| name | help |
| description | Human-facing capabilities menu — renders the router's intent→skill table in plain language, grouped by workflow, so a user can browse what steer can do without knowing any skill name. Read-only; sources the live router table so it can never drift from actual routing. |
| when_to_use | Use when the user wants to browse steer's capabilities rather than run one — "what can steer do?", "what can you do?", "show me the commands", "list the skills", "I'm new here, what's available?". This is discovery, not navigation: for "what should I do next" in a real repo route to /steer:next, and for getting a repo onto the standards route to /steer:setup. |
| argument-hint | [optional: a skill or area to zoom into] |
| disallowed-tools | Edit, Write, NotebookEdit, EnterWorktree |
Browse what steer can do (read-only menu)
/steer:help is the one surface a curious user can point at to see the whole
capability set at a glance. Everything steer does is normally reached by
describing a goal in plain language and letting the router pick the skill (see
00-router.md) — you never have to know a skill name. This skill is for the
person who wants to look at the map anyway: it prints the menu.
It changes nothing. It reads the router table and re-presents it; it never edits,
commits, routes, or runs another skill. If the user then picks something, that's
a separate turn.
Single source of truth — render the router table, don't retype it
The authoritative capability list is the ## Intent → skill front-door table
in ${CLAUDE_PLUGIN_ROOT}/rules/00-router.md. Read that file now and build the
menu from its rows. Do not hardcode the list here — if you transcribe it, the
menu drifts the moment a skill is added or a front door changes. Every row you
show must come from that table as it stands this session, so a new front door
appears in the menu automatically.
Include the front doors (the Intent → skill table) plus the two
user-invocable skills the router surfaces only in its below-table prose —
/steer:standards and /steer:reference (take their one-line purpose from that
prose). Do not surface the internal gateways (user-invocable: false —
tracker-sync, spec-scaffold); they are never a user's entry point. You may
mention that each front door auto-routes to specialized skills, but don't
enumerate those unless the user asks to zoom in.
Phase 1 — Read the router
Read ${CLAUDE_PLUGIN_ROOT}/rules/00-router.md. Take the front-door rows from the
## Intent → skill table: each row's "trying to…" phrase and its target skill.
Phase 2 — Group and render
Present the front doors grouped into these plain-language areas, in this order.
Map each router row to the area its target skill belongs to; omit an area that has
no rows this session (so the menu stays honest if the table changes).
- Get set up — bootstrap or maintain a repo on the standards (
setup and the
specialized init / adopt / sync / doctor it dispatches to; protect).
- Shape the work — think a feature through, absorb a PO's document, capture
and sequence the backlog (
spec, build, intake, issues, adr).
- Do the work — implement a change or fix an issue now, including hotfixes
(
work, work --hotfix).
- Find your bearings — figure out where things stand and what matters most
(
next, audit), automate that sweep on a schedule as an autonomous loop
(loop), get a shareable stakeholder-readable page of one feature (explain),
and this menu itself (help).
- Plumbing — report a defect in the steer plugin upstream (
report), and load
the operating manual or reference prose on chat-only surfaces (standards,
reference — the two below-table entries).
For each entry render one compact line: the plain-language goal first (from
the router's "trying to…" column), then the invocation in backticks —
e.g. - Think a feature through without building it — /steer:spec. Lead with the
goal, not the skill name; the whole point is that the user recognizes their
intent, not that they memorize a command.
Close with one line reminding them they can just say what they want in plain
language — the router will pick the skill — and that /steer:next answers "what
should I do now" in a specific repo, which this menu deliberately does not.
Zooming in (optional argument)
If the user named a skill or area ($ARGUMENTS), skip the full menu and expand
just that one: read the target skill's SKILL.md frontmatter (description +
when_to_use) and summarize what it does, when to use it, and which front door
reaches it (per 00-router.md). Still read-only — describe it; don't run it.
What this skill is not
- Not a navigator: it never reconstructs repo state or recommends an action.
That's
/steer:next. If the user asks "what should I do next", route there.
- Not a dispatcher: it never bootstraps or picks init/adopt/sync. That's
/steer:setup.
- Not a place to restate the rules: the always-on manual loads via the
SessionStart hook (or
/steer:standards on chat surfaces). This is just the
capability index.