Init

mission-control: init

Scaffold a mission-control workspace into the current working directory from the canonical template files bundled with this plugin at ${CLAUDE_PLUGIN_ROOT}/skills/init/templates/.

Step 1: Check for conflicts. Never overwrite.

Check the current directory for each target path below. If a target file already exists, do NOT write it. Collect every conflict, report the list to the user, and skip those files. Only create what is missing. An existing threads/, inbox/, or archive/ directory is fine; only same-path files count as conflicts.

Targets:

• CLAUDE.md
• HOTLINE.md
• INDEX.md
• README.md
• threads/example-thread/README.md
• inbox/.gitkeep
• archive/.gitkeep

Step 2: Create the workspace

Copy each non-conflicting file from ${CLAUDE_PLUGIN_ROOT}/skills/init/templates/ to the same relative path in the current directory, creating directories as needed. Copy the files as-is; do not edit or reformat them.

For example:

cp "${CLAUDE_PLUGIN_ROOT}/skills/init/templates/CLAUDE.md" ./CLAUDE.md

If ${CLAUDE_PLUGIN_ROOT}/skills/init/templates/ does not exist, stop and tell the user the plugin install looks broken; do not improvise file contents.

Step 3: Tell the user the next steps

After scaffolding, report what was created (and any conflicts skipped), then give the next steps from the template README:

hotline setup --telegram-token <token>   # once, machine-wide
hotline init                             # in this folder
hotline start

Then: DM the bot, approve the pairing code with hotline pair <code>, and they're in. Point them at README.md in the new workspace for the full guide, and mention that HOTLINE.md sets the texting voice and is meant to be edited.

Also mention that hotline init pre-approves routine read-only tools (Read, Grep, Glob, LS, …) in .claude/settings.json so a remote texting user isn't buzzed for every navigation step — edits and commands still ask. They can widen or narrow that allowlist in .claude/settings.json to taste.

If they drive OpenCode instead of Claude Code, they run hotline init --harness opencode. That scaffolds a dedicated primary agent at .opencode/agents/hotline.md whose system prompt is hotline's texting mechanics and voice (OpenCode ignores the MCP instructions field, so the agent is how those rules — including the anti-prompt-injection pairing rule — reach it), and merges opencode.json with the hotline MCP server pinned to that agent (HOTLINE_OPENCODE_AGENT=hotline) plus a default permission block. Re-running regenerates the managed agent file but never clobbers a hotline.md they wrote themselves. The voice still comes from HOTLINE.md.

Do not run hotline commands yourself; the user runs them.