antigravity-code-review
Code review with optional Antigravity: prefer google_antigravity_review_diff on git changes; Codex verifies findings; optional writing polish pass.
- License 许可
- MIT
- 所属仓库
- Meapri/agent-hub
- 最近更新
- 2026-08-04
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- antigravity-code-review
- description
- Code review with optional Antigravity: prefer google_antigravity_review_diff on git changes; Codex verifies findings; optional writing polish pass.
03 SKILL.md 正文
Antigravity Code Review
Use this skill for code review with a strict evidence boundary: Codex discovers and verifies every technical finding; Antigravity may only turn the completed finding set into clear prose.
Workflow
- Inspect
git status, the complete relevant diff, surrounding code, tests, call sites, and configuration locally in Codex. Optional assist:google_antigravity_review_difffor a first-pass opinion — still re-verify every claim in Codex before presenting. - Discover bugs, regressions, missing tests, and security risks in Codex. Do not treat Antigravity output as authoritative without local verification.
- Verify each retained finding against the actual file and assign its final severity, file, line, evidence, and remediation in Codex.
- If prose help is useful, send only the verified finding records and residual
test gaps to
google_antigravity_writewithtask: "technical-doc". If the configured Google provider is unavailable, keep the verified local findings and format them in Codex rather than changing transports. - Recheck the returned prose in Codex. Remove invented claims, changed severities, unsupported commands, incorrect file/line references, and model meta-commentary before presenting it.
Review Prompt Shape
Give Antigravity an immutable, already verified finding set:
{
"task": "technical-doc",
"model": "gemini-3.1-pro-preview",
"instruction": "Format these verified findings clearly. Preserve every severity, file, line, fact, and test gap exactly. Do not add findings or technical claims.",
"source_text": "<verified finding records from Codex>",
"project_context": "off",
"retry_count": 1,
"timeout_sec": 180
}
Output Rules
Lead with confirmed findings ordered by Codex-assigned severity. Include only file/line references verified locally. If no confirmed issues remain, say that clearly and mention residual test gaps without asking Antigravity to invent a review narrative.
Boundaries
- Do not send secrets, raw credential material, or unrelated private code.
- Do not send raw repository content when verified finding records are enough.
- Antigravity must not originate findings, change severity, or decide whether a claim is technically correct.
- Do not include speculative or model-originated findings that Codex did not verify locally.
04 同仓库更多 Skills
grok-doctor
Diagnose Grok Codex consent, API key, and MCP readiness.
grok-codex
Use the Grok Codex leaf MCP for consent-gated xAI chat, model listing, and provider status.
claude-doctor
Diagnose Claude Codex consent, API key, and MCP readiness.
claude-codex
Use the Claude Codex leaf MCP for consent-gated Anthropic chat, model listing, and provider status.
release-copilot
Use when the user asks Codex to prepare a release, PR description, changelog, release notes, tag plan, GitHub release plan, version summary, or release checklist from a local git repository. This integrated version uses google_antigravity_release_snapshot, google_antigravity_release_draft, and optional google_antigravity_write polishing.
google-grounded-search
Use consent-gated Google-grounded search for current facts, source checks, versions, policies, prices, schedules, and verification.