agent-skills
Claude Code skill: let your agent drive fleets of local omp (oh-my-pi) model instances toward one goal — provider config, bounded turns, failure salvage, ver…
Install with your AI
Paste into Claude Code, Cursor, or any agent — it reads the repo and wires the tool into your project.
Install and set up agent-skills (git-clone project) into my current project. Found on https://claudeers.com/agent-skills-6 Repo: https://github.com/loktar00/agent-skills Homepage/docs: — Detected install method: git-clone → git clone https://github.com/loktar00/agent-skills Category: devtools. Platforms: cli, api, web. Read the repo's README for exact setup and env vars, then install it and wire it into my project. Claudeers Health Verdict: active; community-verified: false. Confirm the source before running anything.
git clone https://github.com/loktar00/agent-skills
// compatibility
| Platforms | cli, api, web |
|---|---|
| Operating systems | — |
| AI compatibility | claude |
| License | NOASSERTION |
| Pricing | open-source |
| Language | JavaScript |
agent-skills
My global Claude Code skills — the working set I run every agent session with, published as-is. Each skill is a self-contained directory; install any of them by copying it into your skills folder:
git clone https://github.com/loktar00/agent-skills
cp -r agent-skills/<skill-name> ~/.claude/skills/<skill-name>
Then just ask your agent — skills load on demand by name and description.
Orchestration
The centerpiece: one main agent that drives fleets of headless coding-agent harnesses toward a goal — it decides the topology, gives each instance an orchestrator and a slice of the work, and verified results flow back up.
| Skill | What it does |
|---|---|
harness-orchestration | The meta-skill: pick the right harness for a job, mix harnesses across lanes, and when a harness has no adapter yet, learn it empirically and have the harness write its own adapter skill |
omp-orchestration | Drive omp (oh-my-pi) instances: provider config, bounded turns, failure salvage, verification discipline, multi-lane rules — the shared doctrine lives here (tested against omp 17.3.5) |
codex-orchestration | OpenAI Codex CLI (codex exec) — written by Codex itself, headless under orchestration, every flag verified against its own CLI (codex-cli 0.147.0) |
grok-orchestration | xAI Grok CLI (grok -p) — written by Grok itself, same protocol (grok 1.0.5) |
Each self-written adapter ships with its NOTES.md evidence trail: which facts were verified by
running commands vs. prior knowledge. The set is self-extending — harness-orchestration contains
the exact protocol (and TASK.md template) that produced the codex and grok adapters.
Proof-of-work: distilled from ~90 real orchestrated turns across 10 models and 5 providers rebuilding the Warcraft III menu as a live browser scene — including two text-only models that built the full animated 3D scene blind:
Asset generation
Drive a local ComfyUI box headlessly — textures, sound effects, music — with the graphs verified rather than guessed.
| Skill | What it does |
|---|---|
comfyui | The engine: API-format graphs (UI-format workflows are NOT postable), the full endpoint surface, the discovery protocol for learning an unfamiliar model by pulling its shipped template, and the failure modes — including the wrong-VAE class that produces garbage with no error at all. Reference runner in templates/comfy_gen.py |
comfyui-krea2 | Krea2 images (~6 s at 1024²): the verified graph, and why it is not a FLUX model — wrong text encoder errors loudly, wrong VAE fails silently |
comfyui-audio | Stable Audio SFX/ambience (~6 s per 4 s clip) + ACE-Step music: verified graph, the checkpoint-has-no-CLIP trap, and SFX prompting that sits in a game mix |
Verification & QA
Because "it exited 0" is not evidence.
| Skill | What it does |
|---|---|
test-rigor | The evidence standard for any QA: what counts as proof, baseline-before-change, paired before/after under state drift, verdicts that don't overclaim. Per-project trap files plug in under references/ |
perf-proof | Prove a frontend perf change with paired, measured before/after numbers: builds both git refs, serves each production build, drives a real signed-in browser N times per condition, verdict fails on regression. Per-project configs under targets/ (npm install inside the skill dir for Playwright) |
Workflow
| Skill | What it does |
|---|---|
no-agent-attribution | Human-only authorship on everything git/GitHub: blocks AI co-author trailers, "generated with" footers, session links, and bot identities — with a pre-push audit and removal recipes. Auto-invoked on any commit/push/PR |
Model selection
| Skill | What it does |
|---|---|
model-fitness | A local ledger of which models actually succeed at which task types across delegated runs — surfaces per-model success rates and recommends a swap when the record says a model keeps failing a category |
Notes
perf-proofandmodel-fitnesswrite runtime data to~/.perf-proof/and~/.model-fitness/— local only, never in this repo (the perf-proof browser profile holds a real login).- The banner and section art were generated by Grok's
/imagine, orchestrated headlessly — dogfooding the skills that publish them.
License
CC BY 4.0 — use it, adapt it, ship it, but attribution is required: credit "agent-skills by Jason Brown (loktar00)" with a link back to this repo.
// faq
What is agent-skills?
Claude Code skill: let your agent drive fleets of local omp (oh-my-pi) model instances toward one goal — provider config, bounded turns, failure salvage, verification discipline, multi-lane orchestration. It is open-source on GitHub.
Is agent-skills free to use?
agent-skills is open-source under the NOASSERTION license, so it is free to use.
What category does agent-skills belong to?
agent-skills is listed under devtools in the Claudeers registry of Claude-compatible tools.
// embed badge
[](https://claudeers.com/agent-skills-6)
// retro hit counter
[](https://claudeers.com/agent-skills-6)
// reviews
// guestbook
// related in Developer Tools
🔓The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Curs…
🔓Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
🔓Use Garry Tan's exact Claude Code setup: 23 opinionated tools that serve as CEO, Designer, Eng Manager, Release Manager, Doc Engineer, and QA
🔓AI coding assistant skill (Claude Code, Codex, OpenCode, Cursor, Gemini CLI, and more). Turn any folder of code, SQL schemas, R scripts, shell scripts, docs,…