
MCP Inventory Detail
The active MCP inventory matches the root positive inventory. All 11 servers are
declared in .gemini/settings.json under mcpServers:
| Server | Transport | Role |
|---|---|---|
serena |
stdio (uvx) |
LSP-based code intelligence and Serena memory |
chrome-devtools |
stdio (managed $HOME/.local/bin/chrome-devtools-mcp wrapper) |
CloakBrowser live debugging and DevTools |
sequential-thinking |
stdio (bunx) |
Structured reasoning chains |
shadcn |
stdio (bunx) |
shadcn/ui component registry |
dart-flutter |
stdio (dart) |
Dart/Flutter LSP and pub tools |
context7 |
stdio (bunx) |
Library documentation lookup |
github |
HTTP/SSE | GitHub repos, issues, pull requests |
deepwiki |
HTTP/SSE | AI-powered repository documentation |
grep |
HTTP/SSE | Cross-repository code search |
figma |
HTTP (local) | Figma design context and code connect |
openai-docs |
HTTP/SSE | OpenAI API documentation |
Current refreshed package pins are
@modelcontextprotocol/[email protected] and
@upstash/[email protected]. config/mcp-runtime-versions.env owns their
version literals; all three active Antigravity MCP surfaces and
config/rldyour-contract.json must remain exact projections of those pins.
Repository Context / Serena Memory
Normal main history carries source-controlled configuration artifacts:
.gemini/ surfaces, config/, references/, scripts/, tests, and
CHANGELOG.md. Durable agent context files - AGENTS.md, GEMINI.md,
.serena/project.yml, and .serena/memories/ - are tracked on main as
normal source. Runtime-local Serena cache, review scratch files, diagnostics,
markers, local state, tokens, cookies, and credentials stay ignored.
Serena memory domains active for this adapter:
CORE-01-INDEX- durable memory index.GEMINI-01-ADAPTER-SURFACE- native surface facts.RUNTIME-01-BASELINE- pinned runtime version evidence.MCP-01-TOOLS- MCP inventory facts.BROWSER-01-VALIDATION- browser provider routing rules.FLOW-01-SDLC- SDLC flow rules.SECURITY-01-POSTURE- security posture facts.RELEASE-01-VALIDATION- release validation rules.SERENA-01-MEMORY-SYNC- memory maintenance rules.ANTIGRAVITY-01-TRANSITION- Antigravity transition notice.
Memories store durable facts only (Purpose, Current Facts, Evidence, Operational Rules, Last Verified). Chat logs, speculation, secrets, tokens, cookies, and credentials must not be stored.
Freshness contract: re-verify memories after each adapter release and after any upstream Antigravity CLI version change. Memory updates are committed with the source change that makes their facts true.
Security Boundary
Approval Mode and YOLO
Committed .gemini/settings.json sets general.defaultApprovalMode = "auto_edit" - the maximal owner-autonomy posture that Antigravity CLI accepts in
committed configuration. auto_edit auto-approves file edits without prompting.
security.disableYoloMode = false and security.toolSandboxing = false are set
explicitly to preserve the maximal committed owner-autonomy posture currently
proven for Antigravity CLI.
Full shell-command YOLO flags for Antigravity CLI are NOT_PROVEN until the
installed-runtime lane verifies the current agy --help contract. Do not copy
retired legacy runtime --approval-mode=yolo semantics into Antigravity
launchers or docs. Writing an unverified yolo value into committed settings
has no effect and must not be done.
Permissions are not a sandbox. This adapter is designed for an owner workstation with owner-trusted tools and reviewed MCP servers. It is not suitable as-is for untrusted multi-user or production environments.
Secrets
Do not commit Antigravity, Google model, or Google API keys, OAuth material,
service-account JSON, Google Cloud ADC files, cookies, browser profile state, or
MCP provider tokens. .env.example documents accepted local variable names;
real values must remain local and git-ignored.
MCP Trust Boundary
All 11 active MCP servers are owner-reviewed. HTTP/SSE servers (github,
deepwiki, grep, openai-docs) use remote endpoints with env-var tokens
where required. The figma server runs locally at 127.0.0.1:3845. No
unapproved third-party MCP server may be added without an explicit inventory and
release-policy update.
Access and Antigravity Notice
Antigravity CLI adapter 1.7.31 targets enterprise, paid API-key, Vertex AI, Google
Cloud, and explicitly owner-approved authenticated environments. Consumer OAuth
availability after June 18, 2026 is NOT_PROVEN for this adapter because
legacy runtime support is retired for active use in this adapter. See
references/gemini-antigravity-transition.md.
Validation
Static validation (no runtime required):
python3 scripts/validate_gemini_config.py --strict
python3 scripts/validate_gemini_extension_manifest.py --strict
python3 scripts/validate_gemini_commands.py --strict
python3 scripts/validate_gemini_skills.py --strict
python3 scripts/validate_gemini_subagents.py --strict
python3 scripts/validate_gemini_hooks.py --strict
python3 scripts/validate_gemini_mcp_inventory.py --strict
python3 scripts/validate_gemini_browser_routing.py --strict
python3 scripts/validate_gemini_runtime_baseline.py --strict
python3 scripts/validate_instruction_docs.py --strict
python3 scripts/validate_serena_memory_schema.py --strict
python3 scripts/validate_serena_memory_semantics.py --strict
python3 -m pytest -q
Installed-runtime smoke checks (optional; require a deliberately configured owner machine with valid authentication):
agy --version
agy -p 'Return exactly READY.' --output-format json
The NOT_PROVEN policy applies to any validation result that cannot be
confirmed through local static analysis alone (e.g., live MCP connectivity,
consumer OAuth availability after June 18, 2026).
Release / Rollback
Releases follow the root control plane numeric tag policy. A VERSION file
alone is not a release: the GitHub Release and the numeric tag must point to the
same commit that the root pins in config/repositories.json.
- Default version movement is patch (
+0.0.1) after any source, config, or documentation change post-release. - Minor (
+0.1.0) and major (+1.0.0) bumps are owner-directed decisions only. CHANGELOG.mddocuments all notable changes per release.SECURITY.mdlists the currently supported version.- Only the current exact numeric tag receives security fixes.
To roll back: check out the prior numeric tag, verify the root control plane
pins it in expected_head, and run the static validation lane.
Support / License
License: AGPL-3.0-or-later.
Author: Danil Silantyev (github:rldyourmnd), CEO NDDev.
Repository: https://github.com/NDDev-it-com/rldyour-antigravity-cli
Report security issues privately through GitHub Security Advisories for
NDDev-it-com/rldyour-antigravity-cli. Do not open public issues for security
vulnerabilities.
No comments yet
Be the first to share your take.