iu-endpoint

Estrellas0

Forks0

Actualizado25 de mayo de 2026 a las 07:52

Validate the IU unified endpoint and discover better models. Probes every transport, health-checks the models configured in OpenCode (opencode.json), reports per-model backend redundancy + live latency, and diffs the live catalog to surface newer/hotter models worth adopting — for OpenCode and for the Hermes Agent (especially Kimi). Use when checking endpoint health, picking a reliable model/host, or deciding whether to upgrade configured models.

Instalación

Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.

Fuente

jkrumm

jkrumm/dotfiles

Abrir repositorio de GitHub Ver repositorios del creador

Descarga

Ocupaciones relacionadasSOC

Basado en la clasificación ocupacional SOC

Desarrolladores de softwareOcupaciones informáticas y matemáticas·SOC 15-1252

Explorador de archivos

3 archivos

SKILL.md

readonly

Más de este repositorio

mismo repositorio

brain

jkrumm/dotfiles

Read and write the second brain — a git-backed Obsidian vault at ~/SourceRoot/brain (PARA structure + a top-level wiki/ knowledge tree). Use whenever the user mentions "brain", "second brain", "knowledge base", "note", "remember", "look up", "find note", "capture", "wiki", or "okf". Also use for any request to save, retrieve, or organize personal knowledge that belongs in the brain rather than in a repo-specific doc.

2026-08-020

distill

jkrumm/dotfiles

Write, draft, and polish prose through a human-owned 7-step pipeline. Use whenever the user says "distill", "write article", "draft", "summarize into", "create content", "synthesize", or "publish". Also use for any task where the output is a piece of long-form writing — the pipeline keeps prose quality out of coding sessions.

2026-08-020

localai

jkrumm/dotfiles

RETIRED — manage the (now decommissioned) per-machine mlx-audio TTS/STT stack; TTS/STT moved to the cloud audio-gateway (VPS)

2026-07-300

remote-dev

jkrumm/dotfiles

Operate the Mac mini remote dev host — connecting from the MacBook over mosh or herdr --remote, running and reattaching many Claude Code agents, herdr workspaces/panes and its socket API, claude --bg durable daemons, the Uptime Kuma readiness heartbeat, and the failure modes specific to a headless always-on Mac. Use when the user mentions herdr, mosh, the mini, "remote dev", "dev host", detaching or reattaching agents, "did my agent survive", sessions dying on lid-close, or asks how to reach a dev server running on the mini.

2026-07-270

img

jkrumm/dotfiles

Manage the personal image stack — public CDN uploads/transform URLs and the private image-share layer. Use whenever an image needs a URL (public or private) for an article, blog post, vault note, README, OpenGraph tag, or a one-off share; or when the user says "upload this image", "share this image", "private image", "publish image", "host this screenshot", "get me a CDN link", "resize this", mentions "image-share", or asks where an image lives.

2026-07-240

sync

jkrumm/dotfiles

Bidirectionally sync ALL git repos across two machines — this MacBook and the always-on Mac mini — over SSH, using GitHub/GitLab as the transport. Commits and pushes uncommitted work on both sides, then rebases/fast-forwards each so they converge, with per-repo subagents resolving rebases and merge conflicts. This is a MULTI-repo, TWO-machine operation — distinct from /commit, /ship, or /git-cleanup, which act on a single repo. Use whenever the user wants to sync their machines/repos, "catch up" the laptop to the Mac mini or push laptop work back, is about to travel or just got back, has uncommitted work scattered across many repos, or wants to reconcile branches that diverged between the two machines. Trigger even on a bare "sync", "sync my machines", or "get my repos up to date".

2026-07-240

name

iu-endpoint

description

IU Unified Endpoint — validate & discover

The IU unified endpoint (op://common/anthropic, host derived from BASE_URL) fronts many model backends through several transports. This skill answers three questions: what's up right now, which alias gives the most reliable host, and are there newer/better models I should switch to (OpenCode + Hermes).

Key facts (the mental model)

The model alias is the host selector. Each id maps to one or more backend "sinks" (owned_by in the catalog). More backends = more redundant = less likely to 429/timeout. The validator prints backends=N per model.
- Examples: claude-opus-4-6 (3 backends) is steadier than claude-opus-4-7 (1). Kimi-K2.5 (2: Nebius + Azure) is steadier than Kimi-K2.6 (1: Sweden Central, throttle-prone). gpt-5 (9) is the most redundant GPT.
Transports on the same host: /anthropic/v1, /openai/v1 (rich catalog), /azure/openai/..., /gemini/v1beta, /replicate/v1. There is no /bedrock passthrough (404) — Bedrock is only an internal backing.
Resilience in OpenCode: the AI SDK auto-retries 429s with backoff. For hard failover across models, switch alias (or, in Hermes, code a primary→fallback).
Probe quirks: gpt-5* reasoning models need max_completion_tokens (not max_tokens); *-codex models return empty over chat-completions (responses API only) — don't configure them for OpenCode. The validator handles the first.
Auth: /openai + /replicate use Authorization: Bearer; /anthropic uses x-api-key + anthropic-version; /azure uses api-key; /gemini uses x-goog-api-key. Same single key for all. Never print the key.

How to run

bash .claude/skills/iu-endpoint/validate.sh          # full: transports + health + catalog diff + Hermes
bash .claude/skills/iu-endpoint/validate.sh --quick  # transports + configured-model health only

The script reads the key from Keychain (claude-sdk-api-key) and never prints it. Full run takes ~30–60s (it sends a tiny completion per configured model).

What to do with the output

Health triage. Call out any configured model that is not ok:
- THROTTLED (429) → transient capacity; prefer a higher-backends sibling.
- TIMEOUT/ERR(503) → backend down or saturated; note it, suggest alias.
- High latency on a 1-backend model → flag as slow-prone.
Reliability advice. When two aliases serve the same model tier, recommend the higher-backends one as default (e.g. opus-4-6 over opus-4-7 for daily use).
Discover upgrades. In the NOTABLE list, find [NEW] ids that are a newer version or stronger sibling of a configured [cfg] model (e.g. a newer Gemini preview, a higher GPT-5.x, a newer Kimi). For each genuinely better one, propose the exact opencode.json edit (under provider iu or iu-anthropic), with tool_call/reasoning/attachment/limit. Verify it actually completes first with a one-shot curl (use max_completion_tokens for gpt-5*). Only recommend models that return real text.
Hermes advice. Compare the models grepped from ~/SourceRoot/hermes-agent against the best available. Be specific about Kimi (the user runs Kimi in Hermes): K2.6 is single-backend/throttle-prone, K2.5 is dual-backend/steadier — recommend a primary + fallback (e.g. K2.6 primary, K2.5 fallback) and point at where in hermes-agent the model is wired. Flag any newer Claude/Gemini brain worth switching to. Do not edit hermes-agent from here unless asked.
Report concisely. A short health summary, a ranked "consider adopting" list with backend counts, and any concrete config edits. No key, no raw catalog dump.

Updating OpenCode config

config/opencode/opencode.json (symlinked to ~/.config/opencode/opencode.json) holds the curated model set. Editing it needs no make setup (it's a symlink). Keep the set lean and current; this skill is the mechanism for keeping it so. After any edit here, commit in dotfiles.