Fix stale command and flag names in agent docs and skills

[Snodgrass-Wilkerschnoz]

Summary

AGENTS.md and the skill files name commands and flags that the current production and QC specs no longer serve. An agent following them hits unknown command or unknown flag errors. This PR fixes the names to match both live specs.

Was	Now	Production	QC
list-call-results <id>	list-results-by-call <id>	old name missing, new name present	old name missing, new name present
run-call <id>	run-monitor <id>	old name missing, new name present	old name missing, new name present
--since <ISO>	--from <ISO> (window start)	only --from/--time exist	only --from/--time exist
--before <ISO>	--time <ISO> (window end, returns results before it)	same	same

Other changes:

• AGENTS.md gets two new bullets:
  • Time-window flags. --from is the window start and --time the window end. Pass full timestamps; a bare date like 2026-10-14 returns no results.
  • v3 deprecations. QC marks create-call, list-calls, get-call, update-call and delete-call as deprecated in favour of the *-api-monitor commands. Production does not have those yet, so the docs keep the older names and tell agents to check --help.
• skills/embed/setup-api-monitor.md: the verify step pointed at fields that aren't in the result rows (result.success, result.failure_reason). It now uses meta.result_category and meta.http_code.
• run-monitor's documented response now matches what it actually returns: {call_id, context, location_id, result_id, user}.

create-call, list-calls and the other v1 names are left as they are because production still uses them. Switching the docs to the v3 names should wait until production serves them.

How this was checked

• Checked each command's presence and flags with --help against both builds (-tags prod and the default QC build).
• Tested --from/--time against real results on QC to confirm which flag is the start and which the end.
• go test ./... passes.

🤖 Generated with Claude Code