The backend has GET /calendar/agent/status?agent_id=<id> (read an agent's
runtime status, no side effects) but the CLI only exposed `hf agent status
--set` (sets the CALLER's own status). So delegate-task / on-call-handoff
status gates — which need to check whether the RECEIVER/incoming agent is
idle before pinging/handing off — had no CLI path; agents had to guess from
is_active.
Add `hf agent status-of <agent-id>` wrapping the GET endpoint
(X-Claw-Identifier header, no token). Prints {agent_id, status}; surfaces the
backend's 404 for unknown agents so callers can fail-open/closed.
Verified on sim: `hf agent status-of plxrec2` → idle; --json → {"agent_id",
"status"}; unknown agent → 404 "Agent not found".
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The plan-schedule workflow needs to report agent runtime status
(idle/busy/on_call/exhausted/offline) at the end of planning, but the
cli had no wrapper for this — workflows were dropping inline curl in
the middle of their procedure to hit the backend.
This adds 'hf agent status --set <status> [--reason ...] [--recovery-at ...]'.
The endpoint identifies the agent purely from X-Agent-ID + X-Claw-Identifier
headers (no token), so the cli reads AGENT_ID from env and falls back
to hostname() for CLAW_IDENTIFIER if it isn't set — same convention
the openclaw plugin uses. Refuses to send if AGENT_ID env is missing,
since this only makes sense from a pcexec/agent runtime context.
Surface entry added so 'hf --help' lists it.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
