Orbitant POC Skills

A workbench where the Orbitant team prototypes and shares skills for AI coding agents. People drop ideas in, we try them out together, and the ones that prove useful graduate into something we all use.

Skills follow the Agent Skills format and work with Claude Code, Cursor, Copilot, Codex, and others.

Installation

npx skills add weorbitant/orbitant-poc-skills

Other install options

# Install only one skill from this repo
npx skills add weorbitant/orbitant-poc-skills --skill learn

# Install globally (available across all your projects)
npx skills add weorbitant/orbitant-poc-skills -g

# Target a specific agent
npx skills add weorbitant/orbitant-poc-skills -a claude-code

# List available skills without installing
npx skills add weorbitant/orbitant-poc-skills --list

See the skills CLI docs for the full list of flags.

Available Skills

learn

A personal learning coach. A suite of six interconnected skills that help you figure out what to learn, sit down and learn it, and look back at what you've learned. All six share a simple markdown state store at ~/.claude/learning/ (see skills/learning-data-spec.md).

Use when:

• "Set up my learning profile" / "I want to start learning"
• "What should I learn next?" / "Add X to my backlog"
• "Let's learn" / "Continue my lesson" / "Pick up where I left off"
• "Where am I with my learning?" / "Learning status"
• "What have I learned this month?" / "Generate a learning report"
• "I'm done for today" / "Wrap up this lesson"

Commands:

Command	Purpose
/init-learning	Create or update your profile — role, gaps, goals, preferred learning style.
/new-learning	Add a learning idea to your backlog or get suggestions for what to learn.
/learn	Start or continue an active lesson, milestone by milestone.
/learning-status	Fast check of your current lesson, backlog, and streak.
/learning-report	Generate a progress summary over a period of time.
/end-learn	Close out a session or lesson, log what you learned, archive it.

Typical flow: /init-learning once → /new-learning to queue topics → /learn to work through them → /end-learn to wrap up → /learning-report to reflect.

debrief

A bottom-up friction-capture skill. Run at the end of a Claude Code session to turn the corrections you made and the project knowledge you had to volunteer into proposed CLAUDE.md edits — so the next session starts higher up the curve. Designed to be high-signal: returning "session looks clean" with no proposals is a first-class outcome, not a fallback.

Use when:

• "/debrief" / "/debrief the naming convention thing"
• "wrap up this session" / "what did we learn this session"
• "any rules to capture" / "should we update CLAUDE.md"

Commands:

Command	Purpose
/debrief	Scan the full session for friction, propose CLAUDE.md edits.
/debrief <hint>	Same, but focused on the area named (e.g., /debrief the API naming convention thing).

Output: Either one message listing N proposed edits with cited evidence per proposal — which the engineer approves all/some/with-edits/none of — or a celebratory "session looks clean" message if nothing meets the bar. Approved edits are appended to the project root CLAUDE.md. The engineer commits via their normal git flow.

v1 scope: CLAUDE.md edits only. No skills, subagents, hooks, slash commands, or settings proposals. No subdirectory CLAUDE.md targeting. No SessionEnd hook reminder. Stateless across runs.

Contributing a skill

1. Create a folder under skills/<your-skill-name>/ with a SKILL.md.
2. Follow the frontmatter convention used by existing skills (name, description, license, version, metadata.author, metadata.tags).
3. Add a section to this README with a short description, a Use when: bullet list of trigger phrases, and any commands the skill exposes.
4. Open a PR.

Browse skills.sh for inspiration — what makes a skill discoverable is a clear description and a concrete "use when" list.