oh-my-antigravity

A multi-agent skill pack and workflow orchestration layer for Google Antigravity.

Architecture aligned with oh-my-codex; skills ported from the "Oh My" ecosystem.

Give your Antigravity agent 148 specialized skills (99 first-party + 49 vendored upstream) and 35 workflows, plus a CLI that installs, catalogs, routes, and governs them — and that works around Antigravity 2.0's global-skill discovery quirk.

📦 npm: [email protected] · CLI: oma · Node ≥ 18

Why this exists (Antigravity 2.0 fix)

Antigravity 2.0.1 does not reliably advertise the global skills directory (~/.gemini/antigravity/skills/) in the agent's system prompt — so global skills you install are often never discovered, even though the files are in the right place. (Workflows have an explicit global_workflows path; skills do not.)

oma works around this in two ways:

1. Workspace-first install (recommended). Skills are written to both .agents/skills/ (the 2.0 default) and .agent/skills/ (backward compat), which Antigravity searches reliably as {.agents,.agent}/skills.
2. GEMINI.md skills block. On every install, oma injects a managed block into the relevant GEMINI.md (the rules file that is always loaded) listing every skill's name, description, and absolute SKILL.md path. The agent can then find and view_file them even when the global skills directory isn't advertised.

Run oma doctor to see exactly what Antigravity will discover, including whether the GEMINI.md block is present.

Install

Workspace (recommended)

From inside a project (or pass a target dir):

npm install -g oh-my-oma
cd /path/to/your/project
oma setup --scope project-local
# or: oma setup --scope project-local   run from the repo: ./install.sh /path/to/project

Writes: .agents/skills + .agent/skills, .agents/workflows + .agent/workflows, .agent/rules, and a skills block in <project>/GEMINI.md. Restart the Antigravity agent session so it re-scans.

Global (all projects)

oma setup --scope user      # or: ./install_global.sh

Writes: ~/.gemini/antigravity/skills, ~/.gemini/antigravity/global_workflows, ~/.gemini/config/rules, MCP servers into ~/.gemini/config/mcp_config.json, and a skills block into ~/.gemini/GEMINI.md (the discovery fix).

Verify

oma doctor

CLI

oma setup [--scope user|project-local|project] [--force] [--dry-run]
          [--no-skills] [--no-workflows] [--no-rules] [--no-config] [--enable-context7]
oma doctor                       # repo health + what Antigravity will discover
oma route "<task>" [--limit N]   # recommend skills for a task (scored)
oma harness graph|intents|lint|chain <skill>
oma skill list|prefer|conflicts
oma source list|sync|status
oma team start|status|advance|cancel|clear
oma test ...                     # stack detection + test generation
oma help

Architecture

bin/oma.js              CLI entry → src/cli/index.js
src/
  cli/                  setup, doctor, route, harness, skill, source, team, test, notify
  utils/paths.js        single choke point for all Antigravity targets
  config/generator.js   JSON mcp_config.json merge + GEMINI.md skills injection
  merge/                multi-source skill merge + conflict resolution
  catalog/              skill index, manifest, schema validation
  router/               task → skill scoring (boost + index + layer + intent)
  harness/              compose graph, layer-map, intent-registry
  mcp/                  state / memory / trace MCP servers
  team/, notify/, testing/
.agent/skills/local/    99 first-party skills (install source)
.agent/skills/upstream/ vendored upstream packs: superpowers (14), ecc (35)
.agent/workflows/       35 workflows
templates/, schemas/, prompts/, scripts/, .governance/

Skills live in .agent/skills/local/<name>/SKILL.md. The catalog and skill-index are generated; the router and harness read them.

Skills & workflows

148 skills spanning:

• Planning & discovery — plan, analyst, architect, explore, brainstorming, writing-plans, executing-plans, deep-interview, planning-methodology
• Execution — autopilot, ralph, ultrawork, ultraqa, pipeline, start-dev
• Quality & review — aireview, code-review, security-review, critic, verification-before-completion, verification-loop, verify, debug-analysis, systematic-debugging, requesting-code-review, receiving-code-review, finishing-a-development-branch
• Testing — tdd, tdd-guide, test-driven-development, tdd-workflow, tdd-generator, bdd-generator, test-coverage, test-gen, e2e, e2e-testing, eval-harness, qa-tester, checkpoint
• Engineering patterns — api-design, mcp-server-patterns, claude-api, backend-patterns, frontend-design, frontend-patterns, coding-standards, refactor-clean, ai-slop-cleaner
• Research — scientist, deepsearch, research, deep-research, iterative-retrieval, documentation-lookup, exa-search, market-research
• Design & writing — designer, frontend-ui-ux, vision, de-ai-writing, article-writing, brand-voice, content-research-writer
• Meta & skill self-management — skill, skill-create, skill-development, skill-tester, skill-debugger, skill-quality-analyzer, skill-doc-generator, update-codemaps, update-docs, prompt-optimizer, learner, release, doctor, writing-skills, using-superpowers

35 workflows are invoked as slash commands, e.g. /autopilot, /ultrawork, /ralph, /aireview, /research, /tdd, /doctor, /help.

Skill packs

| Pack | Skills | Source | Notes | |---|---|---|---| | local (first-party) | 99 | this repo | Antigravity-native; passes governance for foreign-runtime paths | | upstream/superpowers | 14 | obra/superpowers | Process discipline (TDD, brainstorming, planning, debugging, review) | | upstream/ecc | 35 | affaan-m/everything-claude-code | Engineering patterns (api-design, mcp-server-patterns, claude-api, etc.) |

Caveats

• Parallel-execution skills (swarm, ultrapilot) were removed during the Antigravity migration; orchestration is single-agent / persona-switching.
• superpowers/subagent-driven-development and superpowers/dispatching-parallel-agents fall back to single-session execution under Antigravity (no Task-tool equivalent in Gemini CLI).
• 2 upstream-vs-local name collisions (security-review, verification-loop) are auto-resolved by the merger; see .oma/merge-report.json after install.

Develop / test

npm test                  # governance + catalog + skill-index + doctor
npm run catalog:generate  # regenerate catalog from .agent/skills
npm run governance:skills # lint skills for foreign-runtime leakage
oma harness lint          # validate the compose graph / layer-map / intents

License & credits

MIT. See LICENSE.

• oh-my-codex — the architecture this port follows
• oh-my-claudecode — original skill ecosystem