ship-ready distribution 19 plugins 4 agent CLIs rust-validated MIT SS+ CLI — omp

Ship-ready artifacts. One distribution home.

When an agent tool is refined enough to ship, it lands on DS4CC to install, inspect, or read. Raw local paths come first. The Few Good CLIs™Grok Build, Codex, Kimi Code CLI, OpenCode, and cursor-agent — without turning the stack into a bridge zoo. The omp CLI is the SS+ tier — and it keeps getting better, everyday.

all 19 plugins available curation-gated Titanium L3 substrate Titanium = no MCP
$ curl -fsSL https://raw.githubusercontent.com/VeigaPunk/grok-marketplace/main/scripts/install-xbgst-codex.sh | bash
INSTALL XBGST FOR CODEX → BROWSE PLUGINS NON-CLI GROK CHAT FALLBACK ↗

00New release · XBGST for Codex

The Grok-proven orchestration stack is now packaged for a raw local Codex install. WWKD and the shared XBGST contract remain the core; native Codex delegation does the work.

local first one paste anti-bloat

The complete, inspectable path

Install the Codex marketplace package locally, use xask as a CLI protocol when cross-provider consultation is useful, and keep implementation in native Codex agents. No MCP bridge is required.

READ + INSTALL XBGST →

Optional means optional: the MCP / ChatGPT companion is a secondary adapter for people who deliberately enable it. It is not the install path, not a runtime dependency, and not a claim of publication in the public ChatGPT app directory.

01Pick your host — watch it install

Live terminal. Few Good CLIs™ · five hosts, one repo. Commands below are the real install paths — outputs are representative.

ds4cc — grok build

click a host to replay its install sequence · sound off, green on

01bThe SS+ CLI — omp

SS+ CLI, and it keeps getting better, everyday. omp is not one of the Few Good CLIs™ — it is the tier above the family, and the only CLI listed on this wall.

native substrate no bridge zoo local-first
$ curl -fsSL https://omp.sh/install | sh

Not a grade we hand out — a gate UFO enforces in Rust: omp is the only substrate whose native handoff (omp-native-v1) the Pareto judge validates; every other CLI is contract-shape-only until its receipts land.

How UFO runs on omp — natively, not adapted

  • One authority, host transport — UFO's L1 is omp's own Main; lanes dispatch through OMP task + Agent Hub. No second orchestration runtime, no bridge zoo.
  • Ten seats, fail-closed — Main, one planner, six proposal lanes (scout, reviewer, critic, connector, sentinel, executor), the executor's depth-2 labrat, and a distiller that runs even when an upstream seat failed. The Rust judge rejects any handoff missing a seat; a failed seat stays in the topology as typed evidence.
  • Hash-bound handoff — every assignment byte-matches its SHA-256 and carries exactly one terminal | godspeed marker; contract drift mid-round fails the handoff. Only the Rust judge composes and admits proposals.
  • Autonomous model routing — the chinese_ufo formation pins a route per task class (depth → qwen3.8-max, review/integration → deepseek-v4-pro, mechanical → deepseek-v4-flash) with typed fallback chains. The operator never picks a model per dispatch.
  • Reroute, don't respawn — a failed lane keeps its identity, prompt, and cached context and takes another local route. Failure is telemetry, never a silent retry or a replacement agent.

The wall is omp today because omp passed the gate first. Few Good CLIs™ family members get listed here as they earn proper support — upstream authority, shipped record, copy-only action, and a policy that fails closed on anything else. Admission is a record, not a promise.

02The catalog — curated plugins + the SS+ CLI

02The catalog — 19 curated plugins

Every plugin ships an actionable SKILL.md (fenced commands, no boilerplate) and passes the Rust validator. Filter by name, category chips, or capability. Companion OS tools appear under os tuning.

19 / 19

03UFO Sighting — the live L1 wall

A dedicated terminal wall attached to the tmux session ufo_sighting:1: every active UFO-FSD L1 mission owns one registered, titled pane with stable run/session provenance, laid out on a responsive grid that adapts to the attached client. Work in progress — releases with the /ufo skill under FSD, the full self-driving regime. Takes over the Agent Pip slot.

One pane per mission

┌ ufo_sighting:1 ───────────┐┌ rinnegan-r1 ──────────┐ │
│ │ L1 · 6 lanes settled  │ │
│ │ verdict: 7 edits shipped │ │
│ └───────────────────────┘ │
│ ┌ hatch-twin ───────────┐ │
│ │ L1 · wave live        │ │
│ └───────────────────────┘└────────────────────────────┘

The wall is display and registration state only — never another orchestrator, never a source of runtime authority. A twin is one explicit depth-1 sibling L1 with fresh identities and origin provenance; it never merges judges and never recursively forks.

tmux grid pane registry twin cap: depth-1 WIP

Open the hatch

$ scripts/ufo-sighting open

open launches or focuses the configured terminal on ufo_sighting:1; status reports registered missions, attached clients, and live geometry as JSON. The grid fits up to 15 mission panes (3 rows × 5 columns), sized from the attached terminal.

WIP — lands with /ufo FSD: one L1 orchestrator, a WWKD plan before every judged round, specialist lanes on the native substrate, one Pareto judge, and an honest stop.

03bOS tuning — Omarchy usage limits

Waybar chip for live AI usage windows on Omarchy. Companion product under the os tuning catalog filter — not a marketplace plugin package.

omarchy-usage-tray

Cycles Codex, Grok, Kimi, Cursor, and Alibaba Token Plan meters on the bar. Left-click / scroll = next provider; right-click = identity swap when supported. Installs into ~/.config/waybar/ only.

$ curl -fsSL https://raw.githubusercontent.com/VeigaPunk/omarchy-usage-tray/main/install.sh | bash
OPEN PAGE → github ↗

04Install — Few Good CLIs™

Five supported hosts: Grok Build, Codex, Kimi Code CLI, OpenCode, and cursor-agent. Review third-party plugin sources before installation. Codex + sekhmet L3 is the optimal execution substrate as of 2026-08 (host binary resolve: repo docs).

Grok Build CLI xAI CLI · livepatch ban

# grok-orch (xbgst-stack) — one-liner; does not overwrite config.toml; livepatch not auto-applied
$ curl -fsSL https://raw.githubusercontent.com/VeigaPunk/grok-marketplace/main/scripts/install-xbgst-stack.sh | bash

# optional: merge (do not curl -o overwrite) grok-cli-config.toml
# https://github.com/VeigaPunk/ds4cc-marketplace/blob/main/grok-cli-config.toml

# optional: livepatch ban in the binary (not part of the orch one-liner)
$ git clone git@github.com:VeigaPunk/grok-build-livepatch.git
$ cd grok-build-livepatch && ./scripts/check-and-patch.sh

# ds4cc catalog (separate marketplace; not required for /xbgst)
$ grok plugin marketplace add VeigaPunk/ds4cc-marketplace
$ grok plugin list --available --json
$ grok plugin install myagents --trust
$ grok plugin enable myagents
# install the core set in one loop

for p in myagents godspeed-core mycommands myskills sekhmet xbrd-selector ds4cc; do
  grok plugin install "$p" --trust && grok plugin enable "$p"
done

Config: grok-cli-config.toml (Pages) · Livepatch: grok-build-livepatch hard-bans general-purpose/explore; first-party full tools use agent. xbgst: grok-marketplace / xbrd-grok. Non-CLI fallback: paste GROK_PASTE.md.

Codex OpenAI CLI

# marketplace plugins (hardened host binary: repo docs/TITANIUM-HOST.md)
$ cargo install --git https://github.com/VeigaPunk/xbrd-spark --locked
$ codex plugin marketplace add VeigaPunk/ds4cc-marketplace
$ codex plugin list --available --json
$ codex plugin add sekhmet@ds4cc --json
$ codex plugin add myagents@ds4cc --json
$ codex plugin list --json

Optimal substrate (as of 2026-08): Codex + sekhmet (xbreed L3) — pure namespaced sparks, no worktrees, up to 64 concurrent, double-work tolerant, Rust-only. Adding a plugin installs it enabled. Start a new Codex session to load skills/tools. Cache example: $CODEX_HOME/plugins/cache/ds4cc/myagents/0.2.0. TUI: /plugins + Space. Local: codex plugin marketplace add .

Kimi Code CLI TUI plugins

# commands to enter in the Kimi TUI
/plugins install https://github.com/VeigaPunk/ds4cc-marketplace
/reload
# or register the full packaged catalog in the Kimi TUI
/plugins marketplace https://veigapunk.github.io/ds4cc-marketplace/.kimi-plugin/marketplace.json
/plugins install 
/reload

Invoke as /skill: and /:. Third-party installs ask for trust first — review before approving.

OpenCode bootstrap script

$ git clone https://github.com/VeigaPunk/ds4cc-marketplace.git
$ node ds4cc-marketplace/scripts/install-opencode-agents.mjs --global
# or: --project /path/to/project

Writes 15 the-* agents plus the-netsshark (16 subagents total), and a separate orch primary mode. Structurally verified.

cursor-agent Cursor CLI · Few Good CLIs™

# install Cursor Agent CLI (headless / -p print mode)
$ curl https://cursor.com/install -fsS | bash
$ cursor-agent --version

# oneshot print (no TUI) — trust + text out
$ cursor-agent -p --trust --output-format text --model composer-2.5 -- "hello from ds4cc"

# cloud / Ultra agents also speak this surface — same Few Good CLIs™ seat
$ cursor-agent -p --trust --model cursor-grok-4.6-high-fast -- "run the task"

Now part of the Few Good CLIs™. Headless agentic burn via cursor-agent (also the cloud Agents / Ultra substrate). No native marketplace protocol yet — consume DS4CC skills/plugins from the repo or host docs; use xask / xbrd dispatch templates when crossing models. Docs: cursor.com/docs.

05Optional ChatGPT companion — app.ds4cc.com

A secondary, opt-in Apps SDK wrapper exposes a reviewed read-only subset of the catalog. Raw local installs remain the primary DS4CC path; this bridge is never required for XBGST.

MCP endpoint

https://app.ds4cc.com/mcp

Scoped to an explicitly reviewed, read-only tool surface with required annotations and widget CSP/domain metadata. It is not the public 19-plugin marketplace, and this page makes no claim that it is published in the public ChatGPT app directory.

read-only reviewed subset health route domain-verified

06The Rust validator — no broken plugins, ever

A binary crate (ds4cc-validator) gates every plugin on structure, schema, and skill actionability. CI runs it on each push.

$ cargo run --manifest-path marketplace/validator/Cargo.toml -- marketplace
Validation passed.
$ cargo test --manifest-path marketplace/validator/Cargo.toml
test result: ok. all green
cargo fmt — clean clippy -D warnings — 0 13 integration tests
Dual-layout aware: marketplace/ (web/CI) and repo root (canonical Codex layout at .agents/plugins/marketplace.json) both validate.

What every plugin must prove

  1. Local source integrity — path pinned to ./plugins/
  2. Manifest parses.codex-plugin/plugin.json, name matches entry
  3. Semver — strict X.Y.Z
  4. Full interface block — display, descriptions, developer, category, capabilities
  5. Skills existskills/ + README.md present
  6. Actionable SKILL.md — fenced code or recognized CLI prefix; boilerplate rejected

07Curation — a living marketplace, not a junk drawer

DS4CC keeps the strongest entry per domain and updates the reviewed set as better evidence emerges. Full policy in CURATION.md.

Admission gate

  1. Useful capability with a reproducible demonstration
  2. Provenance — identifiable maintainer or publisher
  3. Current docs — install steps + supported-platform scope
  4. Compatible licensing and required notices
  5. Reviewed security/privacy posture for its permissions
  6. Automated structural checks + manual end-to-end capability check
  7. Claims limited to demonstrated evidence

Claim discipline

Comparative, compatibility, price, and “full functionality” claims require a dated evidence packet: tested hardware/software, comparison baseline, known limitations, repeatable demo.

Proposed, not advertised: DS4CC microG stays off the public listing until its parity matrix, demos, latency numbers, and price source land in the repo.
Removal: entries are demoted or dropped when unmaintained, unsafe, misleading, legally unclear, or dominated by a demonstrably stronger option.

08Research note · separate tabs

Not Titanium policy. Not an MCP install path for sekhmet/Codex Titanium.

Exa.ai — the real deal

Extremely good web research product. First thing in the hop-on tool wave that actually felt sharp. Separate tab — praise only, not “wire this into L3.”

OPEN EXA TAB →

Bloat we refuse

OpenClaw (hard no already), Honcho, Hermes/Nous (unsalvageable after ~$200 DO), mise (enshittified Node manager — use fnm multishell), MCP zoos, emoji panels that will not die. Operator kills with evidence — dual of curation.

OPEN BLOAT TAB →

omegaG — controller-native agent control

DualSense / DualShock 4 shortcut mapper for terminal-first agents. Formerly DS4CC. Standalone product site on this domain — not a marketplace plugin.

OPEN OMEGAG → vercel ↗ pages ↗
Titanium / sekhmet L3: no MCP policy. Pure namespaced sparks. Do not ship Exa-as-MCP (or any MCP) onto the Titanium host.

09Alibaba Cloud Model Studio · public good

OpenAI-compatible Qwen / multi-modal APIs (intl Singapore), free new-user token quota, official bl CLI. We wired this for agent stacks — sharing the path is the point.

What you get

  • Intl base: dashscope-intl.aliyuncs.com/compatible-mode/v1
  • Anthropic-compatible: …/apps/anthropic/v1/messages
  • CLI: npm i -g bailian-clibl text chat
  • Console: Singapore region API keys + docs

Keys are region-bound. Prefer env inject (DASHSCOPE_API_KEY); never commit sk- material.

Start here

Product console (Singapore) and the Alibaba Cloud benefits campaign (referral). New accounts may receive free Model Studio tokens — offer varies.

MODEL STUDIO CONSOLE →

BENEFITS / REFERRAL A927SY →

Disclosure: the benefits link is a referral (referral_code=A927SY) — DS4CC may receive a benefit if you sign up through it; eligibility and offer vary by region and time. Product docs: OpenAI-compatible chat · API key. See also #referrals.

10Referrals — tools we actually use

Disclosed invites only. No dark patterns. We ship the path (docs, base URLs, secret hygiene) because the agent stack is better when builders share what works.

Headliner: Alibaba Cloud Model Studio (intl). OpenAI-compatible Qwen APIs, Anthropic-compatible messages, free new-user token quota (offer varies), official bl CLI. Wired for agent stacks on DS4CC — this is a public good, not a drive-by affiliate dump. Read the wire-up →
featured intl singapore A927SY

Model Studio / Alibaba Cloud

Benefits campaign + Model Studio free-quota path. Same referral code on the official benefits page.

JOIN VIA A927SY →

console wire notes

Disclosure: the Alibaba Cloud benefits campaign link is a referral — DS4CC may receive a benefit if you sign up through it; eligibility and offer vary by region and time.

Kimi

Invite for Kimi — used in the DS4CC agent lane mix (connector / revenger paths).

KIMI INVITE →

Disclosure: the Kimi invite link is a referral — DS4CC may receive a benefit if you sign up through it; eligibility and offer vary by region and time.

OpenCode

Bootstrap path for the OpenCode host in the Few Good CLIs™ marketplace.

OPENCODE REF →

Disclosure: the OpenCode link is a referral — DS4CC may receive a benefit if you sign up through it; eligibility and offer vary by region and time.

Footer mirrors the same three disclosures. No Grok/x.ai invite links by policy.

11Orchestration lanes — live routing

Mirrored from the packaged SSoT (marketplace/plugins/xbrd-gdsp-fknpft/commands/references/xbreed-shared.md § Axis → Profile Mapping + xask Gate). If this section and the SSoT disagree, the SSoT wins and this section is drift.

Role declarations

the-judge            fable 5 · xhigh
the-planner          sonnet · medium
the-scout            sonnet · medium
the-reviewer         sonnet · medium
the-labrat           sonnet · medium
the-executor         sonnet · medium
the-connector        sonnet · medium
the-distiller        sonnet · medium
the-simplifier       sonnet · medium
the-revenger         sonnet · medium
the-sentinel         sonnet · medium
the-critic           sonnet · medium
the-mutation-tester  sonnet · medium
the-scribe           sonnet · medium
the-netsshark        host extension

Execution gates

planner              wwkd · native
scout / labrat / executor xask --spark --gs codex → gpt-5.3-codex-spark
reviewer             xask --gpt55 --gs -e low codex → gpt-5.6-sol (default tier)
critic               Layer-0: heuer-planning → xask --gpt55 --gs -e low codex → gpt-5.6-sol
revenger RECON       xask --gpt55 --gs -e high codex → gpt-5.6-sol
sentinel             xask --gpt55 --gs -e low codex → gpt-5.6-sol
connector            xask --spark --gs codex → gpt-5.3-codex-spark
mutation: single / <=4 xask --spark --gs codex
mutation: >=5 / breadth xask --effort high --gs codex
distiller / simplifier / scribe native

Codex Sol lanes use explicit reasoning and the neutral default service tier; fast is opt-in. Spark lanes resolve through Sekhmet to gpt-5.3-codex-spark (fallback gpt-5.6-luna). Teammate prefixes: ccs- sonnet, cco- fable 5 (judge), cdx- codex, g- local Gemma/HVM.

Local lane — Gemma 4 over HVM4

xask --gs gemma (aliases g, legacy gemini) dispatches through xbreed ask gemmagemma-hvmrun.sh/run-hvm4.sh → Bend 0.2.38 gen-hvm → HVM4 4.0 control gate → Ollama. Default model gemma4:26b via HVM_GEMMA_MODEL. Cloud Gemini CLI is retired — do not call the gemini binary. Any g-* teammate (and connector when routing local breadth) MUST use this lane — not the retired cloud Gemini path.

Ollama/Gemma 4 is the serving substrate. HVM4 controls routing only; tensor inference is not claimed to run in HVM4.