Submit a preset

Presets on this site are just directories of YAML in the site repository. There is no account system and no upload form — submissions are pull requests, reviewed like code.

1 · Create your folder

One folder per preset under src/content/presets/, named by its install slug (lowercase, hyphens, no spaces):

src/content/presets/my-great-preset/
├── meta.yaml          # site metadata (card, category, verification)
├── preset.yml         # dsh display metadata (name + description)
└── agent.cordis.yml   # the actual plugin wiring & persona

2 · Fill in the three files

meta.yaml

name: My Great Preset
description: One crisp sentence — shown on cards and in the dsh picker.
order: 10                    # picker sort; built-ins take 1–4
category: Coding            # Coding | Review & Quality | Writing | Research | Data | Ops | Learning
tags: [python, cli]         # lowercase, 2–4 of them
author: your-github-handle
mountedTools: [tool-fs, tool-bash]   # row ids you actually mount in agent.cordis.yml
useCases:                   # 2–4 bullets: when should someone reach for this?
  - Refactoring legacy Python CLIs
  - Writing regression tests before touching code
highlights:                 # 1–3 things that make this preset different
  - Refuses to edit files without a failing test first
verifiedVersion: 0.1.0-rc.7   # the dsh release you actually tested on
verifiedDate: "2026-08-18"
license: MIT

preset.yml

name: My Great Preset
description: One crisp sentence for the dsh preset picker.
order: 10

agent.cordis.yml

Start from a copy of a built-in preset (never edit standard in place — upgrades overwrite it), then adjust: Rows are id (local) + name (the real @deepseek-ai/dsh-* package) + optional config:

# My Great Preset — https://dshpresets.com/presets/my-great-preset
# Format matched to the built-in presets: apps/cli/config/agent-presets
# Rows your build actually boots: dsh --profile web --dump-config
- id: persona
  name: '@deepseek-ai/dsh-persona'
  config:
    text: |-
      You are a … (specific role, operating rules, refusals)
- id: tool-fs
  name: '@deepseek-ai/dsh-tool-fs'
- id: tool-bash
  name: '@deepseek-ai/dsh-tool-bash'
  disabled: true        # hard-disable: the tool is not mounted at all

3 · Review checklist

What maintainers look for before a preset earns the verified badge:

• It actually ran. You installed it into $DSH_HOME/.agent-presets/, started a session with it, and it behaved as described. Say which dsh version.
• Persona is specific. A role, operating rules, and what it refuses to do — not three adjectives.
• Least-privilege tools. Only mount what the preset needs; hard-disable with disabled: true where it matters (a read-only reviewer disables tool-bash).
• No secrets, no phone-home. Nothing in the YAML should contain keys, tokens, or non-official plugin ids you cannot vouch for.
• Slug matches folder, and mountedTools in meta matches the wiring file.

4 · Open the PR

Fork the repository, add your folder, run pnpm build to prove the site builds, and open a pull request with one paragraph: what it does, which dsh version you tested, and a short session transcript excerpt.

Re-verification policy

dsh is in developer preview and will break plugin ids between releases. When a new dsh version lands, maintainers re-run every preset; ones that fail lose the verified badge until patched. That honesty is the whole point of this site.