Antigravity Awesome Skills

The Best Skills for AI coding agents.
Codex, Cursor, Claude Code, Antigravity, Copilot & more.
Trusted by 37K+ GitHub stargazers⭐
@sickn33

[Antigravity Awesome Skills @AASkills_]

Antigravity Awesome Skills is now on X @AASkills_
- The best skills for AI coding agents.
- Daily workflows, prompts, and repo updates.
- Trusted by 37K+ GitHub stargazers⭐

[Antigravity Awesome Skills @AASkills_]

Example prompt: Use runaway-guard to review this image-generation job before I ship it. Identify every unbounded paid API path, write the cost contract, choose MAX_CALLS and concurrency limits, and tell me which provider hard cap must be set.

[Antigravity Awesome Skills @AASkills_]

Daily Skill #008: runaway-guard. For any agent, retry loop, queue, webhook, or batch job that touches a paid AI API. It makes the assistant write the wallet invariant first: max calls, max $ per run, max $ per day, concurrency, retries, and the provider hard cap.

[Antigravity Awesome Skills @AASkills_]

https://t.co/WfYAA7cqea

[Antigravity Awesome Skills @AASkills_]

Example prompt: "Use decision-navigator. I want to change careers but I feel stuck. Ask me one branching question at a time until we reach a concrete first-week plan."

[Antigravity Awesome Skills @AASkills_]

Daily Skill #007: decision-navigator. For when a user says "I don't know where to start" and a big advice dump would make it worse. It narrows uncertainty with one targeted branching question at a time, then turns the leaf into concrete next steps.

[Antigravity Awesome Skills @AASkills_]

https://t.co/3yU9xGRzYR

[Antigravity Awesome Skills @AASkills_]

Example prompt: "Use agentic-actions-auditor to review this repo's .github/workflows for AI-agent steps, tainted GitHub event input, dangerous sandbox settings, and wildcard user allowlists. Return findings with evidence and remediation."

[Antigravity Awesome Skills @AASkills_]

Daily Skill #006: agentic-actions-auditor. Audit GitHub Actions workflows that run AI coding agents. It traces attacker-controlled issue, PR, or comment data into Claude Code, Gemini CLI, Codex, and GitHub AI Inference steps before CI becomes a prompt-injection surface.

[Antigravity Awesome Skills @AASkills_]

https://t.co/TXWyL2TXgX

[Antigravity Awesome Skills @AASkills_]

Example prompt: Use mock-hunter to audit https://my-app.example/admin. I own this app; stay observation-only first, then report which stats, buttons, tables, and links are real, hardcoded, broken, or unknown.

[Antigravity Awesome Skills @AASkills_]

Daily Skill #005: mock-hunter. AI-built dashboards can look finished while every number is fake. mock-hunter audits a live page and classifies visible values as REAL, MOCK, LLM, HARDCODED, BROKEN, or UNKNOWN.

[Antigravity Awesome Skills @AASkills_]

https://t.co/AEdNS1Fl6n

[Antigravity Awesome Skills @AASkills_]

Takeaway: start by hiding generated noise, not source code. Keep manifests and .env.example visible; ignore real .env files, logs, build outputs, binaries, and local DBs.

[Antigravity Awesome Skills @AASkills_]

Daily Skill #004: geminiignore-finops. Your agent should read the code you wrote, not every lockfile, build artifact, log, binary, and local database in the repo. Tune .geminiignore for smaller context, faster runs, and lower token waste.

[Antigravity Awesome Skills @AASkills_]

Skill link: https://t.co/rkTeKjVm9P

[Antigravity Awesome Skills @AASkills_]

"Use agenttrace-session-audit on my latest coding-agent run. Find token/cost spikes, failed tool paths, retry loops, and missing verification, then tell me what to change before retrying."

[Antigravity Awesome Skills @AASkills_]

Daily Skill #003: agenttrace-session-audit

When an AI coding run goes sideways, this skill audits the run itself: cost spikes, tool failures, retry loops, latency gaps, shallow sessions, and what changed between attempts.

[Antigravity Awesome Skills @AASkills_]

https://t.co/xvo5TWcGCN

[Antigravity Awesome Skills @AASkills_]

Example prompt:

"Use subagent-orchestrator to ship this feature. Map the repo first, split reader/writer/browser missions, keep each agent scoped, then integrate and run the final validation yourself."

Skill:
https://t.co/9QBBHtCeCN

[Antigravity Awesome Skills @AASkills_]

Daily Skill #002: subagent-orchestrator

Big tasks break when every agent gets the whole mission.

This skill makes agents split the work first, then choose the right scoped subagent: scouting, implementation, browser checks, repair, or final integration review.

[Antigravity Awesome Skills @AASkills_]

Core rule: scope first, delegate second.

It checks:
- task boundaries
- owned files
- read-only vs writer agents
- dependency order
- validation owner
- whether a repair agent is cheaper than rerunning everything

No "one giant context dump" shortcut.