The tool-agnostic content layer for this repo is AGENTS.md at the repo root, with its CLAUDE.md shim. This .claude/ directory is the Claude-specific action layer that complements AGENTS.md with project-local skills, slash commands, and per-developer settings.

Layout

skills/ — Project-local Claude Code skills. Each skill is its own subdirectory containing a SKILL.md describing what it does and when to use it. Use this for repo-specific procedures.
commands/ — Project-local slash commands. One Markdown file per command: commands/<command-name>.md. Use this for /foo shortcuts that only make sense in this repo.
settings.local.json — Per-developer Claude Code settings (tool permissions, MCP server paths, IDE preferences). Never committed — this filename is in the repo's .gitignore.

When to add a skill (vs. inlining in AGENTS.md or promoting upstream)

Add a skill here when:

The procedure is repo-specific (it would not be useful in other NBL repos as-is).
The procedure is non-trivial (more than a one-line note that fits naturally inside AGENTS.md).
The procedure is a recipe an agent or engineer might re-run, not a one-off.

When to add a slash command

Add a command here when:

The action is something you find yourself typing the same prompt for repeatedly.
The repo has a non-obvious workflow that benefits from a shortcut.

Conventions

Skill and command names use lowercase-kebab-case, matching the folder naming convention in AGENTS.md.
Each skill directory has a SKILL.md (the entry point); supporting files (references, examples, sample data) live alongside it inside the skill's directory.
Each command is a single Markdown file named for the slash command: commands/<command-name>.md.
Skills and commands document why they make the choices they do — the rationale is more durable than the bare instruction.

How to add your first skill

Pick a kebab-case name describing the action: e.g., parse-linear-issues, render-delivery-row.
mkdir .claude/skills/<skill-name>/ and create SKILL.md inside it.
The SKILL.md opens with a short YAML-ish header (name, description, version) and then the prompt content.
Open a PR — the new directory and its SKILL.md are tracked once committed.

References

AGENTS.md — this repo's primary agent-context file (open standard).
Claude Code skills documentation — what a SKILL.md looks like and how Claude Code resolves them.