Agent Skills Setup

Use the active agent’s native question, progress, skill invocation, and delegation capabilities as described in harness capabilities. Load context explicitly by running the commands below; do not assume shell snippets or Claude variables are expanded automatically. Read shared preferences from ~/.agent-skills/config.json; agent-specific overrides live under agents.<agent>.

You are configuring agent-skills for this machine. The settings are saved to ~/.agent-skills/config.json.

Load the current git user and effective shared configuration explicitly:

git config user.name

~/.agent-skills/scripts/shared/context.sh config

The installer imports legacy Claude preferences automatically, retains the original files, and applies versioned migrations. Codex-specific overrides live under agents.codex. Set common preferences at the top level unless the user asks for an agent-specific difference. Reapplying setup detects installed supported agents; no target flag is required. Use --target claude|codex|both only for an explicit override.

Step 1: Work Folder Location

The work folder is where plans, reviews, and working notes for all your tickets are stored — outside of any repo, so they survive branch deletion and work across multiple repos.

Ask the user where they'd like to store these files. Suggest ~/Work as the default. If they're on a work machine with iCloud or a cloud-synced folder, suggest that as an alternative (e.g. ~/Library/Mobile Documents/com~apple~CloudDocs/Work).

Ask through the available question interface, or conversationally when none is available:

• ~/Work (Recommended) — simple local folder
• ~/Documents/Work — inside Documents
• A custom path supplied by the user

After they answer, expand the path (resolve ~) and show the full absolute path for confirmation.

Step 2: Branch Prefix

The branch prefix is prepended to all branch names you create (e.g. ambrose/eng-42-dark-mode). Suggest the first word of their git user name (lowercased) from the loaded context.

Ask:

• <suggested-prefix> (Recommended) — from your git config
• A different prefix supplied by the user

Step 3: Discover Superset Host (Optional)

If the Superset CLI is available and authenticated, discover a default host for /super-work:

superset hosts list --json

• Exactly one host → note its host_id when consistent with the user’s authorized computer and execution preferences; otherwise ask before selecting it.
• Multiple hosts → ask which to use as the default.
• Zero hosts → leave host_id unset and explain that a host must be connected in Superset before /super-work can create a workspace.
• Missing CLI, authentication, or connection → skip this optional discovery; /super-work will handle setup on first use.

An old device_id is not a host identifier. Preserve unrelated configuration, but do not reuse that value or select a remote computer without authorization.

Step 4: Anything to Leave Off This Machine? (Optional)

setup.sh installs everything by default. The exclude key in the config is a persistent opt-out: setup.sh never installs anything listed there, and on every run it removes anything listed that it installed earlier. This is the place to declare it, so a fresh machine never gets the excluded pieces even once.

Ask a single-choice question through the available interface; allow a free-text answer:

• Nothing — install everything (Recommended)
• Skip the workflow skills — plan-work, plan-review, do-work, do-fixes, code-review (for machines where an org repo already provides them)
• Skip all hooks — pretooluse, notification, message-display, window-title

If they pick "Other", they can name any mix of the items below. Map what they say onto this schema — every key is a list, and "all" in a list excludes everything under that key:

"exclude": {
  "hooks":       ["pretooluse", "notification", "message-display", "window-title"],
  "mcp":         ["playwright"],
  "guidance":    ["core", "personal"],
  "skills":      ["<skill directory name>", "..."],
  "attribution": ["sessionUrl", "commit", "pr"],
  "cli":         ["claude-resume", "reclaude"]
}

shared cannot be excluded (other skills depend on it). Anything unknown is warned about and ignored by setup.sh, so a typo is harmless but worth catching here.

Step 5: Write Config

Save the confirmed values in ~/.agent-skills/config.json. Read the existing file first and only set the keys you collected — other skills store their own keys here (team_repos, linear_team_statuses, projects, …) and a re-run must not wipe them:

Use the shared config writer for each confirmed value. Its final argument is a JSON value; quote strings correctly and set only collected keys. For example:

~/.agent-skills/scripts/shared/context.sh config-set --shared user_prefix '"ambrose"'

~/.agent-skills/scripts/shared/context.sh config-set --shared work_root '"/Users/ambrose/Work"'

Substitute the actual confirmed values. Set optional host_id and only the named exclude categories the same way, passing each nested key separately. Use --shared for common defaults; use agents codex or agents claude key components only for an explicitly requested per-agent preference. Leave every other key intact. Read config again to verify the saved effective values.

Create the work folder if it doesn't exist:

mkdir -p "<absolute-path>"

Step 6: Confirm

Show a summary:

"Setup complete.

• Work folder: <path> ✓
• Branch prefix: <prefix> ✓
• Host: <host-name> ✓ (or "not configured — run /super-work to set up later")
• Excluded: skills: plan-work, do-work (or "nothing — everything installs")

Run bash setup.sh in the agent-skills repo to apply, then /plan-work <LINEAR-ID> to start your first work session. To change exclusions later, edit the exclude key in ~/.agent-skills/config.json (or run setup.sh --without <component>) and re-run setup.sh."

If there was a pre-existing config (shown in loaded context), note what changed.