Skip to content
Agentmart
Submit a tool

stack-guide

How vllm-on-tap runs vLLM on each stack type, supported or planned - one reference per type with its prerequisites, config fields, serve, readiness, destroy…

Skill by using-system

IndexedMIT★ 1
Write a review

Chalk Labs

Need help deploying stack-guide?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/using-system-stack-guide

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

How vllm-on-tap runs vLLM on each stack type, supported or planned - one reference per type with its prerequisites, config fields, serve, readiness, destroy and traps. Read the current stack type's reference only.

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/stack-guide

What's inside

Skill name
stack-guide
Description
How vllm-on-tap runs vLLM on each stack type, supported or planned - one reference per type with its prerequisites, config fields, serve, readiness, destroy and traps. Read the current stack type's reference only.
License
MIT
Instructions
63 lines in SKILL.md · view on GitHub

Files (8)

  • SKILL.md
  • aws.md · references/
  • azure.md · references/
  • gcp.md · references/
  • kubernetes.md · references/
  • local-vllm-docker.md · references/
  • local-vllm-metal.md · references/
  • local-vllm.md · references/

SKILL.md (first lines)

# stack-guide

One reference per stack type in `references/<stack type>.md`. Read only the
current environment's type. The stack type `aca` of an earlier version is
`azure`: read it as `azure`, and write `azure` on the next write of the
environment.

A planned type's reference only says it is planned and has none of the
sections below - tell the user the type is not supported
in this version, and stop.

Every supported reference has these sections, in this order:

- **Prerequisites and install** - the tools the type needs, how to check
  them, and each one's official install command (run only on the user's yes).
- **Config fields** - the `config` keys of an environment of this type, with defaults.
- **Prepare** - what `/vot-config` creates once (optional).
- **Serve** - numbered checks, run in their written order and stopping at the first refusal (the "already served" one leads to the destroy-or-stop question), then the exact command that starts unit `vot-<preset>`.
- **API key** - how requests authenticate and where the key is read from
  (optional).
- **Ready when** - the readiness check and its time bound; poll across
  several bounded tool calls (each under the host's tool-call limit, e.g.
  5 min), never one long loop.
- **Destroy** - the exact command that removes unit `vot-<preset>`.
- **Traps** - one line each.

Common prerequisite (every type): `uvx --version` succeeds - install
`brew install uv` or the official script
https://docs.astral.sh/uv/getting-started/installation/ (on the user's yes).

Placeholders used in the references: `<preset>` (the preset name),
`<model>`, `<served name>`, `<vllm args>` (the merged `vllm_args` rendered
as `--flag value`, `true` as `--flag`, `false` omitted; on a shell command
line each value is single-quoted, `--flag '<value>'`, and in a JSON array
each value is a JSON-encoded string, so a JSON value such as
`{"method":"mtp"}` passes intact; a `'` inside a value is written `'\''`
in both cases, since a reference's JSON array may itself be single-quoted
in the shell), `<port>` and the
other `config` fields by their key, plus the placeholders a reference
defines itself. `[...]` marks a
part included only when its condition applies, without the brackets.

Tracing (every type): when the environment has `otlp_endpoint`, or the
reference's *Serve* checks resolved one, add
`--otlp-traces-endpoint <otlp_endpoint>` to `<vllm args>` and set
`OTEL_SERVICE_NAME=vot-<preset>` and
`OTEL_RESOURCE_ATTRIBUTES=<resource attributes>` in vLLM's environment,
where `<resource attributes>` is
`vot.preset=<preset>,vot.model=<model>,vot.stack=<stack type>`, each
value percent-encoded outside `A-Za-z0-9._~/-` (`<model>` after the
preset's per-stack override).
For an OTLP/HTTP
endpoint (port 4318 or a `/v1/traces` path) also set
`OTEL_EXPORTER_OTLP_TRACES_PROTOCOL=http/protobuf` and pass the full URL
ending in `/v1/traces`, appending that path when the environment's value
lacks it (vllm-guide's *Traces over OTLP*).

Preset environment (every type): each name in the preset's `env` must be
exported in the user's shell (`printenv <NAME>` prints a value) - refuse
naming the missing one - and is passed to vLLM by name, never by value in a
file or on a command line.

From using-system/vllm-on-tap, MIT.

Signals

Stars
1
Forks
0
Open issues
0
Last commit (this folder)
Sep 27, 2026 (1 month ago)
Last release
Sep 27, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from using-system

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get stack-guide running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying stack-guide?

CallGet help
Cite this page

stack-guide skill. Agent Mart. https://agentmart.dev/skills/using-system-stack-guide (last updated 2026-10-24)