Skills are plain files

A skill is just a file: a folder under .claude/skills/ with a SKILL.md inside. There is nothing to install. Put it in your team’s content/ folder, and an agent whose context includes it gets it as a file in its worktree when it launches, like any other context file.

Write one

From Ledgerline’s billing team:

---
name: write-migration
description: Write a forward-only Postgres migration with a backfill plan.
---
Create `api/migrations/<timestamp>_<name>.sql`. Never drop a column in the same
release that stops writing it. Include a backfill query and an index plan.

Saved as billing/content/.claude/skills/write-migration/SKILL.md, it shows up as .claude/skills/write-migration/SKILL.md in the worktree of the next Claude Code agent you launch for billing: the Claude Code default context includes .claude/skills. A Codex agent’s default reads .agents/skills instead, so add .claude/skills to its context or keep a copy there. An agent that’s already running picks it up only when it calls hydrate_worktree; nothing is pushed into a running worktree on its own.

Child teams inherit it. Inheritance works per file: a child team that writes its own .claude/skills/write-migration/SKILL.md replaces that one file, and every other skill in the folder still comes from the parent (see Teams and inheritance). Step by step: Write a skill once for every repo.

Installed files and hand-written files are the same thing: plain content your team edits, reviews and shares through git like everything else.

Start from an example

The CodeSense docs repo comes bundled with every install, and its getting-started/content/examples/ folder holds copyable examples:

• skills/routine/, skills/agent/ and skills/codesense-setup/: skills that help an agent author routines, agents and your setup.
• dashboards/: ready-made HTML pages, such as a sprint analysis and a kickoff console.

Copy the folder you want into your team’s content/.claude/skills/ (or a page into a pane, below) and edit it. The examples are inert where they sit: nothing loads them until you copy them.

Nothing depends on a skill

CodeSense’s MCP tool descriptions carry everything an agent needs to use them, so no workflow breaks because a skill is missing.

Where the other pieces live

• Agents carry their own args, MCP servers, permissions and startup in their agent.yaml. See Agents and runtimes.
• Routines run scheduled or event-driven agent work; they replace the old automations.
• Panes are HTML files at <space>/panes/<id>.html, next to content/ and config/. They show in the right pane, hot-reload as you edit them, and sync like config/. Agents write and open them with the pane tools.
• Claude Code and Codex are built into the app; CodeSense knows how to launch each CLI.