Skip to content

Specialized Dockerfiles for agent flavors (docs, media, data, presentations, content) #7

@pavelanni

Description

@pavelanni

Summary

Create multiple Dockerfiles that bundle domain-specific CLI tools into the DocsClaw image for different business use cases.

Motivation

The base DocsClaw binary is 5 MiB, but agents need external tools for real work. Rather than one bloated image, provide purpose-built flavors that include only the tools relevant to each domain.

Proposed flavors

Flavor Use case Key tools
docs Document conversion, review, summarization pandoc, libreoffice-headless, poppler-utils (pdftotext), wkhtmltopdf
media Image acquisition, processing, description imagemagick, ffmpeg, exiftool, curl/wget
presentations Slide deck creation and rendering marp-cli, mermaid-cli, chromium-headless
data Report generation, CSV/Excel processing gnuplot, csvkit, python3+openpyxl, jq
content Blog posts, newsletters, polished Markdown markdownlint-cli, vale (prose linter)

These are tool-focused, not role-focused. The agent's personality and role come from the ConfigMap; the Dockerfile determines what tools are available.

Design considerations

  • All flavors should use a common base image (UBI minimal or UBI micro)
  • Multi-stage builds to keep image size reasonable
  • Each flavor gets its own Dockerfile: Dockerfile.docs, Dockerfile.media, etc.
  • CI should build and push all flavors to quay.io/redhat-et/docsclaw:<flavor>
  • Consider whether some flavors should bundle MCP servers (e.g., Playwright for media)

Non-goals

  • Developer-focused images (code review, linting) — developers use personal agents
  • "Kitchen sink" image with everything — defeats the purpose

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions