Skill

Ping major LLM providers in parallel and compare real API latency. Run with /ping

Security Scan

The skill's purpose (measuring LLM API latency) matches its code, but there are inconsistencies between declared requirements and what the script actually needs (it reads secrets from pass and relies on curl/bc), so review before installing.

Purpose & Capability

Name/description match the implementation: scripts/ping.sh makes parallel requests to multiple LLM provider APIs and measures latency. The providers and models listed in SKILL.md align with the endpoints called in the script.

Instruction Scope

SKILL.md correctly instructs running scripts/ping.sh and documents that it uses `pass shared/` to retrieve API keys. The script only sends minimal prompts and discards response bodies, returning timing results. It does not attempt to read unrelated system files or exfiltrate data to third parties. However, SKILL.md suggests optionally adapting to environment variables even though the shipped script does not read them; that mismatch is a small scope ambiguity the user should address.

Install Mechanism

No install spec (instruction-only with one script). Nothing is downloaded or written to disk beyond a short-lived temp dir created at runtime. This is low install risk.

Credentials

Registry metadata lists no required env vars or required binaries, but SKILL.md lists optional API keys and the script actually retrieves keys from `pass shared/...`. The script implicitly requires the `pass` binary plus common utilities (curl, bc, mktemp, sort). The requested credentials (provider API keys) are appropriate for the stated purpose, but the metadata omission of required binaries and the mismatch between env-vars listed in SKILL.md and the script's actual secret sourcing is an inconsistency worth noting.

Persistence & Privilege

The skill does not request permanent presence (always: false), does not modify other skills or system configuration, and does not store credentials or enable itself. It runs ephemeral network requests only when invoked.

What to consider before installing

This script will read your LLM provider API keys (it expects them in pass at shared/<provider>/api-key) and send tiny requests to each provider to measure latency. Before using: (1) Confirm you have and trust the providers whose keys will be used; the script will transmit your keys to those provider APIs as part of normal requests. (2) Ensure you have the required binaries installed (pass, curl, bc, mktemp) — the registry metadata does not declare these but the script requires them. (3) If you prefer environment variables, adapt the script to source keys from env vars (the shipped script does not read them). (4) Note Google API key is used in a query parameter (may appear in logs/proxy traces); consider using header-based auth if preferred. (5) Run the script in a safe/test environment first and inspect it locally; it deletes temp files and discards response bodies, but verify it behaves as expected. The inconsistencies are likely sloppy metadata, not malicious intent, but review and adapt the script before granting it access to real credentials.

LLM Speedtest

Ping major LLM providers in parallel and compare real API latency (TTFT).

When to Use

• User types /ping or asks about model latency/speed
• Comparing provider response times
• Checking if a specific provider is slow or down

How It Works

Runs scripts/ping.sh which:

1. Retrieves API keys from pass shared/ (users may need to adapt key sourcing for their setup)
2. Fires parallel curl requests to each provider with a minimal prompt ("hi", max_tokens=1)
3. Measures total round-trip time per provider
4. Sorts results by latency and displays with color badges

Output Format

Results are sorted fastest-to-slowest with color badges:

• 🟢 < 2s — Fast
• 🟡 2–5s — Normal
• 🔴 5–30s — Slow
• ⚫ 30s — Timeout

Example:

⚡ Model Latency — 14:32

🟢 `Gemini       412ms`
🟢 `GPT-4o       623ms`
🟢 `Sonnet       891ms`
🟡 `Grok        2104ms`
🟡 `MiniMax     3210ms`
🟡 `Opus        4102ms`

_real API latency (TTFT)_

Models Tested

Provider	Model
Anthropic	Claude Sonnet 4
Anthropic	Claude Opus 4
OpenAI	GPT-4o-mini
Google	Gemini 2.5 Flash
MiniMax	MiniMax-M1
xAI	Grok 3 Mini Fast

Cost

~$0.0001 per run (1 token per model, cheapest tiers).

Note

This skill uses pass shared/ for API key retrieval. If you don't use pass, you'll need to adapt scripts/ping.sh to source keys from environment variables or another secret manager.