/advisor ask inside an enabled session is a normal consult with resume
and state writes. Only a conversation where the mode is not on runs a
one-off, and that path no longer forwards another chat's transcript_path.
Co-authored-by: Eric Zakariasson <ericzakariasson@users.noreply.github.com>
Consults in a conversation that never ran /advisor no longer re-bind the
project silently; they stay off, or run as a one-off when the user asks.
Re-binding (in the skill and in the hook library on first bind) removes
the previous conversation's pending and last-response markers so the
end-of-turn nudge cannot fire for edits made elsewhere.
Co-authored-by: Eric Zakariasson <ericzakariasson@users.noreply.github.com>
The skill cannot see its own conversation_id, so a bare /advisor in a
second chat used to report 'already enabled' and leave that chat unbound:
hooks stayed silent there and a consult could resume the other chat's
advisor. /advisor now always rewrites state.json (keeping model and
nudge), and consults never resume an advisor id this chat did not save.
/advisor status is the read-only path.
Co-authored-by: Eric Zakariasson <ericzakariasson@users.noreply.github.com>
Hardcoded slugs drift between Cursor surfaces, so /advisor <model> now
resolves against the subagent model slugs the agent already has in
context: exact slug as is, family name to its latest model at the
highest reasoning tier. The only stated default is the latest Grok at
xhigh effort. Examples use Cursor models only.
Co-authored-by: Eric Zakariasson <ericzakariasson@users.noreply.github.com>
Pin advisor-subagent to grok-4.6[effort=xhigh] and make `grok`
(cursor-grok-4.6-xhigh) the default alias; every alias now resolves to
its family's highest reasoning tier. Opus and the other families remain
available via /advisor <model>.
Co-authored-by: Eric Zakariasson <ericzakariasson@users.noreply.github.com>
Cursor counterpart of Claude Code's /advisor. The main model consults a
stronger (or different) model before major decisions, when stuck on an
error, and before declaring a task done; the advisor gets a full briefing
plus the transcript path and returns a verdict with guidance.
- skill `advisor`: /advisor [model|off|status|ask ...|nudge on|off],
checkpoint protocol, briefing template, usable as a Custom Mode
- agent `advisor-subagent`: read-only, pinned to claude-opus-5[effort=xhigh]
by default, overridable per session via the Task model parameter
- hooks: track edits since the last consult, log consults to
.cursor/advisor/log.md, nudge a pre-completion consult once per batch
of edits (quiet when the agent ended on a question for the user)
- register in marketplace.json and the root README
Co-authored-by: Eric Zakariasson <ericzakariasson@users.noreply.github.com>
Cursor Agent