Short version: one graph, shared, and you install each tool's integration once in the repo.

The graph is shared. graphify-out/ is tool-agnostic, so you build it once and both Codex and Antigravity query the same graph. You never need two graphs or two builds.

graphify .        # build the graph once (or: graphify extract .)

The skill is shared too. Project-scoped, both Codex and Antigravity read the skill from the same path, .agents/skills/graphify/SKILL.md, so a single install location serves both.

Each tool just needs its own always-on hook, so run each install once in the repo:

graphify install --project --platform codex     # adds AGENTS.md + the .codex hook
graphify antigravity install                     # adds .agents/rules + .agents/workflows (+ an MCP config hint)

On the current release (0.8.29), use graphify antigravity install for Antigravity so you get its .agents/rules + .agents/workflows always-on layer. (The --project --platform antigravity flag was only laying down the skill and skipping those; that parity gap is now fixed on v8 and ships in the next release.)

For full architecture navigation in both tools you can also point them at the MCP server (graphify.serve graphify-out/graph.json); the antigravity installer prints the exact config block for that.

So: one graph, one shared .agents/skills skill, plus each tool's own always-on file. Re-run graphify update . (or /graphify) as the code changes and both tools stay current off the same graph.