Skip to main content
SubTurtles support Claude Code skills through repeatable --skill flags on ctl spawn and ctl start. Use skills when a task needs domain-specific workflows (frontend QA, Remotion video pipelines, testing patterns, migration playbooks).

What --skill Does

--skill is both a runtime toggle and a tracking field.
1

Capture skill names

ctl parses each --skill <name> flag and stores names as JSON in subturtle.meta (SKILLS=[...]).
2

Pass names into SubTurtle runtime

ctl starts the worker with python -m super_turtle.subturtle ... --skills <name1> <name2>.
3

Enable project skill directory

If skills are non-empty, the loop sets add_dirs = [\"super_turtle/skills\"].
4

Forward to underlying agent CLI

Claude/Codex calls include --add-dir super_turtle/skills, so the agent can discover project-scoped skills.
5

Expose in status/list output

ctl status and ctl list print the configured skill names for that SubTurtle.
ctl does not validate skill names. A typo in --skill still appears in metadata/output, but no matching skill will be loaded.

CLI Usage

spawn example:
start example (existing workspace):
Verify what was set:

Skill Directory Layout

Project-scoped skills should live under:
This layout keeps skill logic versioned with the repo so every SubTurtle run sees the same behavior.

Remotion Example: remotion-best-practices

If you already have the Remotion skill in your local skills library, copy it into the project skill directory:
Then spawn a SubTurtle with that skill:
The skill provides Remotion-specific guidance (animations, sequencing, captions/subtitles, ffmpeg usage, render patterns) without bloating the SubTurtle state file.

Create a Custom Skill

Create a new skill folder and SKILL.md:
Use it:
Keep skills narrow and reusable. Put durable workflows in skills, and keep task-specific details in each SubTurtle’s CLAUDE.md.