Skills and icons

Skills

A skill is a folder with a SKILL.md: instructions (and any scripts or references) for a kind of task, which the agent reads when that task comes up. Your lab’s analysis conventions, a pipeline’s quirks, how to use an in-house tool.

code flower mounts skills read-only and installs them as the agent’s own:

From	Applies to
~/.codeflower/skills/	every sandbox (create the folder)
skills: in config.yaml, a list of folders	every sandbox
-s/--skills DIR, or Skills in the start form	that sandbox

Each folder can be one skill (it has a SKILL.md) or a folder of skills.

~/.codeflower/skills/
  scrnaseq-conventions/
    SKILL.md
  lab-lims/
    SKILL.md
    query.py

The sandbox also brings its own skills: reports (storyteller), the icon gallery, MCP setup, and, when isolated, the network allowlist.

Icons for figures

For schematic figures (study designs, workflows, pathways, graphical abstracts), the agent draws an editable SVG with icons from your icon gallery, ~/.codeflower/icons, shared by all your projects. It plans the figure with you, sketches the layout, finds icons, and iterates on your feedback (select elements in the preview and comment).

The gallery holds only icons you added. Servier Medical Art’s terms forbid automated downloads, so the agent never downloads icons: it tells you what to look for and gives you search links, and you download the file and put it into the sandbox (e.g. upload it to the workspace in the app’s Files). The agent imports it with its credit (SciDraw drawings need their authors, license and DOI).

icons.folder in config.yaml moves the gallery; off turns it off.

Generating icons

When nothing fits, the agent can generate an icon, after asking you, if you configure an image model:

images:
  provider: openai          # or google
  api_key_env: OPENAI_API_KEY

The key is passed to the sandbox as a private file, and an isolated sandbox may reach the provider’s API. Codex can also use its built-in image tool.