Dagger Design Proposals
Guidelines for writing design proposals for Dagger features.
Before Writing
Always research first:
- Check relevant internal docs, such as
internal-docs/dagger-codegen.md, for context - Look at related code in the Dagger codebase:
- GraphQL schema:
core/schema/*.go - CLI commands:
cmd/dagger/*.go - Core types:
core/*.go
- GraphQL schema:
- For public API or workspace proposals, read
internal-docs/version-gating.md - Understand existing patterns before proposing new ones
Structure
# Part N: Title
*Builds on [Part N-1: Title](link)*
## Table of Contents
- [Problem](#problem)
- [Solution](#solution)
- [Core Concept](#core-concept)
- [CLI](#cli)
- [Status](#status)
## Problem
Numbered, concise limitations:
1. **Short title** - One sentence explanation.
2. **Short title** - One sentence explanation.
## Solution
One paragraph summary.
## Core Concept
GraphQL type definitions with inline docstrings:
```graphql
"""
Type description here.
"""
type Example {
"""Method description."""
method(arg: String!): Result!
}
```
Go for implementation examples:
```go
func New(ws dagger.Workspace) *Example {
// ...
}
```
## CLI
Real command examples:
```bash
$ dagger command --flag
OUTPUT
```
## Status
One line.
---
- Previous: [Part N-1](link)
- Next: [Part N+1](link) or "Part N+1: Title (coming soon)"
Style
- Concise - Trust the reader. Remove fluff.
- Tables - Use for comparisons.
- GraphQL for APIs - Type definitions, not Go interfaces.
- Go for implementation - Examples showing how modules use the API.
- Real examples - go-toolchain, node-toolchain, not abstract Foo/Bar.
- Less is more - Remove sections when challenged.
What to Avoid
- Separate "Methods" sections (use GraphQL docstrings)
- "Design Rationale" sections unless specifically valuable
- "Open Questions" that aren't real blockers
- Go for type definitions (use GraphQL)
- Layout examples that might confuse
- Over-explaining
Process
- Gists as source of truth - Publish early, iterate in gist
- Link parts together - Previous/Next at bottom, "Builds on" at top
- Each part stands alone - But builds on previous
- Iterate quickly - User feedback drives changes
Iterating with User
When you have clarifying questions or notes:
- List them first - Present a high-level numbered list of all questions/notes
- One at a time - Walk through each item individually, waiting for user response
- Don't dump - Never present all questions with full details at once
Codebase References
When writing proposals, reference actual Dagger code:
| Topic | Location |
|-------|----------|
| GraphQL schema definitions | core/schema/*.go |
| CLI commands | cmd/dagger/*.go |
| Core types (Directory, File, etc.) | core/*.go |
| Engine internals | engine/*.go |
| SDK codegen | internal-docs/dagger-codegen.md, cmd/codegen/*.go |
| API version gates | internal-docs/version-gating.md, core/schema/*.go |
Example: To understand how Host.findUp works before proposing Workspace.findUp:
# Find the schema definition
grep -r "findUp" core/schema/host.go
# Find the implementation
grep -r "FindUp" core/host.go
Publishing
# Create new gist
gh gist create file.md --desc "Dagger Design: Part N - Title" --public
# Update existing gist
gh gist edit GIST_ID file.md
# Post changelog comment (always do this after updates)
gh api --method POST /gists/GIST_ID/comments -f body="## Changelog
- Change 1
- Change 2"
Always post a changelog comment after updating a gist with significant changes.
Related Skills
Check for other Dagger skills that may help with research:
engine-debugging- Engine debugging workflows, trace replay, cache snapshotsinternal-docs/dagger-codegen.md- SDK codegen, templates, bindingsinternal-docs/version-gating.md- schema views and public API version gatesinternal-docs/- Cache and engine implementation references
Scan to join WeChat group