help

name	help
description	Capability discovery — the canonical answer to 'what can you do?' and 'what do I say to ...?'
risk_class	soft

Help

The canonical "now what?" answer. Lists every capability the vault has, organized by what the user wants to do, with natural-language phrasings that trigger each. Zero token burn unless invoked — no proactive tips, no daily banners, no inline reminders.

Also serves as the author's reference when they forget what they've wired up. It's literally /help's job to be the index.

When to invoke

• The user types help, /help, what can you do, what do I say to …, how do I …, commands, or any equivalent phrasing.
• The user is new to rubber-ducky and the prior turn ended /onboard.
• The user asks about a capability and you're unsure of the exact verb — invoke /help and quote the relevant section back rather than improvise.

Output structure

Render a single, scannable message. Group capabilities by what the user wants to do. Within each group, list the natural-language phrasing first and the underlying skill / CLI second — the user thinks in intent, not in verbs.

Daily routines

• "good morning" / "morning" / "gm" → /good-morning — generate today's daily page and a prioritized brief; sweeps backend drift via /reconcile before gathering context.
• "wrap up" / "end of day" / "eod" → /wrap-up — summarize the day, run the linter, sweep backend drift via /reconcile after the day's local writes, and update tomorrow's focus.

Both routines invoke /reconcile as part of the twice-daily reconciliation ritual — see Workflows at the wiki ↔ backend seam for the full design walk-through.

Tasks

• "I'm starting on …" → /start — flips the task to in-progress, logs it on today's daily page.
• "done with …" / "finished …" → /close — marks the task done, stamps the close date, logs the activity.
• "link to as blocks/relates-to/…" → /link — cross-link two tickets in the same backend.
• "transition to " → /transition — sync the wiki and backend statuses.

Capture

• "something urgent: …" → /asap — file an ASAP item that persists until resolved.
• "I had an idea: …" → /idea — capture an idea for later.
• "remind me on to …" → /remind — date-keyed reminder.
• "meeting note: …" → /meeting-note — file a structured meeting note.

Context

• "learn from this: " / "ingest this" → /ingest-writing — paste any raw material (drafts, emails, articles, project docs). The skill extracts what's useful and routes it into wiki/voice.md, about.md, vocabulary.md, or preferences.md. Append-only.
• "what did I do last week?" / "when did we first discuss X?" → /query — natural-language search of the work history.
• "check the vault for drift" → /lint — workspace health report.

Integrations

• "connect " → /connect — wire up an external system. Surfaces four options inline (official CLI / pp-CLI / MCP / hand-written) and lets you pick.
• "reconcile" / "check drift" / "are we in sync" / "pull active from " → /reconcile — refresh active tasks, surface drift between wiki and each backend, and walk the User through resolution one ticket at a time.

Pushing work out

• "push this to " → /push — draft, preview-confirm, create the external ticket, back-link, audit. Gated by confirm.
• "comment on the ticket: …" → /comment — post a comment with a preview-confirm gate.
• "new ticket for …" → /new-ticket — originate a brand-new ticket on the intake backend, optionally mirror downstream.
• "log the release of " → /release — log a repo release and optionally finalize the included tickets across backends.

Settings

Every setting has a natural-language phrasing. Translating user intent into the right rubber-ducky settings ... call is part of normal conversation; this section tells the user what knobs exist.

Setting key	What it controls	Say this to change it
ingest.auto_on_wrap_up	At /wrap-up, offer to ingest today's drafts as voice samples.	"start / stop ingesting at wrap-up"
ingest.auto_on_onboard	At /onboard, invite the user to paste raw material.	"ask / don't ask me to paste at onboard"
ingest.kinds	Which extraction categories /ingest-writing produces.	"only extract voice and vocabulary"
confirm.<transport>.<verb>	Per-action confirm policy — auto (skip prompt) or preview (show + ask). Unlisted actions default to preview.	"always auto-confirm Jira comments"

Operators can read or write any setting directly:

rubber-ducky settings get <path>           # read a value
rubber-ducky settings set <path> <value>   # write a value (type-validated)
rubber-ducky settings enable <path>        # shorthand for `set <path> true`
rubber-ducky settings disable <path>       # shorthand for `set <path> false`

Every change is audit-logged to wiki/log.md.

Hybrid confirmation

The vault splits actions into two risk classes:

• risk_class: soft — reads, drafts, writes landing in the local vault. Auto-executes.
• risk_class: hard — sends, posts, creates, updates touching external systems. Gated by rubber-ducky confirm. The skill prepares a structured preview, calls confirm request --action <descriptor> --preview <text> (the CLI shows the preview and prompts the user unless settings.confirm.<descriptor> is auto), and only proceeds after confirm consume validates the token.

Per-action overrides under confirm.* are the power-user knob — say "always auto-confirm Jira comments" or "always preview Slack DMs."

Run rubber-ducky inside Obsidian (optional)

Most users run claude from a normal terminal or Claude Code Desktop. For a Chase Hannigan-style visual chrome you can run rubber-ducky inside Obsidian:

1. Open the vault in Obsidian.
2. Settings → Community Plugins → Browse, search "Terminal", install.
3. Enable the plugin.
4. Terminal plugin settings → set the launch command to claude and the working directory to the vault root.
5. Open a terminal pane (default hotkey, or via the command palette "Terminal: Open terminal in vault root").

The vault stays browsable in Obsidian's editor while claude runs in the in-Obsidian terminal pane. Behavior is identical to running from a standalone terminal.

Vault-level commands (operator surface)

The end-user surface is conversational. These exist for operators debugging or scripting:

• rubber-ducky init <dir> — bootstrap a new vault.
• rubber-ducky doctor / doctor lint — workspace health checks.
• rubber-ducky status — show vault info.
• rubber-ducky update — refresh bundled skills.

What this skill does NOT do

• No proactive tips. Help is invoked, not pushed.
• No memorization. This is a reference page generated each time, not cached state. If a capability changes, update this skill.