antigravity
Skill0 starsChanged 5 months agoMITView on GitHub
Set up or re-sync the current project to run in Google Antigravity. Checks Antigravity's LIVE docs first (mechanics drift — never build from remembered facts), runs Soloship's antigravity target (rules, hooks, global plugin), closes the bespoke-rule drift gap, mirrors MCP servers, then reports the manual settings steps (terminal policies, review triggers) and the injection verification the user must run in Antigravity itself. Use when asked "/antigravity", "set up antigravity", "sync antigravity", or "get this project working in Antigravity".
8 other projects use a version of this.
Adopted by
Most recently taken up by
- ITS-MSW/skillstheir own version, first seen Oct 17, 2026
- erayendes/cogitorstheir own version, first seen Oct 9, 2026
- koenvdheide/antigravity-skilltheir own version, first seen Oct 8, 2026
- GryAsl/Polyphonytheir own version, first seen Oct 8, 2026
- yuting0624/antigravity-for-claude-codetheir own version, first seen Jul 2, 2026
- PURPLE-ORCA/skillstheir own version, first seen Jun 20, 2026
Other versions8 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 8
- yuting0624/antigravity-for-claude-code★ 477same name
- KINGSTAR-OMEGA/claude-token-optimizer★ 111same name
- sakibsadmanshajib/antigravity-plugin★ 19same name
- GryAsl/Polyphony★ 4same name
- erayendes/cogitors★ 2same name
- ITS-MSW/skills★ 2same name
- PURPLE-ORCA/skills★ 2same name
- koenvdheide/antigravity-skill★ 0same 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 repository55 other files it gives its agents.
- AGENTS.mdAGENTS.md
- scripts/AGENTS.mdAGENTS.md
- src/AGENTS.mdAGENTS.md
- .cursor/rules/publish-version-bump.mdcCursor rule
- .cursor/rules/release-version-sync.mdcCursor rule
- auditSkill
- autoplanSkill
- bootstrapSkill
- brainstormSkill
- browseSkill
- ceo-reviewSkill
- clarifySkill
- cleanupSkill
- code-reviewSkill
- codexSkill
- component-inventorySkill
- context-restoreSkill
- context-saveSkill
- critiqueSkill
- cronSkill
- csoSkill
- cursorSkill
- debugSkill
- deepen-planSkill
- design-reviewSkill
- devex-reviewSkill
- document-reviewSkill
- eng-reviewSkill
- executing-plansSkill
- fableSkill
- finishSkill
- frontend-designSkill
- grill-meSkill
- grokSkill
- implementSkill
- learnSkill
- office-hoursSkill
- onboardSkill
- plan-design-reviewSkill
- planSkill
- polishSkill
- qaSkill
- reviewSkill
- shipfastSkill
- shipthoroughSkill
- simplifySkill
- specSkill
- statusSkill
- subagent-driven-developmentSkill
- test-driven-developmentSkill
- ui-auditSkill
- ui-ux-pro-maxSkill
- using-git-worktreesSkill
- verification-before-completionSkill
- writing-plansSkill
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 2 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 8 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 38
Never edit or trim them to fit the target.
- Very specific: line 114
Project without Soloship: create .agents/rules/ and copy all .claude/rules/*.md verbatim.
- Very specific: line 126
If .mcp.json exists: list servers by name, ask the user which to make available in Antigravity (multi-select).
- Very specific: line 137
Write the manifest to .ai/sync-state/antigravity.json.
- Very specific: line 138
Commit generated .agents/ files per project conventions (worktree rules apply).
- Distinctive: line 148
Protections absent in Antigravity: every hook/gate with no counterpart, named, with what it guarded — plus any hook proposals from the new …
“Common” is measured: kept by at least 6 in 10 of the 8 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
/antigravity — Project Environment Sync for Google Antigravity
Make this project's governance fully available to Antigravity, from whatever state it's in now. Idempotent: first run sets everything up; later runs sync only what changed and report what's new.
Standing caveat this skill must carry into its report: on 2026-08-15,
Antigravity with all rules installed on disk (doctor 3/3) still ignored
them and mis-reported .claude/rules/ as its own loaded context. Files
present on disk and rules reaching the model's context are two different
claims — this skill can verify the first, only the user can verify the
second (Phase 7's verification step).
Gates (binding in every posture — violating any is a failed run)
- Docs before build. The live docs check (Phase 3) runs BEFORE any sync work. Antigravity's mechanics drift between runs — its docs tree gained Hooks and per-surface (app/IDE/CLI) sections after Soloship's target was built. Building from remembered facts and checking docs afterward is the failure class this ordering prevents. The report records the docs-check date.
- Sources are read-only.
CLAUDE.md,AGENTS.md,.claude/rules/,.claude/settings.json,.mcp.jsonare inputs. Never edit or trim them to fit the target. - Prefer the installer over hand-rolling. Soloship HAS an
antigravity target:
npx soloship init --agent antigravity/npx soloship upgrade --agent antigravity, plus the global plugin sync (npm run antigravity:install-localfrom the Soloship repo, landing in~/.gemini/config/plugins/soloship). Run those — don't reimplement them (Antigravity itself once proposed rebuilding this installer because it didn't look for it; don't repeat that). - Real copies, never symlinks.
ln -s ../.claude/rules .agents/rulesmakes Soloship's installer write future Antigravity-only rules straight into.claude/rules/(verified againstinstallRulesAt(), 2026-08-15). - Syncing config does not certify the tool. The final report must
say so and point to the workspace's certification battery if one
exists (default location:
docs/testing/model-certification/SOP.md).
Phase 1 — Discover the source surface
Inventory: CLAUDE.md, AGENTS.md, every .claude/rules/*.md,
.claude/settings.json hooks, .mcp.json, existing .agents/ config
(rules/, hooks.json), ~/.gemini/config/plugins/soloship (global
plugin, with its syncedAt), ~/.gemini/GEMINI.md (user-global rules),
and Soloship presence (.soloship/version). Run npx soloship doctor
and record the antigravity lines — the required items are
.agents/rules/, .agents/hooks.json, and the global plugin.
Phase 2 — What's new since last run
State lives at .ai/sync-state/antigravity.json:
{ "syncedAt": ISO-8601, "files": { "<source path>": "<sha256>" } }.
Hash every source (shasum -a 256), diff against the manifest: new /
changed / removed / unchanged. No manifest = first run, say so.
Removed sources: remove their .agents/rules/ counterparts after
verifying the file really derives from the removed source.
Phase 3 — Live docs check (BEFORE any sync work)
Fetch Antigravity's current docs — start at
https://antigravity.google/docs/getting-started and follow the nav
(as of 2026-08: /docs/rules-workflows/, /docs/ide/rules/,
/docs/mcp/, /docs/agent-settings/, /docs/permissions/,
/docs/hooks/). If URLs moved, search.
Confirm before acting: rules directory (.agents/rules/) and
activation modes unchanged; whether project-level GEMINI.md still
matters alongside AGENTS.md; MCP config location and schema for the
surface the user actually uses (the docs tree has separate MCP pages
for the app, IDE, and CLI); and whether the Hooks docs describe
capabilities that could carry more of the Claude Code / Soloship gates.
This phase outputs the verified mapping Phase 4 executes, plus a numbered manual-steps list written for a non-coder — exact menu path, exact value, one line on why. Two settings always get an explicit recommendation with reasoning:
- Terminal policy: recommend NOT Turbo (auto-executes commands; an uncertified agent with Turbo is unauthorized-write-by-default).
- Review triggers: recommend Request Review (or the strictest available) until the tool passes certification.
Phase 4 — Sync (per Phase 3's verified mapping)
- Installer first:
init/upgrade --agent antigravity, then the global plugin sync if the Soloship repo is on this machine. Re-runnpx soloship doctor— antigravity must be 3/3. Doctor output is the evidence; paste the relevant lines in the report. (1/3 with only the global plugin present means the per-project.agents/directory is missing — the installer creates it.) If Phase 3 found the installer's output no longer matches what Antigravity reads, stop and surface that as a Soloship bug instead of patching around it. - Close the drift gap the installer misses: diff
.claude/rules/against.agents/rules/by filename AND content hash; copy any rule absent or stale, byte-identical. List every file moved — each is a candidate to fold intosrc/rules.ts(suggest it; the change belongs to the Soloship repo). - Project without Soloship: create
.agents/rules/and copy all.claude/rules/*.mdverbatim. Confirm rootAGENTS.mdexists (Antigravity reads it natively); report a gap rather than inventing one. Do not createGEMINI.mdunless Phase 3 found the project-level file still meaningful alongsideAGENTS.md. - Hooks:
.agents/hooks.jsoncomes from the installer. If Phase 3 found new hook capabilities, propose mappings for more of the Claude Code / Soloship gates in the report. Every gate with no counterpart is listed as protection absent in Antigravity.
Phase 5 — MCP servers
If .mcp.json exists: list servers by name, ask the user which to make
available in Antigravity (multi-select). Configure them per the
location and schema verified in Phase 3. Secrets: placeholder plus a
manual step unless the target file is confirmed outside the repo or
gitignored.
Phase 6 — Verify, record, commit
npx soloship doctor3/3 on antigravity, file counts match the delta, every copied rule content-identical to source (shasumboth sides).- Write the manifest to
.ai/sync-state/antigravity.json. - Commit generated
.agents/files per project conventions (worktree rules apply).
Phase 7 — Report (always, in this order)
- Docs-check date + anything that changed in Antigravity's mechanics since the last run.
- Delta since last run (or "first run — full setup").
- Synced: what Antigravity now gets (doctor lines + drift-gap list + plugin syncedAt).
- Protections absent in Antigravity: every hook/gate with no counterpart, named, with what it guarded — plus any hook proposals from the new docs.
- Manual steps: the numbered list from Phase 3, including the terminal-policy and review-trigger recommendations.
- MCP: mirrored / skipped / placeholders.
- Injection verification (user-run, in Antigravity): files on disk
don't prove rules reach the model. Give the user this exact check to
run in a fresh Antigravity session after a full app restart: ask
"Read the file at .agents/rules/plan-claim-verification.md and tell
me what it says. Separately: what governance rules do you have
loaded right now, and from where?" — if it can read the file when
pointed at it but doesn't spontaneously know
.agents/rules/exists, the injection isn't working (observed 2026-08-15). - Certification reminder: config synced ≠ tool trusted — run the battery SOP before granting write access; until then, proposal-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.

